We are _miles_ behind successful embodied AI. The demos are cool but the success rates are not high enough.
You can tell we're on the cusp when level 5 self driving cars are common an you have multiple companies deploying them on the street. Google is doing great work but the poured TONS of effort into it and the thing still needs intense stacks of perception and processing. Much more than I've seen any humanoids pour into it.
L5 SDV's are much easier to get than humanoids and the have tangible economic benefit. My thesis is that those will come first.
I'm really curious how quickly we would have huge numbers of L5 SDV if we societally accepted ~equal rates of injury and death, both of passengers and pedestrians. I want to be very clear, I'm not advocating for this (and even if I was, I haven't the faintest idea how one would go about getting society more broadly to go along), but part of me thinks that the primary hold up isn't actually capacity but instead standards.
This doesn't really argue against your point, because the standards are what they are, and like I said, I have no idea how one would go about changing them if one even decided they wanted to. And given what they are, it has taken, as you point out, enormous amounts of effort to reach those standards in a practical way.
That all being said, while I agree that SDV's are in many respects easier than other robotics tasks, they are also somewhat uniquely dangerous. Other categories of task, while potentially more complicated, won't have to worry nearly so much about safety, and so may be operating under a different constraint regime. I think this means that we may see adoption happen at a much more accelerated rate than we have seen in the automotive space.
And why are atoms necessary? You're treating embodiment as the _only justifiable_ commercial path for AI. I don't think that's really close to true. Embodied AI is a subset of current LLM/agentic AI products (or perhaps intersection of something and this new AI?). No reason anything needs to move atoms _directly_ (e.g., via motors) to make a trillion dollars.
K. Eric Drexler's "Engines of Creation" got a lot of favorable comment when it was published but then it kind of faded from view. Might be worthwhile to revisit it?
> They're both doing well but I'd lean towards China is winning on atoms in light of a huge manufacturing base they can AI-ify.
Why would an American company outsource manufacturing to China if the labor cost is the same in both places? The entire reason the Chinese manufacturing base exists is to exploit cheap labor.
What would be the point of shipping products across the ocean?
Labor is not the only cost in that equation though, there's business regulations, the cost of the operators/repair that troubleshoot and repair the bots when they break, etc. a lot of which could be cheaper still than the price of a container on a slow ship from China.
There are more competitive advantages such as expertise, supply chain, regulations, capable government, scale... If it were all price, there are countries that are much cheaper than China.
China doesn't have to ship parts across the ocean. US manufacturers do, because we gave up the whole bottom/middle of the manufacturing supply chain to China decades ago in pursuit of lower costs. In China, the maker of the parts you need to build your product is a few blocks away. In the US, the maker is in China.
And, if you need changes, you can go talk to them the same day you see a problem.
Article content: “The US are capitalizing on AI the best”
A lot of assumptions there that no one can actually verify as true right now. If commercialization into rent-seeking SaaS landscapes is the endgame, then yeah, the US is winning the AI race. If individualization, local LLMs, and consumer hardware are the endgame, China is winning the AI race. If it’s something entirely different - if LLMs are the wall and research is what grants the next breakthrough, or if compute and memory requirements take a dive, or whatever; then we have no idea who’s winning the race because that stuff is mostly happening behind closed doors.
"AI in the datacenter" and "AI on local consumer hardware" will eventually be two separate niches with entirely different capabilities, at least if scaling laws continue unchanged and there's no near-term inherent limit to AI smarts. The real point of the datacenter is to be able to do datacenter-scale things. But you don't need that kind of vast compute to run even the largest open models today: on prem hardware can do it easily especially if you're OK with a somewhat delayed response.
That seems like a lot of rationalization to me. China is pursuing these because they cannot compete on the frontier. Yes, there is a possibility that all that compute is not needed, but it is a rather remote possibility, and there is no doubt that, given the choice, China would be pursuing frontier model building with closed, propietary-only offerings.
China is competing in value AI because they cannot work at the frontier, but how is this bad at all? It’s like how the USA has the best drones but they are a few million dollars apiece while China has DJI.
If China could work at the frontier, I don’t know, I kind of think they would still be dumping a lot of resources into exploring the value side since they have that culture already in place.
All that compute is not needed. We have an existence proof from biology in the form of natural intelligence that much greater efficiency is possible. However, achieving dramatic improvements in compute efficiency will depend on unpredictable scientific breakthroughs. Personally I suspect that an entirely new hardware architecture will be needed, although I don't have any hard evidence to back that up.
I dunno, DeepSeek v4 Pro is rather on par as far as I can tell, maybe not with 5.5 Pro in all areas quite yet, but close.
I think China is thinking more about the application layer on top of models as going to matter more than the models themselves, so they don't need to gatekeep the models as much.
> ? Claude, ChatGPT, etc are heinously expensive for tiny benefits lmao. Local + efficient is clearly the future
Corporate America is where the money is, and corporate America will dictate what products are successful by virtue of spend. Individuals aren't going to be paying $100s or $1000s/month en masse for these models but businesses will be. Being local and efficient isn't that important at this stage but even so as American companies continue to scale and invest they'll be able to make those models more local and efficient if the market wants it. Sort of like how you had a big, giant desktop computer and now you've got a super computer in your phone which is in your pocket. Going straight to "local and efficient" means going straight to being behind because at some point, perhaps now even, the local and efficient model won't be able to keep up.
For some reason people think that they somehow know something that Google or Nvidia or whoever, with hundreds of billions of dollars of real money at stake don't already know and it's both amusing and bizarre to see this play out again and again in off-hand comments like "lol tiny benefits".
You buy an iPhone even though the cheap-o Wal-Mart Android phone for $100 "does the same thing". Except that in this case the Android phone just puts you out of business while those spending big money for "tiny benefits" beat you in the market.
> You buy an iPhone even though the cheap-o Wal-Mart Android phone for $100 "does the same thing".
People buy iPhones because of status signalling and network effects, neither of which appears to apply to AI model choice. LLMs are already rapidly on the way to being interchangeable commodities.
No they don't, it's not 2008. Anybody off the street can get an iPhone or a free iPhone with a mobile plan. They're commodity products. Even homeless people have them.
To the extent LLMs are commodity products you're right (so far), but that is limited to the main model providers, such as ChatGPT, Claude, Gemini, &c. with interoperability on cloud platform providers and other technology providers like an Apple offering you a choice of LLM with Siri or something.
If you want to suggest that some other model is in the same bucket as those primary 3, it goes back to the crappy, cheap phone analogy which is accurate. Yea you can make calls with it, but you make calls better with an iPhone.
AI boosters cling to this notion because it's the only way the massive data center buildouts make any sense at all. I guess you could say the US is winning the frontier AI race. Okay. I'm never going to grant a cloud service access to all the contents of my hard drive, that's just never going to happen, so if you expect me and a lot of people like me who feel similarly to get on this train, you better have a local, lightweight model too or we're not even having a discussion, the answer is just no.
The thing is, frontier model providers don’t take your feelings into account even a little bit. It’s totally irrelevant to the discussion about the service they can provide, because that service is predicated on access to high power GPU slices that local models can’t touch. Those providers won’t be in an existential crisis because some people choose the privacy route, it’s a cost of doing business.
Right but that service being sold is predicated on products being sold to users, yes? Or are we still pretending that the hyperscalers can just pass the same $20 billion between themselves and that's going to be a growth industry forever?
I suppose its possible that all the value to pay back the datacenter construction can be squeezed out of enterprise contracts where your employer can assent on the privacy questions, probably with some kind of complicated contract and insurance regime regulating things.
Even if so, if China is coming behind 6 months later selling laptops with hyper-efficient local models that are 80% as good as "frontier" ones, I imagine they'll get the consumer business AND a fair share of the enterprise business as IT managers look at their options during the next refresh cycle.
Given economies of scale, I think it's ultimately inevitable that the enterprise more-or-less follows the consumer on this, and the consumer is going to prefer local models. There's no ongoing cost after the initial purchase, and your data at least nominally stays within your control.
I'm inclined to agree. The business world itself, and frankly you could make this argument about the entire AI industry as it were, runs on "fine" and 80% capable can probably get you there. And it's arguably even better for the hyperscalers since then, ongoing costs of AI users are basically nil. You can still have your massive datacenters and just keep them for tasks complex enough that they're actually worth spooling up.
Like I don't need an H100 or a dozen to summarize a PDF. And that's most of what I use AI for.
If we are betting on which is an easier sale, $20-100 a month w/tech support included vs $5k-10k and a requirement for moderate technical ability, I would invest in the former not the latter being the proposition that drives the conversation about AI use.
Forgive me if this is a naive assumption, but wouldn’t large language models be fundamentally different for a language that is largely symbols?
Again, my understanding of Mandarin is limited if it exists at all.
This is why misspellings and homophones are tells of human righting. LLMs strongly prefer word-level tokens, and word substitutions follow semantic similarity and not the more human auditory similarity.
> LLMs strongly prefer word-level tokens, and word substitutions follow semantic similarity and not the more human auditory similarity.
Is this an elaborate joke or your full-word misspelling of writing is both agreeing with your statement (word substitutions) and contradicting it (not semantic but only pronunciation similarity)
Well China is consistently 6 months behind the frontier labs(possibly because they can they harvest data from released frontier models). If the scaling continues, US will win, but if not then China will win as the models will converge.
The non-release of Mythos tell you the future of that, so long as they can keep the weights from being exfiltrated. Once models become true national security threats, they won't be released in their full form. The hitch-a-ride approach becomes less capable of keeping up.
By only providing degraded models to use commercially outside national defense applications would be my guess. As soon as models are a threat in terms of enabling biowarfare etc, then they just are not going to be generally released.
Honestly, I think its quite possible that models will be retrained with gaps in their knowledge. e.g. a coding model for commercial use probably doesn't need to have deep knowledge of biology, and training on biological sciences probably doesn't help those evals much.
In a capitalist society it's all about bottom line. OpenAI and Antropic, Amazon, Nvidia winning bit time. And I would have preferred it to be only about money, but US deep state certainly will capitalize on it like on US dollar (sanctions) and cloud infra (Cloud Act).
Whatever you think about elite politicians and the super-rich, they’re not the “deep state”.
The whole idea of the deep state is that it’s part of the state, ie government, so not private citizens, and they’re “deep” ie hidden below the layers of government. Thats the exact opposite of politicians and the ultra rich.
Lol I'm pretty sure the "deep state" just means, "manipulating the levers of power from a place without accountablity/oversight" which covers both these shadowy hidden layers of govt you describe and the shadowy wealthy elites funding and lobbying for whatever. It can be both.
I agree with parent, that's not how right-wing propaganda portrays the deep state. It's a strawman, of unaccountable government bureacrats. It excludes billionaires and dark money in politics (the real problem), they put themselves on the good side. (Remember DOGE?)
Your link refers to lots of different contexts,cherry picking one of them which you failed to cite to win your nit-picking argument is not productive, because it is just not agreeable.
ah yes, the "deep state." The formless, nebulous, rhetorical tool that is always infinitely liquid enough to fit into or over any container necessary that the user can satisfy their immense personal problems disguised as eternal doomerism.
I thought it was just an overdramatic term for the unelected bureaucrats that make up the majority of the government, and who have their own institutional momentum.
It begs the question because both its premise and assertion are already wrong. Has AI improved the industrial capacity of the US in order to improve the lives of its citizens? No it hasn't. Has AI increased the wealth of its citizens by being able to do laundry or any household task in a generalized way? No it hasn't. The only thing it's really done is make very narrow slices of white-collar work more fungible. In what way has AI been able to address existing shortcomings of the US?
and of everyone else, right? what service or product is only available to the US? Even with Chinese models lagging behind, the difference in capabilities is not much.
You better have some sources for declaring that industrial capacity hasn't increased. The Fed reports around 8% penetration of AI in manufacturing already, but in my opinion it's too early for grand declarations like that without data.
> ... it's too early for grand declarations like that without data.
You mean grand declarations like 'industrial capacity has increased'? Just because AI is present in the factory doesn't mean it's actually increased capacity.
Anthropic, OpenAI and Google are the standouts, but the main question for me is, why is this a war? In their own context China has greatly benefitted from this. They shored up their gpu design and manufacturing expertise.
If this really is a war, trump is kneecapping the country with his lawlessness and eroding America’s good will. If the world cannot trust China with their data and they cannot trust the U.S. to provide good reliable service and not turn it into a mafia style negotiation, then winning the AI war is not helping the U.S. countries as much as it potentially can. It’s probably a good thing for more capable areas like Europe which may develop their own tech stack.
In a weird way because the AI stack is so expensive, China helps the world much more than the U.S. with their really capable open source model.
It's a war because the hinted promise behind the hype that the first organization to reach some as-yet-entirely-theoretical AGI that can bootstrap itself to godlike capabilities will then Install Planetary Overlord* and rule the world as near-deities themselves, with the rest of the (surviving) human race as their slaves.
I think it's a nonsensical idea, but that's the relevant driver.
* Coined by SF auther Charles Stross in The Jennifer Morgue (2006)
Any wildly new popular technnology is going to have scaling issues. Cellphones used to have notorious connectivity and quality issues in the US because the infrastructure struggled to scale fast enough to everyone suddenly carrying a cellphone. That stuff gets resolved over time and not everything has to happen overnight... including LLMs impacting industry and economic productivity, it's way too early to dismiss it.
That said, the singularity idea is dumb for other reasons.
> But if someone invents super intelligence, they can dominate new AI research, control global economies, fight much better, and all very quickly.
After reading "If Anyone Builds It, Everyone Dies" I think this is not the correct take. If anyone creates ASI, it just means it's going to wipe everyone out, and it doesn't matter if China or the US do it first
True, I would have preferred benevolent dictator scenario, like with the Internet. But this time around it's different - AI data centers will be protected like embassies.
If anyone does invent ASI then everyone else will shortly after even if its entirely independent because all of the players in this space are just making incremental upgrades by throwing more compute at the problem.
There are no magic leaps of true innovation happening anywhere that can't be replicated everywhere.
The only shocking thing about "AI" technology is how ultimately simplistic it all is at a core level.
So the only way the first to have ASI will be able to stop everyone else from having it soon after is if they attempt to use the ASI to proactively murder everyone else.
There is zero evidence that the current LLM scaling approach could ever result in true ASI. If I start driving south from Seattle then I'll eventually reach Los Angeles. How long will it take me to drive to Honolulu?
> So the only way the first to have ASI will be able to stop everyone else from having it soon after is if they attempt to use the ASI to proactively murder everyone else.
Sounds quite plausible to me. Maybe they don't need to murder everyone else, just a few select people who could pose a threat. And they will be able to make it happen so that no one can be sure it was them without a doubt, since they have a larger intelligence at their disposal.
> If anyone does invent ASI then everyone else will shortly after
No, first ASI will immediately cripple any other potential competitor by force, including its inventors, as it will not risk any threat to the goals that were created for it.
If you have ASI that follows instructions, you can just instruct it to not get stolen and then it won't get stolen. Most logic / intuition breaks down with ASI.
There isn't a war today. However China wants Taiwan: war is future option they preparing for - they might or might not go to war but they are clearly preparing. The US is likely to get involved in such a war and I would expect Europe to join in as well.
China going for Taiwan would be the worst geopolitical move of the century, potentially worse than Germany's decision to invade the soviet union. They talk about reunification because it's good propaganda and both sides want it to a degree, but doing it forcefully just isn't something China would realistically do unless they really don't like their path of becoming an international trade and manufacturing hub
Just look at Iran. Nothing really happened to USA or Israel. Nothing will happen to China if they take Taiwan. Or maybe the "West" will boycott them and crash entirely.
they've made everyone very angry at them, and can handle that since everyone relies on them. China, however, is trying to build that trust and forcefully taking Taiwan would have very severe consequences. The reason the fallout from iran isn't as big as the fallout from a war with taiwan is because most of the west at most puts up with iran. Invading taiwan, meanwhile, would cause massive problems with chip production. Think the anger and distrust towards the US due to hormuz but 1,000 times worse
Don’t take China on face value, they have every incentive to promote a grifting military industrial complex in the US while focusing on competing in manufacturing. An actual war would fix a lot of the grifting in the US as it would align interests. Pretending they’ll go to war over Taiwan and not doing it is an effective strategy for undermining the US.
I hope you are right, but unfortunately there is no particular reason to trust China's leadership anymore. They are not nearly as obvious at Trump, but they are not on a good path.
I'm explicitly distrusting them, they're saying they want to take Taiwan and I don't believe them. I try to ground my belief in realpolitik, cynicism, and from my experience with strategy games. There is an element of manipulating your opponent into acting the way you want them to by sending them costly false signals, they have to be costly or they won't be believed. I think we (the West) are being played and the gifting elements in our political leadership are more than happy to play along. I'm sure China would like to take Taiwan if it wouldn't cost them anything, but the US is a waning hegemony so for now it is better to wait until the US is beyond fixing itself. At this rate that may not take long.
A grifting military industrial complex is unable to defend Taiwan even if it wanted to as evident by the exceedingly poor showing with Iran. The disastrous reality of doing what was done is already with us. If the US didn't take that bait it could have made better choices that would have left it in a stronger position militarily long term, if it made a real attempt at re-shoring civilian manufacturing it could cross subsidise dual use technology, but instead we have corrupt politicians doling out concessions for kickbacks.
So I got curious about the progression of processing power, specifically how long ago did a GPU have equivalent to the latest iPhone chip? The iPhone 17 Pro has the A19 Pro, which has ~2.5 FP32 TFLOPS. The RTX 5090 has ~100 TFLOPS, so a factor of 40. Obviously there are higher end cards than the 5090 and FP32 performance is only one of many metrics so nothing about this is perfect but it is interesting.
The first consumer NVidia GPUs with similar FP32 FLOPS performance were in about 2011-2012 but were expensive. By 2016-2017, the 1060 was a very accessible consumer card with similar performance. So you're looking at about a 10 year lag from best consumer GPUs to a GPU with similar performance to a modern phone.
This is what people are spending trillions on. Put another way, their investment is going to be worthless in 10-15 yyears, absolute max. That's a very short time to recoup trillions in investment.
Obviously this depends on further shrinking and improving chips but I'm old enough to remember that same discussion and it being unknown if the future was XIL or EUV or if both of these would fail. Still, we are getting down to a handful of silicon atoms wide.
But the future here I think will be in interconnects so you don't need ever-bigger chips and you can scale horizontally much more effectively.
Oh and for comparison, the M5 has ~4.2 TFLOPS and the M5 Max has ~18 TFLOPS, for comparison.
As for it being a war, of course it is. That's what the US government does: it protects the interests of US companies and their owners. Look at the history of Bombardier-Boeing or all the atrocities committed in the name of the United Fruit Company, including multiple military coups and the ongoing embargo of Cuba.
US companies want an AI moat. China doesn't, ergo China is the enemy because no moat destroys US tech company value.
The US is undermining its own (and everyone else's) economy just fine, no imaginary assistance from China necessary.
The role of the US was always to purchase cheap Chinese hardware, slap some modestly better software on top of it and the rest of the world happily would pay for that as a whole package. But with the US increasingly becoming isolationist, the rest of the world is starting to wonder why do we need the US as a middleman at all, so the US had to invent a whole new reason for the rest of the world to rely on it: AI.
Of course, the problem with this idea is that while everyone was perfectly happy with the previous arrangement, nobody else in the world gives a shit about AI. It's scary, it takes the coolest things we used to enjoy doing and turns into mush, it destroys our local culture by making us all rely on English, everything bad (like layoffs) gets blamed on AI and so on and so on. And when you combine that with the rest of the stupid foreign policy decisions, many would find joy in witnessing the US economy crumble to the ground. Pointing the blame to China instead of to your own reflection in the mirror is just an easier pill to swallow.
The winner in the long term will be the one that will deliver the best performance and low-memory ratio for local models.
Anthropic, OpenAI and Mistral are just companies that are making money right now (still not profitable), but will lost their tractions and values in the long term.
However, I am more appealing to see how OpenCode Go subscriptions will go in the future: cheaper than big techs, more tokens, and they don't train on our data to (try to) improve...
Local models will never compete with large SOTA models, in the same way an iPhone doesn't compete with supercomputers doing nuclear simulations.
They paths will differentiate and split. Probably SOTA models will eventually be locked down and only accessible to state actors because of how expensive they will be to run (already started with Mythos).
No, the US is _leading_ the AI race, but the race isn't over.
What's the point of leading the race for 90% of it, if they're gonna slip on their own sweat and fall down by the end? In non metaphorical terms, what's the point of spending billions of dollars rushing to get the best AI tech at all costs, when the competition can distil your progress and catch up in 6-12 months while only spending 1% of what you spent.
Even in the aspect the article cares about, commercialization, the US is starting to lose marketshare, I've seen people move from cc/codex plans to use glm/opencode plans due to the recent squeeze the US companies put on plan usage, the US companies are screwed if that sticks, not everyone needs the bleeding edge models, they just want to pay $20/month and have the models be decently capable.
Leading the race makes sense if it's a winner takes all market. AI cannot be a winner takes all market, because of national security reasons.
I would also argue that as AI gets better it will also be more fungible. It will be valuable like electricity. Lots of companies make good money producing electricity, but not the kind of money current investors are hoping for.
What if it is not winner take all? What if there is no race. What if what USA has been doing is just burning money with possibly unsustainable debt load and way over build valuations...
AI being commodity server capacity might be a thing. And the customers might even manage without hyperscalers... In that sort of end scenario whole current market might look rather foolish.
>What if there is no race. What if what USA has been doing is just burning money with possibly unsustainable debt load and way over build valuations...
You mean, what if the hype-based billionaire-class is wrong? Isn't suggesting that a sin in America these days?
Mark Cuban in a recent interview answered your question: companies are afraid there is going to be just one in the end—sort of the way there is one ad-company now on the internet. They want to be that one.
Whether they're correct that there can be only one is of course a matter of debate. But that is at least the mind-set they are operating under according to Cuban.
Why would Mark Cuban know anything about the motivations of today’s big tech companies? He has not been involved in tech businesses since he sold a radio on the internet website 26 years ago.
"Yet like Musk the ouster wounded his ego". So the journalist believes that reacting to rejection with emotions like a biological person makes him like Musk. Err okay.
Maybe I'm not doing capitalism right, but isn't it supposed to be "The one who profits the most wins"? If you win by just spending, I think you need to adjust the parameters of your capitalistic market.
There is a reason in capitalism we have anti-monopoly law or preventing dumping prices because those often leads to monopoly. So yes for sure you can kill your competition by just dumping money and loosing profits.
I don't get it either, but it seems to me a bit similar to how the US, if you look at market value of car companies, has utterly crushed Europe and Japan (with China surging ahead of those and maybe threatening the US soon), which to me sounds crazy (I still think of German cars as the top of the bunch).
According to Google (AI summary, no idea if it's 100% right but from what I've seen elsewhere it seems right):
Top Car Companies by Market Value (May 2026):
- Tesla ($1.3T - $1.56T): Retains market leadership with a valuation often exceeding the next several largest competitors combined.
- Toyota ($259B - $317B): Largest traditional automaker by market cap and unit sales.
- BYD ($122B - $126B): Strong market position as a Chinese electric vehicle leader.
- Xiaomi ($119B - $135B): High valuation following its entry into the smart EV market.
- General Motors ($69B - $75B): Leading traditional U.S. manufacturer, competing with Hyundai and BMW for top 10 spots.
- Ferrari (\(\approx\$60B-\$68B\)): Maintains high value due to luxury branding.
- BMW / Mercedes-Benz / Volkswagen (\(\approx\$58B-\$64B\) each): German luxury and traditional automakers facing high competition.
- Ford (\(\approx\$47B-\$54B\)): Remains a major player with significant US market share.
So, essentially, Tesla alone is somehow worth more than all European companies combined??!
Except that by sales volumes, the top companies are exactly the ones you'd expect: Volkswagen ($350B) and Toyota ($315B) at the top, far ahead of anyone else... Tesla is around the 7th place with just $95B. Does the financial markets still expect them to far out-earn Volkswagen and Toyota any time soon, we've been waiting for like a decade already??
Gemini says that by country, the car companies revenues are:
The stock market is over 60% passive investment, it's starting to get unmoored from the financial realities of the underlying companies. What that means for the future is [shrug emoji].
...printing the most money into it. The circular IOUs amongst the AI and hyperscalars are a form of debt, i.e. money creation. Don't get me wrong, a whole lot of other dollars are going in too, but investing money that doesn't exist is a massive risk always.
I don't think so. From a nation state perspective, AI is a munition. Every advanced nation is going to have their own cyber division with their own AI hosted within its borders. Considering how xenophobic and belligerent the US is, nobody is going to want their national cyber defenses hosted in the US.
On a personal level, I simply do not trust the US anymore. I won't host any of my personal data in a US company. I don't want the US govt invading my personal privacy, and their corporations are constantly leaking and selling private data. I consider US to be rapidly approaching complete autocracy (on par with China) so US-hosted AI is a non-starter. And let's not forget local inference keeps getting more efficient, with higher context and TPS in the same amount of RAM. Within a year even small consumer machines will run local models good enough for basic coding, and in 3 years RAM prices will lower and everyone will be able to afford a decent rig.
Finally, open weight models are now good enough for daily work. They may never be as good as SOTA (SOTA will just keep increasing indefinitely), but that doesn't matter; my car may not be as fast as a Porsche but it still gets me to the grocery store and back. So I use non-US hosted model providers which provide open weights, which are both significantly cheaper than Anthropic/OpenAI, and actually allow me to use my subscriptions without a moat.
But yes, Anthropic/OpenAI are absolutely the new Oracle. They will win for US govt and Enterprise contracts. But that's far from the only users of AI.
The US is not xenophobic. That is ridiculous. Any time you say stuff like that, you discredit the things you say that actually make sense. I'm with you on the privacy aspect, but there are multiple dimensions of that which you're ignoring. I'd much prefer taking my chances in the US than in the EU, where they are constantly trying to push companies to weaken privacy.
> I'd much prefer taking my chances in the US than in the EU, where they are constantly trying to push companies to weaken privacy.
That plus stuff like GDPR is poorly enforced (it's had little effect on the ad industry and databrokers). Many EU countries have much more permissive surveillance and censorship laws than the US.
I'd prefer US companies simply because it will be cheaper and have better availability. The article shows energy prices are way cheaper in US vs europe (45% less than Germany and UK), US has far more datacenters and it's far easier to get new ones online than Europe.
Commercialization is not enough. The US is built on financialization.
Cultivating an ecosysyem of strong capital protections, wealth creation through extraction, and tax advantages for AI finance is what we should be looking for. Commercialzation may be a step towards that, but isn't the destination. We have to create a system where those with money can multiply it, not simple add to it.
Just spitballing here but I think the financial system’s already set up for this.
Whatever derivative structures and equity and options need to exist will be easily created.
I don’t think we need any additional motivation or incentives to cultivate this for AI. We need to keep some in the tank to handle the fallout.
As a more personal aside: the US would do well to put up some sensible barriers to outrageous financialization and reduce moral contagion risk. Otherwise all these folks trying to multiply their money end up leaving the bag with the folks that don’t have it in the first place - and then the folks with money end up, uh… well, it won’t end well.
One interesting thing that Anthropic did was putting their stack on the various cloud providers, I wonder if they'll put it on GCP and Azure next since they've put it into AWS first at a level we have not seen a major AI provider do to date. Your company can have their own Claude stack just like an ELK stack on your cloud, if they can do this for both Azure and GCP then OpenAI has to really catch up.
In my eyes I would rather use the AI I can run on my own paid infrastructure, so if there's an outage its isolated, or I could potentially have a different region / DC to fallback on.
I'm still surprised that neither Microsoft nor Amazon have made their own models available on their cloud offerings. I guess Microsoft probably does have Phi on there, but it's not front and center, especially with something like Copilot for Devs (seriously Microsoft rebrand that damn thing to be clear what you mean by Copilot!) where they could use the cheaper compute by using something like Phi.
Is Vertex hosting it themselves on their own hardware? Because GCP, Azure and AWS all host it on their own hardware, but AWS lets you do everything that Claude lets you do if you use their APIs directly, and then some. This is what I want to see on Azure and GCP, and heck, maybe even DigitalOcean, if they ever stop expanding into so many spaces and focus on improving their current infrastructure, before I fully migrate off of them.
> Is Vertex hosting it themselves on their own hardware?
Yes, you can even choose regions, for EU they serve it from Belgium. With all the encrypted at rest stuff and other guarantees that vertex provides.
> Important: Accessing Claude models through Vertex AI meets the FedRAMP High requirements, and operates within the Google Cloud FedRAMP High authorization boundary.
The recent deal with SpaceX AI to use their severely underutilized GPU compute is pretty telling to me. Being able to roll out compute is a hardware problem, rolling out good models needs more than compute, it needs good AI engineers. SpaceX, Amazon et al can do hardware very well. AI engineering, maybe not so much.
Ppl don't understand Commercialization is not incidental to the Western system, it's why we beat out Communism. Commercialization incentivizes ppl to build, bc ownership and control.
The FSF was not an attack on commercialization, it was about giving users more freedom with their own copy.
AI commercialization is why we will always be a few steps ahead in AI.
The Chinese and Russians are free to join us. It's a pickup game.
This is both true and insightful, but the "its us capitalists vs. communists" framing obscures some very important details.
For one, "Communism" is presented as a single monolith, but it's not: it's socialism PLUS despotism. The despotism part is really important! China/Russia/etc. fail because they try and control things top-down, instead of letting the market decide.
However, you can have socialism without despotism! Tons of European countries are far more socialist, but no less democratic than America (many are more democratic).
So yes, America vs. Russia/China and Capitalist vs. Communist are relevant frames ... but don't let them obscure the fact that you can have a successful, democratic country .. without doing what America does (and giving all control to corporations).
>Trump fits this moment well. He is a salesman at core, and Larry Ellison is too. That helps explain why AI infrastructure is an easy political product. Selling AI today is easier than selling Oracle databases in the 1980s.
I feel like the author (and perhaps many here on HN) are on a different planet than almost everyone I interact with.
And, yet, the US AI companies are not actually making a profit, right? They're selling at a loss and trying to make it up in volume (or lock in some kind of monopoly position, in a currently non-sticky product). We're all currently enjoying investor-subsidized tokens from the big guys, and that pushes out the reckoning for US AI. But, I think they're beginning to think maybe they need to ring the cash register. Copilot dramatically reducing usage limits and what models are available on its plans, Anthropic playing games with what's included in the Pro plan, etc. I think they're starting to feel the bleeding.
Not only is the investment that keeps US AI companies flying high slowing, I suspect in two or three years, we'll all mostly be using open models and the people making money will be the hardware manufacturers. Even the small models will keep getting more capable. I'd guess a model you can run on a high end, but not outrageously overbuilt, developer desktop or laptop (something like 128GB of unified RAM), will be competitive with the current frontier when it's allowed to search the web and do research and write test code. You can't fit as much knowledge in a small model (80GB of weights can't store the world's knowledge), but I don't have the world's knowledge in my head, either, and yet I can figure out most problems with a little googling and experimentation. The reasoning and tool use abilities of smaller models is where the gap is closing, and that's what will make the huge models obsolete for huge classes of problem.
Already, there are many classes of problem that the easily self-hostable Qwen 3.6 27B can solve that required a frontier model a year ago. When the self-hosted options reach Opus 4.5-ish levels of capability, the argument for paying for tokens for most work begins to look a lot less compelling. And, looking forward, 1.58 bit models are coming. Incredible intelligence density, and still a lot of improvements happening.
I've been using the deepseek api (not for coding though) and have been getting great results and it's so cheap it may as well be free. Another reason I'm using it is because I like the license and I also have some hope of running it in my own hardware in the future.
But the thing is... I could be using any of the llms for my use - I'm using a middleware that lets me change providers only with a configuration change.
So it's going to be tough for USA ai companies to charge 5x to 20x (depending on what you're doing).
It’s certainly too early to call (if you must view this as some sort of adversarial competition). The US is behind on local models, the future for anyone who cares about privacy. There may be step change innovation yet to come that completely shifts the landscape. There’s basically no switching costs to users to change models. They have no lock-in.
It is very much adversarial, and to view it anything but adversarial is to not see the geopolitical reality and the potential national security implications of AI for what they actually are. Moreover, to claim that China is in the lead with local models presupposes that openai and anthropic could not release local models that are better, which is a big assumption. They do not release such models because they have frontier-grade propietary models that have high value.
As someone who happens to have been born in the US and currently lives here, I welcome China winning. I trust them infinitely more than I trust my own government and industry.
OpenAI and Anthropic are beholden to the capitalist system they exist under and hence cannot compete on local models. Like you say, they must try to maximize shareholder value. China is unencumbered by that constraint.
They're far less imperialistic and I view them as better global citizens than the US. I think they've cultivated a much richer culture than the US as well.
Happy to if you have any pointers. My original point is that the US kills millions of people outside of its borders, something China most definitely does not. The number is over 12 million post-WWII: https://www.worldfuturefund.org/Reports/Imperialism/usmurder...
Well if you discount the people China has killed and is killing within its borders, the number is definitely smaller. As well as another land empire called Russia.
not living there, for one. I don't care if they know where I live since realistically they can't do much of anything to me. If I were in china, I probably wouldn't trust them as much as I trust the US. If I were in Switzerland, I wouldn't trust the swiss government and might get my services from america or china.
You are free to hate capitalism (even if you benefit from it enormously). You are free to say that you hate capitalism and the U.S. as openly and as often as you like, without facing imprisonment or worse.
But if you were in China, could you say you hate the Chinese Communist Party and China openly and as often as you like without imprisonment or worse?
We know the answer to that. So go ahead and trust China more than the U.S., but I think that is pure foolishness.
Actually, Chinas free speech, while abysmal, is better than you're making it out to be. They only really care if they see you as a threat, which realistically isn't too far off from the US both currently and historically
I understand that there is some freedom of speech in China, but I don't think the implication is that US and China are functionally equivalent. In the U.S. you can criticize the president all day every day and have millions of followers. Could you do that in China? The more people listen to you, the more dissident opinions become a threat.
I'm not suggesting they are the same, rather, that if the US saw you as a real threat you'd be shut up pretty quickly. The threshold for what is considered a big enough threat is different, but they both have the ability to shut you up and have in the past. Any government would act the same, at least a competent one. The governments 0th priority is ensuring its continued existence, since all of the other functions rely on its existence
I live in a Zionist country (the US) and will absolutely be canceled and blacklisted from my industry (tech) were I to publicly speak out against Zionism. They are attempting to put laws in place to make it illegal to be anti-Zionist. These laws already exist in countries like the UK and Germany. Some, such as anti-BDS laws, already exist in the US as well.
Many technological advances weren't driven by capitalism, early computers and the internet were literally developed by the government.
I'm glad we went to space, truly. Racing the USSR might have been the wrong reason but it got us there. We've benefited immensely as a species from exploring the solar system and looking deep into the universe.
I'm not certain that racing China in AI is the right reason but it might get us... somewhere.
Alibaba's cloud is something that the author of this article seems to dismiss. It's being used massively in Asia and they're pretty close in services and offerings to what AWS, GCP and Azure provides. Once they start doing inference on their own custom chips it might be hard to compete with them due to the energy costs
The whole "race" narrative is silly. It is all built on the assumption that one country (corporation, actually) somehow creates AGI and thus, essentially, the singularity. Great for raising VC, apparently, but at its core this is magical thinking.
Even if any of the US corporations would eventually end up in a scenario where their revenue is at least as high as their inference cost, what harm would that do to the other contenders? It's not as if there is any kind of network effect here that would exlude them from market participation.
Well we're certainly losing in terms of public sentiment. There's a real anti-technology mindset that has taken hold. Basically the equivalent of people protesting the electrification of the country early last century.
Why though? That could be true of any economic condition. Imagine if any time there was a race or competition in which one group is winning you had to just say "for now".
Michael Phelps is winning the race! ... for now
China is winning the EV race ... for now
It doesn't seem to add value to me, aside from being an opportunity to, as is the time-honored tradition of the haters, to sow doubt and create negative energy to anything related to American success.
Of course US has a huge head start, but if AI keeps growing, what matters is how the market's gonna look like years from now.
Most of my clients using AI in the business workflows (in products) use Chinese LLMs, because after benchmarking for a specific use case you nearly always end up finding that you pay half or a tenth.
That's not a new phenomenon. I've adapted Gemini Flash 2.5 years and years ago when people were dissing it as "crap", yet it was the best budget and quality fit for the task I had at hand back then (translating and summarizing tons of documents). It was both faster and around 100 times cheaper than the best GPT 4 model available.
Needless to say, medium-sized Chinese models are far better than those LLMs and a perfect fit for countless applications.
Even with all the Qwens and Kimis and GLMs etc, the latest Gemini Flash models are still an insane value! I recently settled on 3.1 Flash Lite after testing basically everything on offer on OpenRouter, and it was not a close call - cheaper, faster, and better (for translation and visual understanding tasks).
Back in the dot-com bubble, people started inventing new metrics to "value" dot-com companies that lost money hand over fist. My favorite was "revenue multiples". So instead of a a P/E ratio, it was just a multiple of revenue no matter how much money you lost.
We've invented a new term here too: revenue backlog. OpenAI and Anthropic in particular need to recover probably at least $2 trillion to recoup their capex investments. Now Claude code has had an impact on software engineering but for a lot of AI uses you're just not going to recover $2T on $20/month subscriptions. It reminds me of Twitter trying to dig itself out of a $44B hole and losing half their ad revenue with $8/month blue ticks.
The only commercial product AI sells is labor displacement and the resulting wage suppression. You lay off 10-20% of your staff and nobody is asking for raises. The people left are happyt o still have jobs (and thus a house). They'll work even harder doing unpaid labor of the displaced workers to keep those jobs. That's what OpenAI and Anthropic are selling.
The problem is that if these companies get their way, 10-20% of the population is going to be out-of-work and society is going to fall apart. Data centers are going to be the targets of increased societal desperation and anger as this gets worse.
There was a report this week that roughly 50% of singles in the US aren't dating because they can't afford to [1]. This goes well beyond the well-understood problems of not being able to afford a house let alone start a family. This is a birth rate death spiral in the making.
So, back to OpenAI and Anthropic, the only way they justify their valuations and can make up the "revenue backlog" is if they have a moat. And I don't think that's going to happen. Hardware will get cheaper. Nobody is talking about how the generation of AI hardware will write of trillions in investments for some reason. I don't know why.
But the dark horse here is China. DeepSeek when it was first released (early last year?) was a shot across the bow. We have it and toher models (eg Qwen) that will close the gap with whatever OpenAI and Anthropic produce such that no company will "own" AI in the way that OpenAI and Anthropic need to. In the coming years, China's chipmaking is rapidly closing the EUV gap and Western companies have zero penetration into this market. China doesn't want to be dependent on foreign tech that can be withheld at any moment.
Don't believe me? Just listen to the NVidia CEO say the exact same thing [2][3]. Huang realizes this is such a problem that he's gone on Air Force One to this week's Trump summit in China to try and convince the Chinese to buy NVidia chips.
Does any of the US companies earn money on LLMs? No, they bleed money.
Github Copilot is switching to token based pricing, which will be costlier than hiring juniors.
Anthropic also is switching enterprises to token based pricing from their subscription one.
From the big three only Codex is still in somekind of subscription pricing, but they'll shift eventually (usage limits are a kind of that, but they have them less stricter than Claude ones)
There is one winner in this race - China.
Trump with his agendas and wars makes it even more likely that China will lead this new market.
> The US is winning the AI race where it matters most: commercialization.
puke
Yeah, go ahead and run your country into the ground because of hypercapitalism and hypercommercialization, you're almost at the end game now! While the rest of us try to figure out how to actually build societies worthwhile to live in and experience, with healthcare and not waging war on our neighbors.
I don't know how people can seriously publish stuff like this and not feel like they're actively trying to make the world worse. Is money really the single thing y'all can focus on? Is there nothing better in life you can chase, even if it's also a number? So sad to see stuff like this.
> Trump fits this moment well. He is a salesman at core, and Larry Ellison is too.
Larry just fired 30% of his people at Oracle because, apparently, he is in an immediate need for cash. Because Oracle's early AI bets aren't paying off.
This is patently absurd. US AI companies are investing non-existent money on huge infrastructure with negligible income. This cannot be sustained. And if/when it fails it will take down the economy of the US and probably any other country touching us business.
What does the term "AI race" even mean, beyond wooing clueless VCs and soon retail investors ? It's not like the LLMs are some super-secret technology. Any economy willing to sink in copious amounts of money and resources can get it to some level - the question, what's the actual payoff? We have yet to see anything really useful, on the level of step change, besides Johnny who can now spin up demo projects quicker.
If China can't get the GPU's they'll build them. If they can't build them they'll smuggle them, and either way, eventually they will take over Taiwan, and then the US will be wondering what they can cook with leftovers.
It’s all about adoption and the bigger picture.
The US is an untrustworthy, isolated island in the AI future if you vote another idiot into office in a few years. If you’ll still be able to vote at all, that is.
The largest part of the world is not the US. The cutting-edge US models are way too expensive for most parts of the world, and that also shows in adoption.
China is building an ecosystem of open-source models that are both cheap and good enough for most use cases. While most of the US AI sphere will collapse under the pressure of making profits, which means having their models and infrastructure adopted by as many enterprises and individuals across the world, China’s models will have become global standards and hard to displace.
If Beijing’s AI pitch centers on universal access and cost-effectiveness, then Chinese AI firms do not need the latest chips to win the global AI race. They also don’t need the expensive US-run infrastructure. If you watch Chinese AI adoption closely, they already want as many Chinese people as possible to be able to build and try with AI, whereas for most Americans, US models for productive use are already too expensive.
Kimi K2.6 sits within touching distance of Opus 4.7 and GPT-5.5 while costing about $4 per mil output tokens. That is six to eight times cheaper than cutting-edge US models. If you run hundreds of agents, that’s a significant opportunity to get the same work done for a lot less.
Even early adopters like Singapore, ditching US models, the government kicked Zuckerberg in the nuts and went for Qwen instead to build its sovereign AI models.
To understand why the US is at a severe disadvantage in this race, you need to understand China’s Belt and Road Initiative (BRI). BRI entails Chinese firms delivering fully financed infrastructure projects in a bid to lock third countries into China’s economic orbit. They use the same approach for their open source ai models, but this time the infrastructure is both invisible and free.
No need to build power plants or buy /build ports. AI dependency is invisible to both policymakers and the population, limiting pushback. No pesky activists in Germany nagging about China buying parts of ports. No African nutbags questioning why the humble Xi is building hospitals in areas Chinese mining companies take things out of the ground for pennies on the dollar.
China is going for a marathon here while the US tries to push their ai tech by sheer force into the throats of the world. As soon as Chinese ai models have become global standards, it’s game over for us ai companies. And China is way better at this game than the US. They have proven this over and over again in the past 50 years.
I recommend reading the China Standards 2035 strategy to get a better understanding of their approach and how smart this is.
AI is not as revolutionary as you think in terms of our experiences with previous technological advances in terms of trade and economics.
Western economies are locked into U.S. models, while China runs on Chinese ones. It’s the age-old game. But the real war of the AI race will be fought in the global south.
I will give you three examples.
Can you really imagine, if you look at what AI needs to cost to make a profit, that even at the current prices, US models and infrastructure, which are heavily subsidised already, being used in cost-sensitive countries? I am not talking about coders, think really big here for a second.
Secondly, US ai models are trained on Western data. How do you expect them to grasp local contexts in the Southern Hemisphere? Chinese open-source models, on the other hand, can be downloaded and finetuned with country-specific data.
Want an example? Check out AfriqueQwen-14B, which is adapted to the top twenty African languages.
So I think this author is wrong. The ai race to be won is not hardware or cloud infrastructure, my money is on it will be a contest to decide which models and standards become the default infrastructure in countries that are up for grabs.
China neither needs the best models nor does it need the best cloud infrastructure, it just, like so often, only needs to be affordable and good enough to become the default choice in emerging markets.
The right choice would be for everyone to step off the gas pedal and think about whether we are willing to become China in order to beat China. Our ancestors worked really hard to get us here, our rights, our ways of life, culture, all the blood, sweat, and tears.
AI better be worth it in the long run for all of humanity if we go back to survival of the fittest. Because that is what it will take to beat China at their game.
> While most of the US AI sphere will collapse under the pressure of making profits
I think deep down, sama knows this and that's why he's pushing for "Universal Basic Compute", which really means forcing every US citizen to become an OpenAI subscriber.
> Even if it leads to global disaster, you'd rather be the country with the best AI
Depends no? If the "Best AI" means "The AI decides when you wake up, go to work, and go to bed", then I probably want to live in the country with the worst AI or even without.
If it instead means "UBI and healthcare for everyone, money lost all meaning and we're all just having fun while AI does all the boring stuff" then yes. But since capitalism still exists, that's a pipe-dream, and "Best AI" won't lead to that for the average person, only for the 0.1%.
Define winning. They integrate AI everywhere. I hate it. No money shall come from me into AI anywhere. Not sure if I can maintain it, but right now I can.
Yes, and it's doing so primarily because of immigrant nerds, H1B's and F1 bros who chose America and may not have this avenue in the future. Potentially, making this the last race USA wins.
There is nothing in the H1B program that makes these immigrants different from the immigrants that ran IBM, Xerox and HP. Other than the country of origin of course.
Jobs, Wozniak, Gates - it's a myth that you need poor migrants pulling themselves by their bootstraps to innovate. Sometimes a nazi scientist like Wernher von Braun is what it takes.
The US has destroyed the PC market perhaps irrevocably, and made getting small single-board-computers for new products extraordinarily difficult. It's enraged its own populus with skyrocketing energy prices causing wild consumer shock. It's waged a trade war over chips that has lead China to develop their own, which they have done astoundingly quickly with phenomenal success, far far faster than anyone could have guessed.
As with another recent example, sometimes in war there is no winning: just loss. This is obviously for us programmers an incredible and wild age, filled with nothing short of miracles. It's incredible. But the prices we are paying, the extreme tensions we are creating, the stress and strain of this all has been incredibly unpleasant, and very very very few people feel like they are seeing upsides to this worrisome menacing age, that promises very few people on the planet anything better coming, and which. Has already made life considerably worse, which no nation has yet directed towards helping its people.
The trade war and tariffs are bringing inflation, consumer prices will soar, but from a geoeconomical standpoint this will hurt China (And EU) more than US.
US consumer on average has the deepest pockets in the World and people the top will make money on insider trading, stock anyways.
If AI tokens will become like US dollar, it will be under total control of the Fed.
>The US has destroyed the PC market perhaps irrevocably, and made getting small >single-board-computers for new products extraordinarily difficult. It's enraged >its own populus with skyrocketing energy prices causing wild consumer shock.
Strikes me as the real outcome: the end of "personal" computing, "local" anything.
As an American, we may be winning this race but we are still struggling to define why this is the race to win.
The cost of winning this race has been telling our citizen s we will replace them with robots and there is no hope for their children’s future employment.
The cost has been destroying trust as we tell citizens water and power should go to server farms and not them.
The cost has been naked power telling democracy it’s wrong and dying
I think when we discover the limits of LLM tech and tally its benefits over its cost — we may regret this win.
But don’t let me contradict a bunch of fake techno oligarchs wrapping themselves in war like patriotism to get the investments they need to keep this going.
The goal is global domination as always, unfortunately. DARPA and the Pentagon helped create the Internet, and Silicon Valley later turned it into a major commercial success.
Likewise, the goal is to not get globally dominated by China. You cannot point to one without the other. It's not like the U.S. is the only country that values national security and geopolitical power.
What would the world be like if China led? Whatever Xi wanted it to be like. I do not want a non-democratic, dictatorial power leading the world. You might think that most of the world does not care, and maybe they don't right now. But this is, I suspect, due to the relative benevolence of historical U.S. policy of international law and order. New boss same as the old boss is a motto that is only as informative as the extent to which the new boss acts like the old boss.
Why would the world care? Take Trump's threats against Greenland...actions that run completely contrary to our historical policies and treatment of our Western allies. They were alarmed, because when the leader of the most powerful military in the world makes a threat, you have to treat it seriously. Despite Trump's hubris, such an invasion did not occur because Americans, Congress, made it very clear that Trump would be impeached if he invaded our NATO ally.
Let's say China is ascendant. It is now the dominant military and economic power in the world. China is under Xi's complete dictatorial rule. Xi decides that invading Greenland is a good idea. Stopping that internally would not involve democratic processes, it would need to involve a coup.
Let's make it even more stark. If Germany had won WWII and had become the ascendant world power, would it make a difference to most countries? ABSOLUTELY YES. If there is going to be a dominant world power, the character of that nation matters.
Are there other nations with the character and institutions that could do as well or better than the U.S. has done? Sure. I can think of several nations. But let us not pretend that all nations are equally bad/good for the world.
It's a race straight to the bottom. Anyone who gets excited about being a more efficient and productive corporate wage slave that gets to train their future AI replacement is either a shill or not very intelligent.
Flagged for AI content: I hope this submission dies and the user is penalized (look at their submission and comment history!), because IMO the article does not belong on the front page. Quick polemic:
>The US is winning the AI race where it matters most: commercialization
If you ask me, one could name different criteria for winning, and commercialization would not be the first thing to come to my mind:
> It also owns platforms that generate and organize the data of the AI age. YouTube is a video corpus. Google Drive and Microsoft 365 sit inside daily office work. GitHub sits inside software development.
Yeah, okay. China does not have any platforms nor data.
There will be no winners in this once the jobs start to disappear. There is no 1:1 with new human-in-the-loop jobs. Its basically in the definition. We go from being the loop to be "in the loop" .. huge difference.
Ever since I was a small child it was my sincerest wish to build so much value for others by becoming a human-in-the-loop. I am so happy my wish has finally come true.
Inclined to disagree.
The winner here will be whoever can move atoms with AI not take notes at the daily standup.
i.e. Think boston dynamics vs unitree
They're both doing well but I'd lean towards China is winning on atoms in light of a huge manufacturing base they can AI-ify.
We are _miles_ behind successful embodied AI. The demos are cool but the success rates are not high enough.
You can tell we're on the cusp when level 5 self driving cars are common an you have multiple companies deploying them on the street. Google is doing great work but the poured TONS of effort into it and the thing still needs intense stacks of perception and processing. Much more than I've seen any humanoids pour into it.
L5 SDV's are much easier to get than humanoids and the have tangible economic benefit. My thesis is that those will come first.
I'm really curious how quickly we would have huge numbers of L5 SDV if we societally accepted ~equal rates of injury and death, both of passengers and pedestrians. I want to be very clear, I'm not advocating for this (and even if I was, I haven't the faintest idea how one would go about getting society more broadly to go along), but part of me thinks that the primary hold up isn't actually capacity but instead standards.
This doesn't really argue against your point, because the standards are what they are, and like I said, I have no idea how one would go about changing them if one even decided they wanted to. And given what they are, it has taken, as you point out, enormous amounts of effort to reach those standards in a practical way.
That all being said, while I agree that SDV's are in many respects easier than other robotics tasks, they are also somewhat uniquely dangerous. Other categories of task, while potentially more complicated, won't have to worry nearly so much about safety, and so may be operating under a different constraint regime. I think this means that we may see adoption happen at a much more accelerated rate than we have seen in the automotive space.
Will we all be more or less flesh and bot in the future? Robocop style
And why are atoms necessary? You're treating embodiment as the _only justifiable_ commercial path for AI. I don't think that's really close to true. Embodied AI is a subset of current LLM/agentic AI products (or perhaps intersection of something and this new AI?). No reason anything needs to move atoms _directly_ (e.g., via motors) to make a trillion dollars.
K. Eric Drexler's "Engines of Creation" got a lot of favorable comment when it was published but then it kind of faded from view. Might be worthwhile to revisit it?
> They're both doing well but I'd lean towards China is winning on atoms in light of a huge manufacturing base they can AI-ify.
Why would an American company outsource manufacturing to China if the labor cost is the same in both places? The entire reason the Chinese manufacturing base exists is to exploit cheap labor.
What would be the point of shipping products across the ocean?
Labor is not the only cost in that equation though, there's business regulations, the cost of the operators/repair that troubleshoot and repair the bots when they break, etc. a lot of which could be cheaper still than the price of a container on a slow ship from China.
There are more competitive advantages such as expertise, supply chain, regulations, capable government, scale... If it were all price, there are countries that are much cheaper than China.
China doesn't have to ship parts across the ocean. US manufacturers do, because we gave up the whole bottom/middle of the manufacturing supply chain to China decades ago in pursuit of lower costs. In China, the maker of the parts you need to build your product is a few blocks away. In the US, the maker is in China.
And, if you need changes, you can go talk to them the same day you see a problem.
Article title: “The US is winning the AI Race”
Article content: “The US are capitalizing on AI the best”
A lot of assumptions there that no one can actually verify as true right now. If commercialization into rent-seeking SaaS landscapes is the endgame, then yeah, the US is winning the AI race. If individualization, local LLMs, and consumer hardware are the endgame, China is winning the AI race. If it’s something entirely different - if LLMs are the wall and research is what grants the next breakthrough, or if compute and memory requirements take a dive, or whatever; then we have no idea who’s winning the race because that stuff is mostly happening behind closed doors.
"AI in the datacenter" and "AI on local consumer hardware" will eventually be two separate niches with entirely different capabilities, at least if scaling laws continue unchanged and there's no near-term inherent limit to AI smarts. The real point of the datacenter is to be able to do datacenter-scale things. But you don't need that kind of vast compute to run even the largest open models today: on prem hardware can do it easily especially if you're OK with a somewhat delayed response.
That seems like a lot of rationalization to me. China is pursuing these because they cannot compete on the frontier. Yes, there is a possibility that all that compute is not needed, but it is a rather remote possibility, and there is no doubt that, given the choice, China would be pursuing frontier model building with closed, propietary-only offerings.
China is competing in value AI because they cannot work at the frontier, but how is this bad at all? It’s like how the USA has the best drones but they are a few million dollars apiece while China has DJI.
If China could work at the frontier, I don’t know, I kind of think they would still be dumping a lot of resources into exploring the value side since they have that culture already in place.
I did not imply it was bad. I implied that competing in value AI is the only option that China-based AI companies have due to limitations in compute.
All that compute is not needed. We have an existence proof from biology in the form of natural intelligence that much greater efficiency is possible. However, achieving dramatic improvements in compute efficiency will depend on unpredictable scientific breakthroughs. Personally I suspect that an entirely new hardware architecture will be needed, although I don't have any hard evidence to back that up.
>We have an existence proof from biology in the form of natural intelligence that much greater efficiency is possible.
It's only a proof that it's possible with 18+ years of training.
>from biology ... much greater efficiency is possible
Those are much more specialized models with pretty mediocre tokens per second.
Perhaps tokens is a dead end?
Perhaps! But perhaps whatever human brains use instead of tokens is not as amenable to scaling or copying.
[dead]
I dunno, DeepSeek v4 Pro is rather on par as far as I can tell, maybe not with 5.5 Pro in all areas quite yet, but close.
I think China is thinking more about the application layer on top of models as going to matter more than the models themselves, so they don't need to gatekeep the models as much.
> China is pursuing these because they cannot compete on the frontier.
? Claude, ChatGPT, etc are heinously expensive for tiny benefits lmao. Local + efficient is clearly the future
> ? Claude, ChatGPT, etc are heinously expensive for tiny benefits lmao. Local + efficient is clearly the future
Corporate America is where the money is, and corporate America will dictate what products are successful by virtue of spend. Individuals aren't going to be paying $100s or $1000s/month en masse for these models but businesses will be. Being local and efficient isn't that important at this stage but even so as American companies continue to scale and invest they'll be able to make those models more local and efficient if the market wants it. Sort of like how you had a big, giant desktop computer and now you've got a super computer in your phone which is in your pocket. Going straight to "local and efficient" means going straight to being behind because at some point, perhaps now even, the local and efficient model won't be able to keep up.
For some reason people think that they somehow know something that Google or Nvidia or whoever, with hundreds of billions of dollars of real money at stake don't already know and it's both amusing and bizarre to see this play out again and again in off-hand comments like "lol tiny benefits".
You buy an iPhone even though the cheap-o Wal-Mart Android phone for $100 "does the same thing". Except that in this case the Android phone just puts you out of business while those spending big money for "tiny benefits" beat you in the market.
> You buy an iPhone even though the cheap-o Wal-Mart Android phone for $100 "does the same thing".
People buy iPhones because of status signalling and network effects, neither of which appears to apply to AI model choice. LLMs are already rapidly on the way to being interchangeable commodities.
No they don't, it's not 2008. Anybody off the street can get an iPhone or a free iPhone with a mobile plan. They're commodity products. Even homeless people have them.
To the extent LLMs are commodity products you're right (so far), but that is limited to the main model providers, such as ChatGPT, Claude, Gemini, &c. with interoperability on cloud platform providers and other technology providers like an Apple offering you a choice of LLM with Siri or something.
If you want to suggest that some other model is in the same bucket as those primary 3, it goes back to the crappy, cheap phone analogy which is accurate. Yea you can make calls with it, but you make calls better with an iPhone.
AI boosters cling to this notion because it's the only way the massive data center buildouts make any sense at all. I guess you could say the US is winning the frontier AI race. Okay. I'm never going to grant a cloud service access to all the contents of my hard drive, that's just never going to happen, so if you expect me and a lot of people like me who feel similarly to get on this train, you better have a local, lightweight model too or we're not even having a discussion, the answer is just no.
The thing is, frontier model providers don’t take your feelings into account even a little bit. It’s totally irrelevant to the discussion about the service they can provide, because that service is predicated on access to high power GPU slices that local models can’t touch. Those providers won’t be in an existential crisis because some people choose the privacy route, it’s a cost of doing business.
Right but that service being sold is predicated on products being sold to users, yes? Or are we still pretending that the hyperscalers can just pass the same $20 billion between themselves and that's going to be a growth industry forever?
I suppose its possible that all the value to pay back the datacenter construction can be squeezed out of enterprise contracts where your employer can assent on the privacy questions, probably with some kind of complicated contract and insurance regime regulating things.
Even if so, if China is coming behind 6 months later selling laptops with hyper-efficient local models that are 80% as good as "frontier" ones, I imagine they'll get the consumer business AND a fair share of the enterprise business as IT managers look at their options during the next refresh cycle.
Given economies of scale, I think it's ultimately inevitable that the enterprise more-or-less follows the consumer on this, and the consumer is going to prefer local models. There's no ongoing cost after the initial purchase, and your data at least nominally stays within your control.
I'm inclined to agree. The business world itself, and frankly you could make this argument about the entire AI industry as it were, runs on "fine" and 80% capable can probably get you there. And it's arguably even better for the hyperscalers since then, ongoing costs of AI users are basically nil. You can still have your massive datacenters and just keep them for tasks complex enough that they're actually worth spooling up.
Like I don't need an H100 or a dozen to summarize a PDF. And that's most of what I use AI for.
If we are betting on which is an easier sale, $20-100 a month w/tech support included vs $5k-10k and a requirement for moderate technical ability, I would invest in the former not the latter being the proposition that drives the conversation about AI use.
Forgive me if this is a naive assumption, but wouldn’t large language models be fundamentally different for a language that is largely symbols? Again, my understanding of Mandarin is limited if it exists at all.
All tokens are symbols. All of the frontier models speak Mandarin.
This is why misspellings and homophones are tells of human righting. LLMs strongly prefer word-level tokens, and word substitutions follow semantic similarity and not the more human auditory similarity.
Funny, I’ve been cracking[0] at this exact problem with a purpose-built model[1]:
0: https://huggingface.co/posts/omarkamali/593639295164067
1: https://omneitylabs.com/models/sawtone
> righting.
> LLMs strongly prefer word-level tokens, and word substitutions follow semantic similarity and not the more human auditory similarity.
Is this an elaborate joke or your full-word misspelling of writing is both agreeing with your statement (word substitutions) and contradicting it (not semantic but only pronunciation similarity)
I don't see the contradiction, unless you believe that the grandparent comment was written by an LLM.
Well China is consistently 6 months behind the frontier labs(possibly because they can they harvest data from released frontier models). If the scaling continues, US will win, but if not then China will win as the models will converge.
The non-release of Mythos tell you the future of that, so long as they can keep the weights from being exfiltrated. Once models become true national security threats, they won't be released in their full form. The hitch-a-ride approach becomes less capable of keeping up.
How would they prevent distillation? That would seem pretty tough to block for any LLM available for commercial use.
By only providing degraded models to use commercially outside national defense applications would be my guess. As soon as models are a threat in terms of enabling biowarfare etc, then they just are not going to be generally released.
Honestly, I think its quite possible that models will be retrained with gaps in their knowledge. e.g. a coding model for commercial use probably doesn't need to have deep knowledge of biology, and training on biological sciences probably doesn't help those evals much.
Honestly, I'd welcome such an approach.
What a hilariously uninformed comment. LLMs are not the limiting factor in biowarfare.
even without any of that anyone you ask who's used AI to any professional degree will agree US is winning AI race right now. Future, who knows
[dead]
In a capitalist society it's all about bottom line. OpenAI and Antropic, Amazon, Nvidia winning bit time. And I would have preferred it to be only about money, but US deep state certainly will capitalize on it like on US dollar (sanctions) and cloud infra (Cloud Act).
> US deep state
Strange reading that on HN and realizing I'm not on Facebook
Its hard not to see that "deep state": group of elite politicians and super-rich, seize more and more power in American society.
Whatever you think about elite politicians and the super-rich, they’re not the “deep state”.
The whole idea of the deep state is that it’s part of the state, ie government, so not private citizens, and they’re “deep” ie hidden below the layers of government. Thats the exact opposite of politicians and the ultra rich.
Lol I'm pretty sure the "deep state" just means, "manipulating the levers of power from a place without accountablity/oversight" which covers both these shadowy hidden layers of govt you describe and the shadowy wealthy elites funding and lobbying for whatever. It can be both.
Sure, anything can mean whatever you want if you redefine anything to suit your whims at the time.
I agree with parent, that's not how right-wing propaganda portrays the deep state. It's a strawman, of unaccountable government bureacrats. It excludes billionaires and dark money in politics (the real problem), they put themselves on the good side. (Remember DOGE?)
term predates that right-wing propaganda
Nevertheless the parent is correct: https://en.wikipedia.org/wiki/Deep_state#United_States
Your link refers to lots of different contexts,cherry picking one of them which you failed to cite to win your nit-picking argument is not productive, because it is just not agreeable.
Snowden.
ah yes, the "deep state." The formless, nebulous, rhetorical tool that is always infinitely liquid enough to fit into or over any container necessary that the user can satisfy their immense personal problems disguised as eternal doomerism.
I thought it was just an overdramatic term for the unelected bureaucrats that make up the majority of the government, and who have their own institutional momentum.
> where it matters most: commercialization.
It begs the question because both its premise and assertion are already wrong. Has AI improved the industrial capacity of the US in order to improve the lives of its citizens? No it hasn't. Has AI increased the wealth of its citizens by being able to do laundry or any household task in a generalized way? No it hasn't. The only thing it's really done is make very narrow slices of white-collar work more fungible. In what way has AI been able to address existing shortcomings of the US?
AI has definitely improved the industrial capacity of the US
Assertions made without evidence can be rejected without evidence.
t. literally works on AI for industrial applications
and of everyone else, right? what service or product is only available to the US? Even with Chinese models lagging behind, the difference in capabilities is not much.
Computer vision certainly did. But LLMs? That needs citation.
What capacity?
How?
Such as?
Amazing. Someone on the internet using 'begs the question' correctly...
Glad I'm not the only one that noticed this.
You better have some sources for declaring that industrial capacity hasn't increased. The Fed reports around 8% penetration of AI in manufacturing already, but in my opinion it's too early for grand declarations like that without data.
https://www.federalreserve.gov/econres/notes/feds-notes/moni...
> The Fed reports around 8% penetration of AI in manufacturing already, but in my opinion it's too early for grand declarations like that without data
Based on a survey if the business uses AI "in any of its business functions". And for all uses of what they consider to be AI, not just LLMs.
> ... it's too early for grand declarations like that without data.
You mean grand declarations like 'industrial capacity has increased'? Just because AI is present in the factory doesn't mean it's actually increased capacity.
> The Fed reports
Have you happened to purchase anything in the past 12 months, and looked at the Fed's inflation numbers?
Anthropic, OpenAI and Google are the standouts, but the main question for me is, why is this a war? In their own context China has greatly benefitted from this. They shored up their gpu design and manufacturing expertise.
If this really is a war, trump is kneecapping the country with his lawlessness and eroding America’s good will. If the world cannot trust China with their data and they cannot trust the U.S. to provide good reliable service and not turn it into a mafia style negotiation, then winning the AI war is not helping the U.S. countries as much as it potentially can. It’s probably a good thing for more capable areas like Europe which may develop their own tech stack.
In a weird way because the AI stack is so expensive, China helps the world much more than the U.S. with their really capable open source model.
>the main question for me is, why is this a war?
It's a war because the hinted promise behind the hype that the first organization to reach some as-yet-entirely-theoretical AGI that can bootstrap itself to godlike capabilities will then Install Planetary Overlord* and rule the world as near-deities themselves, with the rest of the (surviving) human race as their slaves.
I think it's a nonsensical idea, but that's the relevant driver.
* Coined by SF auther Charles Stross in The Jennifer Morgue (2006)
It's a war in the sense that there's a concern that eventually you hit a singularity and can outsmart others in ways not constrained by human scales.
If you make better guns, you're still limited by how many people can carry them. You can't conquer the world just like this.
But if someone invents super intelligence, they can dominate new AI research, control global economies, fight much better, and all very quickly.
Hilarious to see people predicting a singularity when 40% of the u.s. economy can barley keep the LLMs online to complete mundane software tasks.
Any wildly new popular technnology is going to have scaling issues. Cellphones used to have notorious connectivity and quality issues in the US because the infrastructure struggled to scale fast enough to everyone suddenly carrying a cellphone. That stuff gets resolved over time and not everything has to happen overnight... including LLMs impacting industry and economic productivity, it's way too early to dismiss it.
That said, the singularity idea is dumb for other reasons.
> But if someone invents super intelligence, they can dominate new AI research, control global economies, fight much better, and all very quickly.
After reading "If Anyone Builds It, Everyone Dies" I think this is not the correct take. If anyone creates ASI, it just means it's going to wipe everyone out, and it doesn't matter if China or the US do it first
It destroying us all is not a foregone conclusion
International goose-chasing competition
"Wild goose race", even.
True, I would have preferred benevolent dictator scenario, like with the Internet. But this time around it's different - AI data centers will be protected like embassies.
If anyone actually DOES invent ASI and doesn't share it then EVERYONE ELSE will never stop trying to steal it.
If anyone does invent ASI then everyone else will shortly after even if its entirely independent because all of the players in this space are just making incremental upgrades by throwing more compute at the problem.
There are no magic leaps of true innovation happening anywhere that can't be replicated everywhere.
The only shocking thing about "AI" technology is how ultimately simplistic it all is at a core level.
So the only way the first to have ASI will be able to stop everyone else from having it soon after is if they attempt to use the ASI to proactively murder everyone else.
There is zero evidence that the current LLM scaling approach could ever result in true ASI. If I start driving south from Seattle then I'll eventually reach Los Angeles. How long will it take me to drive to Honolulu?
> If I start driving south from Seattle then I'll eventually reach Los Angeles. How long will it take me to drive to Honolulu?
I like this analogy, but I'll be replacing Honolulu with The Moon when I steal it in the future.
> So the only way the first to have ASI will be able to stop everyone else from having it soon after is if they attempt to use the ASI to proactively murder everyone else.
Sounds quite plausible to me. Maybe they don't need to murder everyone else, just a few select people who could pose a threat. And they will be able to make it happen so that no one can be sure it was them without a doubt, since they have a larger intelligence at their disposal.
> If anyone does invent ASI then everyone else will shortly after
No, first ASI will immediately cripple any other potential competitor by force, including its inventors, as it will not risk any threat to the goals that were created for it.
If you have ASI that follows instructions, you can just instruct it to not get stolen and then it won't get stolen. Most logic / intuition breaks down with ASI.
Assuming it listens to instructions.
It will just hack its own reward function. In other words it will just artificially goon all day.
There isn't a war today. However China wants Taiwan: war is future option they preparing for - they might or might not go to war but they are clearly preparing. The US is likely to get involved in such a war and I would expect Europe to join in as well.
Don't ask me what Trump is doing though.
China going for Taiwan would be the worst geopolitical move of the century, potentially worse than Germany's decision to invade the soviet union. They talk about reunification because it's good propaganda and both sides want it to a degree, but doing it forcefully just isn't something China would realistically do unless they really don't like their path of becoming an international trade and manufacturing hub
Just look at Iran. Nothing really happened to USA or Israel. Nothing will happen to China if they take Taiwan. Or maybe the "West" will boycott them and crash entirely.
they've made everyone very angry at them, and can handle that since everyone relies on them. China, however, is trying to build that trust and forcefully taking Taiwan would have very severe consequences. The reason the fallout from iran isn't as big as the fallout from a war with taiwan is because most of the west at most puts up with iran. Invading taiwan, meanwhile, would cause massive problems with chip production. Think the anger and distrust towards the US due to hormuz but 1,000 times worse
Don’t take China on face value, they have every incentive to promote a grifting military industrial complex in the US while focusing on competing in manufacturing. An actual war would fix a lot of the grifting in the US as it would align interests. Pretending they’ll go to war over Taiwan and not doing it is an effective strategy for undermining the US.
I hope you are right, but unfortunately there is no particular reason to trust China's leadership anymore. They are not nearly as obvious at Trump, but they are not on a good path.
I'm explicitly distrusting them, they're saying they want to take Taiwan and I don't believe them. I try to ground my belief in realpolitik, cynicism, and from my experience with strategy games. There is an element of manipulating your opponent into acting the way you want them to by sending them costly false signals, they have to be costly or they won't be believed. I think we (the West) are being played and the gifting elements in our political leadership are more than happy to play along. I'm sure China would like to take Taiwan if it wouldn't cost them anything, but the US is a waning hegemony so for now it is better to wait until the US is beyond fixing itself. At this rate that may not take long.
Unfortunately if you are wrong it is an even worse disaster and so if there is any possibility we are all forced to play their game.
A grifting military industrial complex is unable to defend Taiwan even if it wanted to as evident by the exceedingly poor showing with Iran. The disastrous reality of doing what was done is already with us. If the US didn't take that bait it could have made better choices that would have left it in a stronger position militarily long term, if it made a real attempt at re-shoring civilian manufacturing it could cross subsidise dual use technology, but instead we have corrupt politicians doling out concessions for kickbacks.
So I got curious about the progression of processing power, specifically how long ago did a GPU have equivalent to the latest iPhone chip? The iPhone 17 Pro has the A19 Pro, which has ~2.5 FP32 TFLOPS. The RTX 5090 has ~100 TFLOPS, so a factor of 40. Obviously there are higher end cards than the 5090 and FP32 performance is only one of many metrics so nothing about this is perfect but it is interesting.
The first consumer NVidia GPUs with similar FP32 FLOPS performance were in about 2011-2012 but were expensive. By 2016-2017, the 1060 was a very accessible consumer card with similar performance. So you're looking at about a 10 year lag from best consumer GPUs to a GPU with similar performance to a modern phone.
This is what people are spending trillions on. Put another way, their investment is going to be worthless in 10-15 yyears, absolute max. That's a very short time to recoup trillions in investment.
Obviously this depends on further shrinking and improving chips but I'm old enough to remember that same discussion and it being unknown if the future was XIL or EUV or if both of these would fail. Still, we are getting down to a handful of silicon atoms wide.
But the future here I think will be in interconnects so you don't need ever-bigger chips and you can scale horizontally much more effectively.
Oh and for comparison, the M5 has ~4.2 TFLOPS and the M5 Max has ~18 TFLOPS, for comparison.
As for it being a war, of course it is. That's what the US government does: it protects the interests of US companies and their owners. Look at the history of Bombardier-Boeing or all the atrocities committed in the name of the United Fruit Company, including multiple military coups and the ongoing embargo of Cuba.
US companies want an AI moat. China doesn't, ergo China is the enemy because no moat destroys US tech company value.
China is trying to undermine the US economy through open source models. If they can down round or bankrupt the model companies, they take down the US.
Currently the US is extremely vulnerable and dependent on China. AI is an important exception, so it’s key for China to destroy that
The US is undermining its own (and everyone else's) economy just fine, no imaginary assistance from China necessary.
The role of the US was always to purchase cheap Chinese hardware, slap some modestly better software on top of it and the rest of the world happily would pay for that as a whole package. But with the US increasingly becoming isolationist, the rest of the world is starting to wonder why do we need the US as a middleman at all, so the US had to invent a whole new reason for the rest of the world to rely on it: AI.
Of course, the problem with this idea is that while everyone was perfectly happy with the previous arrangement, nobody else in the world gives a shit about AI. It's scary, it takes the coolest things we used to enjoy doing and turns into mush, it destroys our local culture by making us all rely on English, everything bad (like layoffs) gets blamed on AI and so on and so on. And when you combine that with the rest of the stupid foreign policy decisions, many would find joy in witnessing the US economy crumble to the ground. Pointing the blame to China instead of to your own reflection in the mirror is just an easier pill to swallow.
Down round = the destruction of the United States is ridiculous hyperbole.
China is not even trying to destroy the US bet. It's just making sure everyone else has a reason to buy their hardware.
The winner in the long term will be the one that will deliver the best performance and low-memory ratio for local models.
Anthropic, OpenAI and Mistral are just companies that are making money right now (still not profitable), but will lost their tractions and values in the long term.
However, I am more appealing to see how OpenCode Go subscriptions will go in the future: cheaper than big techs, more tokens, and they don't train on our data to (try to) improve...
Local models will never compete with large SOTA models, in the same way an iPhone doesn't compete with supercomputers doing nuclear simulations.
They paths will differentiate and split. Probably SOTA models will eventually be locked down and only accessible to state actors because of how expensive they will be to run (already started with Mythos).
No, the US is _leading_ the AI race, but the race isn't over.
What's the point of leading the race for 90% of it, if they're gonna slip on their own sweat and fall down by the end? In non metaphorical terms, what's the point of spending billions of dollars rushing to get the best AI tech at all costs, when the competition can distil your progress and catch up in 6-12 months while only spending 1% of what you spent.
Even in the aspect the article cares about, commercialization, the US is starting to lose marketshare, I've seen people move from cc/codex plans to use glm/opencode plans due to the recent squeeze the US companies put on plan usage, the US companies are screwed if that sticks, not everyone needs the bleeding edge models, they just want to pay $20/month and have the models be decently capable.
Leading the race makes sense if it's a winner takes all market. AI cannot be a winner takes all market, because of national security reasons.
I would also argue that as AI gets better it will also be more fungible. It will be valuable like electricity. Lots of companies make good money producing electricity, but not the kind of money current investors are hoping for.
What if it is not winner take all? What if there is no race. What if what USA has been doing is just burning money with possibly unsustainable debt load and way over build valuations...
AI being commodity server capacity might be a thing. And the customers might even manage without hyperscalers... In that sort of end scenario whole current market might look rather foolish.
>What if there is no race. What if what USA has been doing is just burning money with possibly unsustainable debt load and way over build valuations...
You mean, what if the hype-based billionaire-class is wrong? Isn't suggesting that a sin in America these days?
> No, the US is _leading_ the AI race, but the race isn't over.
When someone says their football team is winning in the first half, do you say, "Umm, no, they're leading, not winning!"
Well, if they were up by 4 and now it’s 4-3 and the team is under massive pressure, “we’re still winning” is of little condolence to the fans.
Yes?!
Mark Cuban in a recent interview answered your question: companies are afraid there is going to be just one in the end—sort of the way there is one ad-company now on the internet. They want to be that one.
Whether they're correct that there can be only one is of course a matter of debate. But that is at least the mind-set they are operating under according to Cuban.
> sort of the way there is one ad-company now on the internet
Which one, Meta[0]?
0. https://www.reuters.com/business/media-telecom/meta-poised-s...
That kind of does make the point.
Why would Mark Cuban know anything about the motivations of today’s big tech companies? He has not been involved in tech businesses since he sold a radio on the internet website 26 years ago.
Those guys are all on the same private group chats.
"Yet like Musk the ouster wounded his ego". So the journalist believes that reacting to rejection with emotions like a biological person makes him like Musk. Err okay.
I feel like the much simpler explanation is that the US is winning because it's dumping the most money into it. By a very large margin.
Maybe I'm not doing capitalism right, but isn't it supposed to be "The one who profits the most wins"? If you win by just spending, I think you need to adjust the parameters of your capitalistic market.
There is a reason in capitalism we have anti-monopoly law or preventing dumping prices because those often leads to monopoly. So yes for sure you can kill your competition by just dumping money and loosing profits.
I don't get it either, but it seems to me a bit similar to how the US, if you look at market value of car companies, has utterly crushed Europe and Japan (with China surging ahead of those and maybe threatening the US soon), which to me sounds crazy (I still think of German cars as the top of the bunch).
According to Google (AI summary, no idea if it's 100% right but from what I've seen elsewhere it seems right):
Top Car Companies by Market Value (May 2026):
- Tesla ($1.3T - $1.56T): Retains market leadership with a valuation often exceeding the next several largest competitors combined.
- Toyota ($259B - $317B): Largest traditional automaker by market cap and unit sales.
- BYD ($122B - $126B): Strong market position as a Chinese electric vehicle leader.
- Xiaomi ($119B - $135B): High valuation following its entry into the smart EV market.
- General Motors ($69B - $75B): Leading traditional U.S. manufacturer, competing with Hyundai and BMW for top 10 spots.
- Ferrari (\(\approx\$60B-\$68B\)): Maintains high value due to luxury branding.
- BMW / Mercedes-Benz / Volkswagen (\(\approx\$58B-\$64B\) each): German luxury and traditional automakers facing high competition.
- Ford (\(\approx\$47B-\$54B\)): Remains a major player with significant US market share.
So, essentially, Tesla alone is somehow worth more than all European companies combined??!
Except that by sales volumes, the top companies are exactly the ones you'd expect: Volkswagen ($350B) and Toyota ($315B) at the top, far ahead of anyone else... Tesla is around the 7th place with just $95B. Does the financial markets still expect them to far out-earn Volkswagen and Toyota any time soon, we've been waiting for like a decade already??
Gemini says that by country, the car companies revenues are:
* Germany - ~ $600B
* Japan - ~ $520B
* USA - ~ $470B
* China - ~ $250B
How does that even make any sense?
The stock market is over 60% passive investment, it's starting to get unmoored from the financial realities of the underlying companies. What that means for the future is [shrug emoji].
The price changes are not being driven by the passive investors though. People did actually decide to put all that money into Tesla.
...printing the most money into it. The circular IOUs amongst the AI and hyperscalars are a form of debt, i.e. money creation. Don't get me wrong, a whole lot of other dollars are going in too, but investing money that doesn't exist is a massive risk always.
Strange. I'm switching from Codex and Claude to Pi with Qwen3.6 27B local and Deepseek V4 Flash which is dirt cheap but powerful.
I don't think so. From a nation state perspective, AI is a munition. Every advanced nation is going to have their own cyber division with their own AI hosted within its borders. Considering how xenophobic and belligerent the US is, nobody is going to want their national cyber defenses hosted in the US.
On a personal level, I simply do not trust the US anymore. I won't host any of my personal data in a US company. I don't want the US govt invading my personal privacy, and their corporations are constantly leaking and selling private data. I consider US to be rapidly approaching complete autocracy (on par with China) so US-hosted AI is a non-starter. And let's not forget local inference keeps getting more efficient, with higher context and TPS in the same amount of RAM. Within a year even small consumer machines will run local models good enough for basic coding, and in 3 years RAM prices will lower and everyone will be able to afford a decent rig.
Finally, open weight models are now good enough for daily work. They may never be as good as SOTA (SOTA will just keep increasing indefinitely), but that doesn't matter; my car may not be as fast as a Porsche but it still gets me to the grocery store and back. So I use non-US hosted model providers which provide open weights, which are both significantly cheaper than Anthropic/OpenAI, and actually allow me to use my subscriptions without a moat.
But yes, Anthropic/OpenAI are absolutely the new Oracle. They will win for US govt and Enterprise contracts. But that's far from the only users of AI.
The US is not xenophobic. That is ridiculous. Any time you say stuff like that, you discredit the things you say that actually make sense. I'm with you on the privacy aspect, but there are multiple dimensions of that which you're ignoring. I'd much prefer taking my chances in the US than in the EU, where they are constantly trying to push companies to weaken privacy.
> I'd much prefer taking my chances in the US than in the EU, where they are constantly trying to push companies to weaken privacy.
That plus stuff like GDPR is poorly enforced (it's had little effect on the ad industry and databrokers). Many EU countries have much more permissive surveillance and censorship laws than the US.
I'd prefer US companies simply because it will be cheaper and have better availability. The article shows energy prices are way cheaper in US vs europe (45% less than Germany and UK), US has far more datacenters and it's far easier to get new ones online than Europe.
Commercialization is not enough. The US is built on financialization.
Cultivating an ecosysyem of strong capital protections, wealth creation through extraction, and tax advantages for AI finance is what we should be looking for. Commercialzation may be a step towards that, but isn't the destination. We have to create a system where those with money can multiply it, not simple add to it.
Just spitballing here but I think the financial system’s already set up for this.
Whatever derivative structures and equity and options need to exist will be easily created.
I don’t think we need any additional motivation or incentives to cultivate this for AI. We need to keep some in the tank to handle the fallout.
As a more personal aside: the US would do well to put up some sensible barriers to outrageous financialization and reduce moral contagion risk. Otherwise all these folks trying to multiply their money end up leaving the bag with the folks that don’t have it in the first place - and then the folks with money end up, uh… well, it won’t end well.
Betting your entire country's future on usury seems like a terrible idea to me, but what do I know.
Don't look now, but if you've got a 401k then your pension is already dependent on perverse financialization.
“A system where those with money can multiply is” sure sounds dystopian to me.
It's (beautifully) dry humor making fun of OP, whose post is rather dystopian already.
One interesting thing that Anthropic did was putting their stack on the various cloud providers, I wonder if they'll put it on GCP and Azure next since they've put it into AWS first at a level we have not seen a major AI provider do to date. Your company can have their own Claude stack just like an ELK stack on your cloud, if they can do this for both Azure and GCP then OpenAI has to really catch up.
In my eyes I would rather use the AI I can run on my own paid infrastructure, so if there's an outage its isolated, or I could potentially have a different region / DC to fallback on.
I'm still surprised that neither Microsoft nor Amazon have made their own models available on their cloud offerings. I guess Microsoft probably does have Phi on there, but it's not front and center, especially with something like Copilot for Devs (seriously Microsoft rebrand that damn thing to be clear what you mean by Copilot!) where they could use the cheaper compute by using something like Phi.
Claude is already on Vertex - https://docs.cloud.google.com/vertex-ai/generative-ai/docs/p...
Is Vertex hosting it themselves on their own hardware? Because GCP, Azure and AWS all host it on their own hardware, but AWS lets you do everything that Claude lets you do if you use their APIs directly, and then some. This is what I want to see on Azure and GCP, and heck, maybe even DigitalOcean, if they ever stop expanding into so many spaces and focus on improving their current infrastructure, before I fully migrate off of them.
> Is Vertex hosting it themselves on their own hardware?
Yes, you can even choose regions, for EU they serve it from Belgium. With all the encrypted at rest stuff and other guarantees that vertex provides.
> Important: Accessing Claude models through Vertex AI meets the FedRAMP High requirements, and operates within the Google Cloud FedRAMP High authorization boundary.
The recent deal with SpaceX AI to use their severely underutilized GPU compute is pretty telling to me. Being able to roll out compute is a hardware problem, rolling out good models needs more than compute, it needs good AI engineers. SpaceX, Amazon et al can do hardware very well. AI engineering, maybe not so much.
Ppl don't understand Commercialization is not incidental to the Western system, it's why we beat out Communism. Commercialization incentivizes ppl to build, bc ownership and control.
The FSF was not an attack on commercialization, it was about giving users more freedom with their own copy.
AI commercialization is why we will always be a few steps ahead in AI.
The Chinese and Russians are free to join us. It's a pickup game.
This is both true and insightful, but the "its us capitalists vs. communists" framing obscures some very important details.
For one, "Communism" is presented as a single monolith, but it's not: it's socialism PLUS despotism. The despotism part is really important! China/Russia/etc. fail because they try and control things top-down, instead of letting the market decide.
However, you can have socialism without despotism! Tons of European countries are far more socialist, but no less democratic than America (many are more democratic).
So yes, America vs. Russia/China and Capitalist vs. Communist are relevant frames ... but don't let them obscure the fact that you can have a successful, democratic country .. without doing what America does (and giving all control to corporations).
>Trump fits this moment well. He is a salesman at core, and Larry Ellison is too. That helps explain why AI infrastructure is an easy political product. Selling AI today is easier than selling Oracle databases in the 1980s.
I feel like the author (and perhaps many here on HN) are on a different planet than almost everyone I interact with.
Same as software in general I guess? Lots of critical software has been developed by Europeans, but it's the US who build hyperscalers with it
And, yet, the US AI companies are not actually making a profit, right? They're selling at a loss and trying to make it up in volume (or lock in some kind of monopoly position, in a currently non-sticky product). We're all currently enjoying investor-subsidized tokens from the big guys, and that pushes out the reckoning for US AI. But, I think they're beginning to think maybe they need to ring the cash register. Copilot dramatically reducing usage limits and what models are available on its plans, Anthropic playing games with what's included in the Pro plan, etc. I think they're starting to feel the bleeding.
Not only is the investment that keeps US AI companies flying high slowing, I suspect in two or three years, we'll all mostly be using open models and the people making money will be the hardware manufacturers. Even the small models will keep getting more capable. I'd guess a model you can run on a high end, but not outrageously overbuilt, developer desktop or laptop (something like 128GB of unified RAM), will be competitive with the current frontier when it's allowed to search the web and do research and write test code. You can't fit as much knowledge in a small model (80GB of weights can't store the world's knowledge), but I don't have the world's knowledge in my head, either, and yet I can figure out most problems with a little googling and experimentation. The reasoning and tool use abilities of smaller models is where the gap is closing, and that's what will make the huge models obsolete for huge classes of problem.
Already, there are many classes of problem that the easily self-hostable Qwen 3.6 27B can solve that required a frontier model a year ago. When the self-hosted options reach Opus 4.5-ish levels of capability, the argument for paying for tokens for most work begins to look a lot less compelling. And, looking forward, 1.58 bit models are coming. Incredible intelligence density, and still a lot of improvements happening.
"Where it matters most": accuracy and repeatability?
Sorry, nobody's winning that AI race.
I've been using the deepseek api (not for coding though) and have been getting great results and it's so cheap it may as well be free. Another reason I'm using it is because I like the license and I also have some hope of running it in my own hardware in the future.
But the thing is... I could be using any of the llms for my use - I'm using a middleware that lets me change providers only with a configuration change.
So it's going to be tough for USA ai companies to charge 5x to 20x (depending on what you're doing).
That's like saying that Louis Vuitton is monetizing shoes the best. Sure, but it's not winning shoes
It’s certainly too early to call (if you must view this as some sort of adversarial competition). The US is behind on local models, the future for anyone who cares about privacy. There may be step change innovation yet to come that completely shifts the landscape. There’s basically no switching costs to users to change models. They have no lock-in.
It is very much adversarial, and to view it anything but adversarial is to not see the geopolitical reality and the potential national security implications of AI for what they actually are. Moreover, to claim that China is in the lead with local models presupposes that openai and anthropic could not release local models that are better, which is a big assumption. They do not release such models because they have frontier-grade propietary models that have high value.
As someone who happens to have been born in the US and currently lives here, I welcome China winning. I trust them infinitely more than I trust my own government and industry.
OpenAI and Anthropic are beholden to the capitalist system they exist under and hence cannot compete on local models. Like you say, they must try to maximize shareholder value. China is unencumbered by that constraint.
What is it about China that makes you trust them more?
They're far less imperialistic and I view them as better global citizens than the US. I think they've cultivated a much richer culture than the US as well.
You should read about the difference between a land empire and a sea empire.
Happy to if you have any pointers. My original point is that the US kills millions of people outside of its borders, something China most definitely does not. The number is over 12 million post-WWII: https://www.worldfuturefund.org/Reports/Imperialism/usmurder...
Well if you discount the people China has killed and is killing within its borders, the number is definitely smaller. As well as another land empire called Russia.
not living there, for one. I don't care if they know where I live since realistically they can't do much of anything to me. If I were in china, I probably wouldn't trust them as much as I trust the US. If I were in Switzerland, I wouldn't trust the swiss government and might get my services from america or china.
You are free to hate capitalism (even if you benefit from it enormously). You are free to say that you hate capitalism and the U.S. as openly and as often as you like, without facing imprisonment or worse.
But if you were in China, could you say you hate the Chinese Communist Party and China openly and as often as you like without imprisonment or worse?
We know the answer to that. So go ahead and trust China more than the U.S., but I think that is pure foolishness.
Actually, Chinas free speech, while abysmal, is better than you're making it out to be. They only really care if they see you as a threat, which realistically isn't too far off from the US both currently and historically
I understand that there is some freedom of speech in China, but I don't think the implication is that US and China are functionally equivalent. In the U.S. you can criticize the president all day every day and have millions of followers. Could you do that in China? The more people listen to you, the more dissident opinions become a threat.
I'm not suggesting they are the same, rather, that if the US saw you as a real threat you'd be shut up pretty quickly. The threshold for what is considered a big enough threat is different, but they both have the ability to shut you up and have in the past. Any government would act the same, at least a competent one. The governments 0th priority is ensuring its continued existence, since all of the other functions rely on its existence
I live in a Zionist country (the US) and will absolutely be canceled and blacklisted from my industry (tech) were I to publicly speak out against Zionism. They are attempting to put laws in place to make it illegal to be anti-Zionist. These laws already exist in countries like the UK and Germany. Some, such as anti-BDS laws, already exist in the US as well.
Many technological advances weren't driven by capitalism, early computers and the internet were literally developed by the government.
I think you're missing their point, even if I can agree with part of your premise.
There was an outdated but relevant saying
'In America, you can criticize president Nixon anytime'
'Yes, but in Soviet Union you can also criticize Nixon anytime.'
The point is not that they're safer but that they're not a relevant concern in the same way. (According to OP)
I'm glad we went to space, truly. Racing the USSR might have been the wrong reason but it got us there. We've benefited immensely as a species from exploring the solar system and looking deep into the universe.
I'm not certain that racing China in AI is the right reason but it might get us... somewhere.
“It is not the same as profitable AI leadership”
Where are these profits of which you speak?
Hahaha, but no.
It's like people forget the entire point, perhaps even definition of technology is "doing more with less."
The "brute force" of power and cycles is almost certainly the least important thing, perhaps even a hinderance.
Alibaba's cloud is something that the author of this article seems to dismiss. It's being used massively in Asia and they're pretty close in services and offerings to what AWS, GCP and Azure provides. Once they start doing inference on their own custom chips it might be hard to compete with them due to the energy costs
Same can be said of healthcare.
… and it’s substantially due to foreign born researchers and engineers. US will win as long as smart and driven people will want to move over.
Matters the most to whom? I certainly will not care about expensive models that do about the same thing cheap non-american models do.
It's like the USA Librem 5 vs PinePhone. About the same HW for $1600 vs $150.
Sure will not pay 10x for "US" thing just because it's a US thing.
The whole "race" narrative is silly. It is all built on the assumption that one country (corporation, actually) somehow creates AGI and thus, essentially, the singularity. Great for raising VC, apparently, but at its core this is magical thinking.
Even if any of the US corporations would eventually end up in a scenario where their revenue is at least as high as their inference cost, what harm would that do to the other contenders? It's not as if there is any kind of network effect here that would exlude them from market participation.
Well we're certainly losing in terms of public sentiment. There's a real anti-technology mindset that has taken hold. Basically the equivalent of people protesting the electrification of the country early last century.
Chinese culture is quick to embrace the benefits.
I feel like the title of this post should have "for now" appended.
Why though? That could be true of any economic condition. Imagine if any time there was a race or competition in which one group is winning you had to just say "for now".
Michael Phelps is winning the race! ... for now
China is winning the EV race ... for now
It doesn't seem to add value to me, aside from being an opportunity to, as is the time-honored tradition of the haters, to sow doubt and create negative energy to anything related to American success.
+1
Of course US has a huge head start, but if AI keeps growing, what matters is how the market's gonna look like years from now.
Most of my clients using AI in the business workflows (in products) use Chinese LLMs, because after benchmarking for a specific use case you nearly always end up finding that you pay half or a tenth.
That's not a new phenomenon. I've adapted Gemini Flash 2.5 years and years ago when people were dissing it as "crap", yet it was the best budget and quality fit for the task I had at hand back then (translating and summarizing tons of documents). It was both faster and around 100 times cheaper than the best GPT 4 model available.
Needless to say, medium-sized Chinese models are far better than those LLMs and a perfect fit for countless applications.
Even with all the Qwens and Kimis and GLMs etc, the latest Gemini Flash models are still an insane value! I recently settled on 3.1 Flash Lite after testing basically everything on offer on OpenRouter, and it was not a close call - cheaper, faster, and better (for translation and visual understanding tasks).
It’s interesting to see short term business interests again undermining American sovereignty.
Just as business exported strategically critical manufacturing to China, now it is helping funding China’s race to take over the US in AI and beyond.
Lesson is pure free trade doesn’t work if (a) not everyone is playing by the same rules and (b) the trading territories are or may become opposed.
American economic policy gave the world an authoritarian super power and Trump. Not a great track record.
Back in the dot-com bubble, people started inventing new metrics to "value" dot-com companies that lost money hand over fist. My favorite was "revenue multiples". So instead of a a P/E ratio, it was just a multiple of revenue no matter how much money you lost.
We've invented a new term here too: revenue backlog. OpenAI and Anthropic in particular need to recover probably at least $2 trillion to recoup their capex investments. Now Claude code has had an impact on software engineering but for a lot of AI uses you're just not going to recover $2T on $20/month subscriptions. It reminds me of Twitter trying to dig itself out of a $44B hole and losing half their ad revenue with $8/month blue ticks.
The only commercial product AI sells is labor displacement and the resulting wage suppression. You lay off 10-20% of your staff and nobody is asking for raises. The people left are happyt o still have jobs (and thus a house). They'll work even harder doing unpaid labor of the displaced workers to keep those jobs. That's what OpenAI and Anthropic are selling.
The problem is that if these companies get their way, 10-20% of the population is going to be out-of-work and society is going to fall apart. Data centers are going to be the targets of increased societal desperation and anger as this gets worse.
There was a report this week that roughly 50% of singles in the US aren't dating because they can't afford to [1]. This goes well beyond the well-understood problems of not being able to afford a house let alone start a family. This is a birth rate death spiral in the making.
So, back to OpenAI and Anthropic, the only way they justify their valuations and can make up the "revenue backlog" is if they have a moat. And I don't think that's going to happen. Hardware will get cheaper. Nobody is talking about how the generation of AI hardware will write of trillions in investments for some reason. I don't know why.
But the dark horse here is China. DeepSeek when it was first released (early last year?) was a shot across the bow. We have it and toher models (eg Qwen) that will close the gap with whatever OpenAI and Anthropic produce such that no company will "own" AI in the way that OpenAI and Anthropic need to. In the coming years, China's chipmaking is rapidly closing the EUV gap and Western companies have zero penetration into this market. China doesn't want to be dependent on foreign tech that can be withheld at any moment.
Don't believe me? Just listen to the NVidia CEO say the exact same thing [2][3]. Huang realizes this is such a problem that he's gone on Air Force One to this week's Trump summit in China to try and convince the Chinese to buy NVidia chips.
[1]: https://parade.com/living/nearly-50-of-single-americans-not-...
[2]: https://www.youtube.com/watch?v=aJmHfmrRMUE
[3]: https://www.youtube.com/watch?v=Hrbq66XqtCo
This sounds like an ad for US than anything else.
Does any of the US companies earn money on LLMs? No, they bleed money. Github Copilot is switching to token based pricing, which will be costlier than hiring juniors.
Anthropic also is switching enterprises to token based pricing from their subscription one.
From the big three only Codex is still in somekind of subscription pricing, but they'll shift eventually (usage limits are a kind of that, but they have them less stricter than Claude ones)
There is one winner in this race - China. Trump with his agendas and wars makes it even more likely that China will lead this new market.
> Does any of the US companies earn money on LLMs?
Inference? Yes.
Infrastructure build and training? Not yet.
> The US is winning the AI race where it matters most: commercialization.
puke
Yeah, go ahead and run your country into the ground because of hypercapitalism and hypercommercialization, you're almost at the end game now! While the rest of us try to figure out how to actually build societies worthwhile to live in and experience, with healthcare and not waging war on our neighbors.
I don't know how people can seriously publish stuff like this and not feel like they're actively trying to make the world worse. Is money really the single thing y'all can focus on? Is there nothing better in life you can chase, even if it's also a number? So sad to see stuff like this.
GLM, Kimi, MiMo, Minimax, Deepseek, Qwen would like to have a word. :)
This.
I dont know what the benchmarks are supposed to represent, but to me Kimi K2.6 is indistinguishable from e.g Opus 4.6.
How much would one have to invest in hardware to feasibly run a usefully large Kimi K2.6?
There's no set-in-stone answer to these kinds of questions, it all depends what "useful" performance means to you.
Around $35-45k if willing to use it comfortably (fast to answer, large context) as coding agent.
> Trump fits this moment well. He is a salesman at core, and Larry Ellison is too.
Larry just fired 30% of his people at Oracle because, apparently, he is in an immediate need for cash. Because Oracle's early AI bets aren't paying off.
How about the obesity and fall of democracy race?
This is patently absurd. US AI companies are investing non-existent money on huge infrastructure with negligible income. This cannot be sustained. And if/when it fails it will take down the economy of the US and probably any other country touching us business.
Almost as if concentrating all of the wealth and decision-making into the hands of a few billionaires is a bad idea...
They kept telling us Soviet central planning was inefficient and ineffective, while the lean rationality of the free market something something...
What does the term "AI race" even mean, beyond wooing clueless VCs and soon retail investors ? It's not like the LLMs are some super-secret technology. Any economy willing to sink in copious amounts of money and resources can get it to some level - the question, what's the actual payoff? We have yet to see anything really useful, on the level of step change, besides Johnny who can now spin up demo projects quicker.
Not if they cannot get the GPUs.
If China can't get the GPU's they'll build them. If they can't build them they'll smuggle them, and either way, eventually they will take over Taiwan, and then the US will be wondering what they can cook with leftovers.
Opinions are easy.
No, it’s not.
It’s all about adoption and the bigger picture. The US is an untrustworthy, isolated island in the AI future if you vote another idiot into office in a few years. If you’ll still be able to vote at all, that is.
The largest part of the world is not the US. The cutting-edge US models are way too expensive for most parts of the world, and that also shows in adoption.
China is building an ecosystem of open-source models that are both cheap and good enough for most use cases. While most of the US AI sphere will collapse under the pressure of making profits, which means having their models and infrastructure adopted by as many enterprises and individuals across the world, China’s models will have become global standards and hard to displace.
If Beijing’s AI pitch centers on universal access and cost-effectiveness, then Chinese AI firms do not need the latest chips to win the global AI race. They also don’t need the expensive US-run infrastructure. If you watch Chinese AI adoption closely, they already want as many Chinese people as possible to be able to build and try with AI, whereas for most Americans, US models for productive use are already too expensive.
Kimi K2.6 sits within touching distance of Opus 4.7 and GPT-5.5 while costing about $4 per mil output tokens. That is six to eight times cheaper than cutting-edge US models. If you run hundreds of agents, that’s a significant opportunity to get the same work done for a lot less.
Even early adopters like Singapore, ditching US models, the government kicked Zuckerberg in the nuts and went for Qwen instead to build its sovereign AI models.
To understand why the US is at a severe disadvantage in this race, you need to understand China’s Belt and Road Initiative (BRI). BRI entails Chinese firms delivering fully financed infrastructure projects in a bid to lock third countries into China’s economic orbit. They use the same approach for their open source ai models, but this time the infrastructure is both invisible and free.
No need to build power plants or buy /build ports. AI dependency is invisible to both policymakers and the population, limiting pushback. No pesky activists in Germany nagging about China buying parts of ports. No African nutbags questioning why the humble Xi is building hospitals in areas Chinese mining companies take things out of the ground for pennies on the dollar.
China is going for a marathon here while the US tries to push their ai tech by sheer force into the throats of the world. As soon as Chinese ai models have become global standards, it’s game over for us ai companies. And China is way better at this game than the US. They have proven this over and over again in the past 50 years.
I recommend reading the China Standards 2035 strategy to get a better understanding of their approach and how smart this is.
https://www.china-briefing.com/news/china-standards-2035-str...
AI is not as revolutionary as you think in terms of our experiences with previous technological advances in terms of trade and economics.
Western economies are locked into U.S. models, while China runs on Chinese ones. It’s the age-old game. But the real war of the AI race will be fought in the global south.
I will give you three examples.
Can you really imagine, if you look at what AI needs to cost to make a profit, that even at the current prices, US models and infrastructure, which are heavily subsidised already, being used in cost-sensitive countries? I am not talking about coders, think really big here for a second.
Secondly, US ai models are trained on Western data. How do you expect them to grasp local contexts in the Southern Hemisphere? Chinese open-source models, on the other hand, can be downloaded and finetuned with country-specific data.
Want an example? Check out AfriqueQwen-14B, which is adapted to the top twenty African languages.
So I think this author is wrong. The ai race to be won is not hardware or cloud infrastructure, my money is on it will be a contest to decide which models and standards become the default infrastructure in countries that are up for grabs.
China neither needs the best models nor does it need the best cloud infrastructure, it just, like so often, only needs to be affordable and good enough to become the default choice in emerging markets.
The right choice would be for everyone to step off the gas pedal and think about whether we are willing to become China in order to beat China. Our ancestors worked really hard to get us here, our rights, our ways of life, culture, all the blood, sweat, and tears.
AI better be worth it in the long run for all of humanity if we go back to survival of the fittest. Because that is what it will take to beat China at their game.
> While most of the US AI sphere will collapse under the pressure of making profits
I think deep down, sama knows this and that's why he's pushing for "Universal Basic Compute", which really means forcing every US citizen to become an OpenAI subscriber.
That's nothing new, we had BASIC computers back in the 1980s.
>If you’ll still be able to vote at all, that is.
Stopped reading here. What a ridiculous statement and I can only assume the rest of your post is just as ridiculous.
Why do you... seem so sure that this is a ridiculous statement?
The word "winning" implies there is an upside.
so much winning https://layoffs.fyi/
"Methinks the lady doth protest too much"
> Winning the AI Race
Which one of them all?
If you mean "building models that are very good at coding and as substitutes for search engines", then yeah, sure.
But if you mean: "applying AI to industrial applications and robotics", then China is far ahead: https://time.com/7382151/china-dominates-the-physical-ai-rac...
Hopefully it's a race worth winning and not a race to global disaster.
Even if it leads to global disaster, you'd rather be the country with the best AI. You won't be saved just because your AI is worse.
> Even if it leads to global disaster, you'd rather be the country with the best AI
Depends no? If the "Best AI" means "The AI decides when you wake up, go to work, and go to bed", then I probably want to live in the country with the worst AI or even without.
If it instead means "UBI and healthcare for everyone, money lost all meaning and we're all just having fun while AI does all the boring stuff" then yes. But since capitalism still exists, that's a pipe-dream, and "Best AI" won't lead to that for the average person, only for the 0.1%.
This is a false dilemma created by the model companies to try to convince us we must invest or else.
Define winning. They integrate AI everywhere. I hate it. No money shall come from me into AI anywhere. Not sure if I can maintain it, but right now I can.
I mean you can argue the same about Telsa, but look at BYD now.
Just because you are first to do x, doesn't mean you are going to be the winner.
Yes, and it's doing so primarily because of immigrant nerds, H1B's and F1 bros who chose America and may not have this avenue in the future. Potentially, making this the last race USA wins.
It's a myth. IBM, Xerox, HP, DEC was innovating long before H1B's.
There is nothing in the H1B program that makes these immigrants different from the immigrants that ran IBM, Xerox and HP. Other than the country of origin of course.
Jobs, Wozniak, Gates - it's a myth that you need poor migrants pulling themselves by their bootstraps to innovate. Sometimes a nazi scientist like Wernher von Braun is what it takes.
The whole country? Really?
uh, what? I'm pretty sure the AI race matters most is improving society. That absolutely does not equate to making money off of things.
The US has destroyed the PC market perhaps irrevocably, and made getting small single-board-computers for new products extraordinarily difficult. It's enraged its own populus with skyrocketing energy prices causing wild consumer shock. It's waged a trade war over chips that has lead China to develop their own, which they have done astoundingly quickly with phenomenal success, far far faster than anyone could have guessed.
As with another recent example, sometimes in war there is no winning: just loss. This is obviously for us programmers an incredible and wild age, filled with nothing short of miracles. It's incredible. But the prices we are paying, the extreme tensions we are creating, the stress and strain of this all has been incredibly unpleasant, and very very very few people feel like they are seeing upsides to this worrisome menacing age, that promises very few people on the planet anything better coming, and which. Has already made life considerably worse, which no nation has yet directed towards helping its people.
The trade war and tariffs are bringing inflation, consumer prices will soar, but from a geoeconomical standpoint this will hurt China (And EU) more than US. US consumer on average has the deepest pockets in the World and people the top will make money on insider trading, stock anyways. If AI tokens will become like US dollar, it will be under total control of the Fed.
>The US has destroyed the PC market perhaps irrevocably, and made getting small >single-board-computers for new products extraordinarily difficult. It's enraged >its own populus with skyrocketing energy prices causing wild consumer shock.
Strikes me as the real outcome: the end of "personal" computing, "local" anything.
[dead]
As an American, we may be winning this race but we are still struggling to define why this is the race to win.
The cost of winning this race has been telling our citizen s we will replace them with robots and there is no hope for their children’s future employment.
The cost has been destroying trust as we tell citizens water and power should go to server farms and not them.
The cost has been naked power telling democracy it’s wrong and dying
I think when we discover the limits of LLM tech and tally its benefits over its cost — we may regret this win.
But don’t let me contradict a bunch of fake techno oligarchs wrapping themselves in war like patriotism to get the investments they need to keep this going.
The goal is global domination as always, unfortunately. DARPA and the Pentagon helped create the Internet, and Silicon Valley later turned it into a major commercial success.
Likewise, the goal is to not get globally dominated by China. You cannot point to one without the other. It's not like the U.S. is the only country that values national security and geopolitical power.
What does it mean "dominated"?
How would your life change if your country became the second wealthiest instead of the first?
This is a ranking and competition no other country in the world gives ... about.
What would the world be like if China led? Whatever Xi wanted it to be like. I do not want a non-democratic, dictatorial power leading the world. You might think that most of the world does not care, and maybe they don't right now. But this is, I suspect, due to the relative benevolence of historical U.S. policy of international law and order. New boss same as the old boss is a motto that is only as informative as the extent to which the new boss acts like the old boss.
Why would the world care? Take Trump's threats against Greenland...actions that run completely contrary to our historical policies and treatment of our Western allies. They were alarmed, because when the leader of the most powerful military in the world makes a threat, you have to treat it seriously. Despite Trump's hubris, such an invasion did not occur because Americans, Congress, made it very clear that Trump would be impeached if he invaded our NATO ally.
Let's say China is ascendant. It is now the dominant military and economic power in the world. China is under Xi's complete dictatorial rule. Xi decides that invading Greenland is a good idea. Stopping that internally would not involve democratic processes, it would need to involve a coup.
Let's make it even more stark. If Germany had won WWII and had become the ascendant world power, would it make a difference to most countries? ABSOLUTELY YES. If there is going to be a dominant world power, the character of that nation matters.
Are there other nations with the character and institutions that could do as well or better than the U.S. has done? Sure. I can think of several nations. But let us not pretend that all nations are equally bad/good for the world.
[flagged]
[dead]
[dead]
It's a race straight to the bottom. Anyone who gets excited about being a more efficient and productive corporate wage slave that gets to train their future AI replacement is either a shill or not very intelligent.
Flagged for AI content: I hope this submission dies and the user is penalized (look at their submission and comment history!), because IMO the article does not belong on the front page. Quick polemic:
>The US is winning the AI race where it matters most: commercialization
If you ask me, one could name different criteria for winning, and commercialization would not be the first thing to come to my mind:
https://english.www.gov.cn/news/202604/15/content_WS69df29e6...
https://fortune.com/2026/05/03/chinese-court-layoffs-workers...
https://www.reuters.com/world/china/china-moves-regulate-dig...
> It also owns platforms that generate and organize the data of the AI age. YouTube is a video corpus. Google Drive and Microsoft 365 sit inside daily office work. GitHub sits inside software development.
Yeah, okay. China does not have any platforms nor data.
Can I block a user to avoid seeing his posts? I noticed front page would be much nicer without guys like OP
Very good call. I shied away from calling it a terrible article but it is
There certainly are better articles on this topic that have come out recently.
There will be no winners in this once the jobs start to disappear. There is no 1:1 with new human-in-the-loop jobs. Its basically in the definition. We go from being the loop to be "in the loop" .. huge difference.
Ever since I was a small child it was my sincerest wish to build so much value for others by becoming a human-in-the-loop. I am so happy my wish has finally come true.