There are no switching costs for users to move to a new model.

> Chinese open source is falling further and further behind American companies

This is simply not true?

Do not have any empirical evidence, but reality is that China's semiconductor capabilities are not at par with Taiwan yet and the US is able to influence Nvidia's sales to China as well as access to other vendors (TSMC) and technologies, giving the West an unfair advantage.

Just like Chinese EVs and Chinese renewables eventually beat the West, I have no doubt that China can probably eventually pull ahead, but I think it is probably accurate to say that China is currently still behind (how far is hard to say) because they have a slight technology handicap imposed by the US.

Your comment is responding to an issue that is different from what GP said. GP was talking about Chinese open source particularly, i.e. their open source models, which AFAIK have consistently been keeping up with (albeit a few steps behind) the closed source OpenAI and Anthropic models.

Hardware capacity is a separate issue entirely.

    > have consistently been keeping up with (albeit a few steps behind) 

I mean, this sentence is self contradictory, no?

    > Hardware capacity is a separate issue entirely.

It seems like hardware capabilities are at the very heart of both training and inference which is why Nvidia, TSMC are hitting record income and capitalization. Feels like divorcing hardware from the equation is discounting a big part of winning this race.

> I mean, this sentence is self contradictory, no?

By benchmarks, the Chinese models are ahead of where the proprietary US models were ... something like 6 or 12 months ago. And all the benchmarks are a bit fuzzy anyway on whether a small gap is trivial or significant. The Chinese aren't having any problems keeping up on model quality. The gap isn't going to lead to any difference that matters unless the US pulls a rabbit out of it's hat.

Plus dollar-for-performance they might be leading in practice, it is hard to compete with self hosted.

You can keep up even if you’re behind. If someone is running a race and you’re constantly two seconds behind their time, you are steps behind but keeping up.

> I mean, this sentence is self contradictory, no?

As others have pointed out, no, not at all. For specifics, see the chart from this link posted by another commenter: https://hai.stanford.edu/news/inside-the-ai-index-12-takeawa... . If anything, Chinese models are closing the gap with US models, not falling behind.

> Feels like divorcing hardware from the equation is discounting a big part of winning this race.

Depends on what race you're talking about. When it comes to "who has the most powerful models", I'd argue it's actually not really that significant - China obviously has the power to train good models.

Probably a better way to phrase it would be keeping “pace“. Yes, they are still behind but by about the same amount as they always have, they aren’t drifting further behind. Like two marathon runners, one a a mile ahead, but both maintaining the same mile time.

China would probably be very confused that you're asserting they're not on par with Taiwan.

In semiconductors.

Otherwise, no one would need to buy from Nvidia or contract with TSMC.

This article might catch you up: https://en.wikipedia.org/wiki/Political_status_of_Taiwan

I'm aware; I'm Taiwanese :)

> There are no switching costs for users to move to a new model.

This depends on how many proprietary APIs are in the way of the model itself.

Or whether the model still works the way you want. For example, a lot of people were pretty unhappy with Claude 4.7 and preferred the way Claude 4.6 worked. If you're relying on a service, then you're stuck with whatever changes the provider decides to make. And the provider is chasing a demographic that's profitable, if you happen to fall out of that demographic then tough luck.

But if you run your own models then you're not subject to anybody's whims anymore. You have full control of how your software works and what it does.

That’s absolutely a reason to just use a web chat interface. Zero cost switching between providers when one isn’t performing. Second, your data are spread between providers which reduces surveillance effectiveness.

You can't really do any serious work with a chat interface though. Also, you're still wasting time figuring out if your providers do what you expect. If you run a local model then you know it's going to work the way you set it up in perpetuity.

This is absolutely false as a recent study from Stanford clearly states https://hai.stanford.edu/news/inside-the-ai-index-12-takeawa...

> This is simply not true?

Yes, this is purely delusional.

What's often understated is how much of an advantage the US has because it speaks the language of global commerce and technology, which for the entire 20th century and the first quarter of the 21st has been English. That's huge. It means teenagers reading man pages are reading fluently.

At some point, though, the balance could tip. It's impossible to say, and it'd be irresponsible to try to predict it, but there isn't any reason English is natively superior, any more than French was 150 years ago, or Latin 600 years ago. But it's a major advantage the US has that isn't acknowledged often enough.

It's an advantage, but I don't see that changing for a very long time:

1. English became the lingua franca right when the world really became globalized. So everyone from Europe to Asia to Africa has wanted to learn English as a second language for decades. So even if American power went away, I still don't see English falling from its perch. I often say it's really hard for Americans to learn another language because if you go to another country hoping to learn that language, so often you'll find many/most people just want to speak to you in English.

2. The only other power I could see surpassing the US in the mid term is China (and that's in no way guaranteed), but the Chinese language (Mandarin), and especially Chinese writing is inherently more difficult for foreigners to learn. I'd also argue the Chinese writing system is inherently more poorly suited to the digital world.

I know it's a common pop science factoid, but there's actually no evidence that language difficulty has much to do with becoming a lingua franca.

Russian is commonly viewed as a difficult language, but it become a regional lingua franca in their sphere of influence. The only reason we aren't speaking Russian is because they lost the cold war.

I do agree that Mandarin speakers might become more open to Pinyin if more foreigners started learning the language. I'd also point out that English and Romance speakers find Mandarin difficult. For Mandarin speakers, is their own spoken language actually difficult for them? They might find English to be a difficult language.

English is one of the most difficult languages to learn, because there's so many irregular sentence/word constructions + irregular pronunciations due to vowel shift + foreign loan words like French/Latin that must be pronounced differently.

Mandarin eliminates all of these problems. The tones and characters are difficult, sure, but questions and answers being grammatically identical along with consistent pinyin is a lifesaver.

Most of that is due to weird spelling, not inherent to the spoken language.

The hardest part about Chinese is "weird spelling" because the written language is a separate language than the spoken one.

If you're using pinyin it's already easier than English.

English is the best language for this because it easily incorporates quirks/foreign loan words. It will always win over more perfect languages because it just absorbs new concepts. It's purpose/existence from the start was to absorb cultures/concepts. It IS the embodiment of joining cultures + move fast/break things over entrenching/codifying.

> English is one of the most difficult languages to learn

I'm curious if this is just an opinion, or if there is data on this. As a native English speaker, I would have thought this to be true. But I have literally never heard this from people who learned English as a second language, and I was always surprised when these folks said they thought English was pretty easy to learn. My guess was that the "loosey goosey" rules of English are a double edged sword. Yes, there are tons of irregular sentence constructions and exceptions to the rules, but at the same time, most native English speakers (at least in the US) don't actually care all that much if you get all the exceptions right, so if you follow the general rules, people will understand you and respond. I mean, tons of native speakers mix up lay/laid/lie/lied (or sometimes throw in a layed for good measure) all the time, and people don't really care.

This is in contrast to other languages where people will give you blank stares (or worse, take offense) if you get one verb conjugation wrong.

> It's an advantage, but I don't see that changing for a very long time:

It’s an interesting question: for how long will it remain important to know multiple languages in the age of LLMs? Of course, it’s better to know foreign language(s) — no doubt about that — but for day-to-day work, unless you’re living abroad, it seems that their practical utility will slowly decrease. And speech-to-speech translation will likely continue to improve as well.

are you including pinyin in your writing system analysis?

I think English is definitely a reason that I took for granted. To add to that from my experience:

- The culture is, I think, the root of the flywheel. The entrepreneurship and competitive intensity is unlike anywhere else I've lived (not an American). It's okay to go bankrupt. It's okay to fail multiple times and burn millions in VC money, in fact it's encouraged! Take a break and raise another round and go again, VCs like second time founders. In my home country having one business go under is the worst thing imaginable.

- The capital markets, even YC (one of the lower tier accelerators by now) gives you 500k for 7%, sometimes pre-revenue. That is an absurd proposition elsewhere

- Surrounding yourself with top talent raises the ceiling for what you think is possible and accelerates your career really fast. It's inspiring for me to be around so many smart and successful people.

What's your definition of "successful people"?

Self actualized, high optionality

Rich, rich

The language of global commerce and technology has not and has never been English

It is money.

Specifically, right now, petro-dollars. For a while before that, it was pounds

The writer is asking how much longer that will continue to be true that it is petro-dollars.

https://en.wikipedia.org/wiki/World_currency

right, it was english, but then it became english

> but there isn't any reason English is natively superior, any more than French was 150 years ago, or Latin 600 years ago.

Actually, there is. English is relatively unique in its ability to incorporate loan words and features of other languages. This is in part due to its history as a merger of 10k French (thus, Latinate) words into an otherwise Germanic language. But it's also due to the unfortunate history of the British empire, followed by American hegemony, which spread English to many other cultures who freely adapted it.

Whether this is enough to justify a continuing status as "the international language" is obviously debatable. But English is different from almost all other human languages, not because it is better, but because it is just ... more

The ability of English to easily incorporate loanwords is because it has lost almost all word flexion from Old English, with very few exceptions, like the plural marker "-s".

Because most grammatical markers are isolated prepositions, there are no problems caused by phonetic mismatches with the words to which they are associated, like it happens in the languages where a borrowed noun must fit into a declension pattern, which can produce phonetically awkward words.

While among the European languages, for English it is indeed the easiest to borrow new words, one can easily construct an artificial language that would be even better than English from this point of view, and which would remedy various problems of English, like the necessity of knowing separately a written form and a spoken form for every word, or the existence of a lot of semantic ambiguities that do not exist in other languages, or various difficulties to express various nuances using the existing modal verbs, or the too verbose methods for expressing certain verbal tenses, moods and voices.

Thus English does not really have any technical advantages. Its moat is the inertia caused by its so widespread use in the present, which will prevent any other language to replace it, regardless of how much simpler and better that language would be.

> a lot of semantic ambiguities that do not exist in other languages,

This is actually another strength of English, not a weakness.

> one can easily construct an artificial language that would be even better than English from this point of view,

The history of artificial languages (Esperanto in particular) is not encouraging in this regard.

> Thus English does not really have any technical advantages.

I wasn't really trying to suggest a technical advantage, but rather a cultural one. English users, as a worldwide bloc, are really incredibly open to loan words, modified grammar, and even whole new vocabulary. All of this happens in other languages too, but the culture of english makes or allows it to happen much faster and much more broadly.

I’m on a motorhome holiday in Norway right now. The younger people I’ve spoken to, from the Netherlands, through Germany and Denmark and into Norway have as good English as me. As with most American-exceptionalism, you ain’t that special. On previous holidays in France, often held up as “never-willingly-speak-English”, we’ve had similar experiences.

Older people here in Northern Europe often seem to speak English quite well, in France less so.

I'm English, my Danish friends have less of an English accent and are considerably more literate than the average of the people I interact with at work over most days.

It isn't a moat, My partners written English surpasses mine and it is her third language.

I, an American, was on a business trip in Sweden then a holiday in Scotland. It was easier to understand the Swedes than the Scots...

Err…becuase we speak Scottish not English.

This.

But this advantage is vanishing. While automated translation is still not good enough for someone fluent in English to tolerate, it's more than good enough already and the progress have been insane over the past few years.

I don't think English speakers are going to have any edge moving forward.

The majority of AI revenue is probably VC money sloshing around in a closed system, e.g. a VC funds some AI company and they pay OpenAI/Claude. These startups also pay for other AI startup products and make it mandatory for their employees to use them. I would venture a guess that 50-80% of the AI revenue would dry up if VCs stopped funding AI startups.

The gains are so obvious that nobody can cite a source proving them

source: revenue, people opening their wallets

Tokenmaxing?

Source: trust me bro

Okay.

Just cite me any sort of study or financial data showing that AI provides long-term financial gains for any company besides small startups

Do you need a study for when a trading firm reports PnL? Likewise when labs report 80x growth?

There are applied AI cos making 100-400M+ in just a few years of incorporation, does that count as financial gain?

Academia is currently 6-12mo behind the frontier of the industry due to secrecy and publication times, so any "long term" study, even for a year, would be out of date on arrival

When we're talking lab revenue, we're taking what companies are spending on AI.

The question is not whether companies are investing in AI, it's whether they're getting anything in return. Or, whether execs are just as anxious and confused about the story being sold as everyone else, taking the ludicrous amount of capital being put behind it as evidence that there's a "there" there, and hopping on the train out of pure FOMO and hedging, whether they're actually getting anything out of it at all.

I think this is reasonable

If we start to see spend go down because projects fail and companies run the ROI calculation and determine it's not worth it, then ill stand corrected and happily admit that

Why don't you list these AI companies, then?

can just ask chatgpt but off the top:

code wrappers - cursor (special case), lovable, replit

part model part applied - perplexity, 11labs, cartesia, suno

applied branches of model labs - codex, claude code, deployment cos & fde teams

ai roll ups - thrive, longlake, some stealth ones

applied - cognition, sierra, fin, harvey, legora, glean

part data part applied - scale

Margins vary, but many of these companies' revenue are already a chunk higher than what was last publicly reported

Wouldnt be surprising to see some of them 2-5x rev in the next few years

I'm working in a large enterprise that is leveraging AI aggressively.

Anecdotally, I'd wager that the modest/incremental but real gains from boring, daily application pale in comparison to the wasted cycles on terrible ideas, disrupted roadmaps due to poor business decision making, and the uncritical injection of insane, LLM generated bullshit into official business documents (fake KPIs for unmeasurable outcomes, references to nonsensical or non-existent process, data-driven decisions backed by hallucinated data. etc.).

I'm deeply skeptical that organizations will see real, lasting gains. I think they'll see some acceleration of copy/paste-adjacent workflows and gains in non-work like generating slide templates, but that's about the limit of it.

As prices rise to meet actual cost, I shudder to think about the idiotic, reactionary ripples it will send through corporate leadership, with everyone scrambling to evade responsibility at the same time and blaming their tech teams for failing to deliver on bullshit/impossible AI initiatives.

TL/DR yeah, I'd also like to see some real numbers.

I'll push back against most of the points in your comment.

    > The demand for AI is currently overwhelming. As in, can't build data centers and GPUs melting overwhelming, companies growing 3x in a month while already at multi-billion revenues.

This isn't a sign of a successful, sustainable business; it's what a bubble looks like. Between the aggressive marketing (including astroturfing!) that LLM companies are engaged in, the perceived stock market advantage companies can gain by shoving LLMs into their offering, and the missile-gap-style approach that many businesses are taking around this, this centre cannot possibly hold.

    > The models get better and better, Chinese open source is falling further and further behind American companies

American companies are, to be fair, flaunting safety and ignoring the wider social impacts of this technology, and both the US federal and state governments seem to be more than willing to go with the flow on that, probably at least partly because of a recognition that the LLM industry is propping up a significant part of the US economy.

    > The productivity gains are, at this point, obvious

They are, emphatically, not. For me and my peers (most of us, individual contributors in software -- and emphatically, those of us working at companies who haven't fully leaned into vibe coding), our jobs have become babysitting claude agents and spending most of our time cleaning up its messes and doing code review. Short-term, sure, this might lead to some productivity gains, but long-term, this is going to lead to mass burnout.

    > The best talent works (or wants to work) in America and get compensated obscene amounts, the most capital flows through America, this is still by far the best place to start a technology business in the world

Unfortunately, the US is in the midst of cracking down on immigration, and the international perception of the country is increasingly that it is an unsafe one.

    > I think American technology was on the decline for the past few years before LLMs, but for the foreseeable future as long as American companies control the talent flywheel I think the new world of tech is going to be much more American than before.

What I see in the US's LLM-backed economy is what I see in many businesses in this same economy, increasingly: the blanket of AI is being used to paper over serious, systemic issues in the organization, but this clearly won't hold. In a world where we have an ounce of responsibility for what we produce, and where customers care about the quality (notably, quality as in correctness) of what's being delivered, this will eventually collapse.

Thank you for your perspective!

I think it's obvious that demand is overwhelming supply right now. I agree that we don't know how much of the demand is due to perception, perverse incentives, or poor management, and how much of the demand is 'real'. I personally believe that the demand is mostly real and will continue to go up, but I don't have a crystal ball.

I also acknowledge that the productivity gains are highly dependent on your specific company's implementation and the work that you're doing. I think the role of a technical IC (which I am as well) is going to be managing fleets of agents, and many people who aren't suited to that type of work will leave the industry (and many people who are will join).

I generally agree with you on the points about American politics, I don't think the way they are cracking down on immigration is very wise.

As for correctness - it's a nontrivial problem to deploy AI in prod that works and doesn't blow up over millions of runs+. Hence why the initial value has accrued to the intelligence layer (labs) but the bulk of the remaining value will accrue to the applied layer in my opinion.

I will buy your entire supply of money for $0.50 per dollar.

Our demand for compute and software is infinite, but our price sensitivity is also high.

So the people who "aren't suited" just get thrown out with the trash. I'm sure that will help with public perception of AI.

The other side of this is that they're replaced with people who aren't qualified to evaluate the work output of the LLMs.

> The demand for AI is currently overwhelming.

Wait until they charge the real pice, if I sold a dollar for 10ct I'd also have a lot of demand.

I'm burning billions of tokens on chatgpt "deepresearch Pro extended" for things I wouldn't even bother googling, the second I have to pay even 2x the price I won't use that anymore

I hear this analogy (selling a dollar for 10ct) but it's unclear to me how we can cleanly map intelligence to cents.

If the LLM was GPT-1, most people wouldn't even use it for free. So clearly there's another axis here?

The analogy is implying that the revenue generated by providers is dwarfed by the total expenditure on inference & continuously training the next best model. These providers have large operational costs and the presumption is that they are providing a dollar ~worth~ of product for 10 cents. Worth being calculable based on the actual capital & operational costs of providing the service.

anthropic models are profitable fully loaded (rev - inference cost - training cost)

According to who? Remember that they told us gpt 3 was "too dangerous to release"

Until their finances are open you can't trust anything they say

Yea I don't know if that's true or not. I'm just saying that is what they are getting at.

The estimates I've seen are that running inference at scale on a Deepseek V3 sized model (so 700B parameters) costs roughly $0.70/mtok or so given current H100 rental costs. Sonnet charges $15/mtok on the API so the delta between the true cost and the API cost is quite large, to the point where even many subscription users are likely profitable.

Can't that be countered by the fact that you can pay a reasonable price (something like 20 or 30 bucks) for small businesses independent flat-rate inference subscriptions of models like GLM-5.1? They aren't being subsidized, they just balance normal and power users around their flat rate. Just check something like synthetic.new, Ollama Cloud or OpenCode Go.

He's not denying that there is demand, he just has a different view on what's happening:

When developers say that LLMs make them more productive, you need to keep in mind that this is what they’re automating: dysfunction, tampering as a design strategy, superstition-driven coding, and software whose quality genuinely doesn’t matter, all in an environment where rigour is completely absent.

They are right. LLMs make work that doesn’t matter easier – it’s all monopolies, subscriptions, VCs, and lock-in anyway – in an industry that doesn’t care, where the only thing that’s measured is some bullshit productivity measure that’s completely disconnected from outcomes.

...

One group thinks this will make the world ten times richer. The other thinks it’ll be a catastrophe.

(from an earlier post, https://www.baldurbjarnason.com/2026/the-two-worlds-of-progr...)

Reasonable conclusion, if you think the entire software industry is rotten then accelerating rot won't do much

I personally disagree with that worldview. (I read the article and the guy's tone is lowkey salty)

The reality is it's insanely hard to convince people (/especially/ consumers. //especially// technical consumers) to pay up to use software. Anyone who has tried to sell software as a startup knows, customers are laser focused on outcomes and value and anything that raises an eyebrow means you're toast

Ofc there are perverse incentives and I think those are bad

> The reality is it's insanely hard to convince people (/especially/ consumers. //especially// technical consumers) to pay up to use software.

The industry is in an extremely bimodal situation, which drives most of that rot.

You have the startups and small businesses who can't get businesses or customers to pay up. And you have the SaaS giants, who already have their customers and can charge whatever they want.

And this is where the "rotten software industry" and doubts about AI feasibility intersect: Both of these business archetypes lack a clear use case for AI.

If you're small, congratulations you can now spend thousands a month on tokens and still have $0 of revenue. AI doesn't really help you "catch up" to customer expectations as now you're also having to compete with the myriad of slop-shops and in-house AI software development.

If you're a giant, well... why bother? Why give OpenAI or Anthropic a million dollars in tokens? They don't need to make the software better nor do need any "AI efficiency" to do layoffs.

I'm curious as to where your perspective comes from.

My view is they both have a clear use case for AI, because every business has a use case for more intelligence on tap. Enterprises big and small already shell out billions upon billions for AI so I'm not sure how your premise holds

In fact AI has resulted in more startups than ever starting to take market share from the incumbent software companies (and the market has started to price that in)

> Enterprises big and small already shell out billions upon billions for AI so I'm not sure how your premise holds

By your logic, shouldn't these enterprise's cash flow be expanding due to AI instead of shrinking?

Every business has a use case for more intelligence on tap, but it is abundantly clear that LLMs are not in any way intelligent. They still frequently make egregious errors in what they do, because they are just token predictors with no intelligence or understanding of what they are doing. Yes, even the state of the art frontier models. This in turn means you have to either baby sit them, or accept a much higher rate of failures than a human would produce. Either option kills any potential productivity gains.

> My view is they both have a clear use case for AI, because every business has a use case for more intelligence on tap.

They all do, but for small companies it won't be a benefit, it will be table stakes. It will also not increase revenue for them, it will reduce it because more competitors will be introduced, and customers won't be able to easily differentiate the true slop from the expert-guided and curated slop. The only alternative will be to become more of a slop shop, i.e. replace expensive programmers with cheaper AI, lowering your quality. Or to shut down.

For big companies who have always had terrible quality that didn't matter at all to their bottom line, of course it's a good investment. They can fire programmers. Do buybacks.

> because every business has a use case for more intelligence on tap.

This is the part I would contest.

Obviously there's some disagreement about how much AI is actually meaningfully intelligent for any given task, but even outside of that:

Turning "intelligence" or staff skill into revenue is not automatic or trivial. You can hire the smartest process engineer on the planet, but if your assembly line has no major inefficiencies, they just can't do anything. They can build you a 2nd assembly line but if there's not the market demand to buy that much product, it's pointless.

For software development: "More code" does not really translate into "more revenue".

If you are a SaaS giant, you don't need "More code". You're not in the business of doing software development. You're in the business of rent seeking. No need to replace people with AI, you can just fire them and replace them with nobody.

And if you're a small firm, code isn't your USP. Everyone has code. Good god especially now with AI, every dumbass' startup can have a trillion lines of code. As has oft been observed (long before AI), the code is a liability. There's not really much efficiency to be had by using AI, because the devs are already minimally writing code.

agents are not used just for code

they're quite generally intelligent, and you can leverage them in nearly every business function. marketing, sales, ops, research etc. which basically every business needs

clearly we still need humans running companies for the intangibles but im saying the premise doesn't hold up under the market reality

I wonder if this is a sign of bad value. Long ago you'd be willing to pay. The relationship was clearer , simpler, stabler. No sudden change of price or rules, no constant false improvement. It was less flexible, and riskier on a way, but it cleaned the noise.

My 2cts

I mean software as a customer/user is by and large completely miserable anymore.

People don't like spending money on yet another Solution that will screw them over then shut down, screwing them over again regarding whatever they were using it for. Look at the reactions and expectations when Beloved Name gets bought by private equity, or how people count down the days until Useful Service starts enshittification and becomes terrible.

People have been trained to leapfrog from service to service while watching for signs of rot because nobody makes software or has long-term plans that actually care about the customer's long-term well being anymore. If Founder McSynergy can make an extra dollar by throwing the entire userbase into a volcano, well, that's just a sacrifice he's willing to make.

> Anyone who has tried to sell software as a startup knows, customers are laser focused on outcomes and value

So the solution is to reduce the cost to zero, instead of competing to provide the best outcome and highest value?

If you've ever tried to start your own company in the US it's a grueling, insane warzone of competition

That results in the winners providing insane value to both customers and equity holders

It's hard for me to reconcile your post as being authentic. From what I see, current "AI" is simply a geo-political tool, and a tool for governments to maintain power and authority. It is not real AI, since it cannot learn.

Real AI is being suppressed and it seems that it will not be allowed to exist in the mainstream, especially in the US.

This is the ultimate "no true Scotsman" post.

> The models get better and better, Chinese open source is falling further and further behind American companies.

Prior restraint is going to put a damper on American state of the art for the foreseeable future.

https://thezvi.substack.com/p/the-ai-ad-hoc-prior-restraint-...

In the longer term, companies won’t be able to build AI infrastructure fast enough to keep up. The construction capacity isn’t there. The hardware production capacity isn’t there. Raw materials, energy, water—not enough of any of it. The supply chain is a fragile, grotesque joke.

> as long as American companies control the talent flywheel

The companies are eating their seed corn. Senior devs are going to age out and there won’t be enough juniors coming up the ranks to replace them. The oncoming demographic crisis multiplies this problem.

Americans decided to sabotage their own public education system for generations. They were able to bridge the gap with foreign undergrad/grad students for a while but that well has been poisoned, probably for good.

Thank you for sharing the article, it's an interesting perspective and I'm inclined to agree with the point about prior restraint.

I'm sad that America is making it more difficult for foreign talent to come in. But with the flip-flops between D/R in the white house it's really hard to predict what immigration looks like even 5 years from now

I see two possible outcomes:

1. People really voted for getting violent criminals out, in which case there is going to be a massive backlash against the current policies.

2. People are really convinced that immigrants are making their lives worse, in which case as things actually get worse with the lack of immigration, they will probably double down. Politicians can keep using immigrants as a scapegoat, and fascism here we come!

> But with the flip-flops between D/R in the white house it's really hard to predict what immigration looks like even 5 years from now

Flip-flop means unsafe. Flip flop means you are one election away from being mistreated in a lawless way by goverment and bad luck away from being mistreated by their sympatizants.

Unless there are actual prosecutions and those made to stick, there is no flip flop safety.

And given one of parties is increasingly anti-democratic, unless there is flip and actual change of constitution next flip in next 3 years, flip flop will stop at the fascist side.

What are you talking about even. Chinese models are what pretty much every AI company in the US is using now because you can run them on prem and customize them, and because hosted versions cost a fraction of US ones. https://www.youtube.com/watch?v=9baDOfwUzHQ

And that's in the US, the rest of the world is all using Chinese models as well. Which means these models get far more collaboration from the global research community being developed in the open. They will set the standards in terms of how APIs work. And they will be what everyone uses going forward.

The closed approach simply can't compete with that. The same way Linux destroyed Windows on servers, open AI models will destroy proprietary solutions as well.

Indeed! China is leaning heavily into AI as state policy, as the solution to its looming demographic crisis. Any advantage the US has is going to be brief. It'll be like comparing the high speed trains in China with the high speed trains in California...

Can this be backed up with any numbers, especially in the US? Every company I've seen using an AI something has obviously been using the API of one of the bigger companies. If this is a valid approach with proof it's basically as good, it would be something I would recommend to my company

Here's a recent Stanford study showing that Chinese models are basically just as good https://hai.stanford.edu/news/inside-the-ai-index-12-takeawa...

For most use cases, you don't actually need frontier performance either. Customization, cost, and data sovereignty are far bigger practical concerns. If you can run your own model on prem and tune it exactly what you need, then you're both saving money and getting better quality output.

It's also wroth noting that tooling can go a long way to improve the quality of output from the models as well, and this is very much an under explored area right now. For example, ATLAS agentic harness does a clever trick where it gets the model to generate multiple candidates then uses a second lightweight model as a heuristic to score them keeping the promising ones. And this drastically improves coding capability.

https://github.com/itigges22/ATLAS

There's also a paper along similar lines discussing how using a harness to force a project structure also allows it to work on much larger projects successfully.

https://arxiv.org/abs/2509.16198

So, I don't think that raw power of the model is even the most important part at this point. We can squeeze a lot more juice out of smaller models we can run locally by using them more effectively.

We're basically in the mainframe era of this tech, but the pendulum always swings to tech getting more optimized and moving to edge devices over time. And I think we're already starting to see this happen with local models becoming good enough to do real work.

That's interesting but it didn't answer my question in any way. You made a claim about usage in companies.

I assumed you were asking about capabilities since that is a question that can be answered. There isn't any comprehensive reporting by companies, so it's just the anecdotal reports the video I linked discusses.

And there are also occasional statements like the one by Airbnb here disclosing what they use https://www.bloomberg.com/news/articles/2025-10-21/airbnb-ce...

ai generated video script

"Chinese models are what pretty much every AI company in the US is using now" - just untrue. you think people inside Cursor use composer for most of their work? haha

the talent at the labs far surpasses the global research community its just not comparable

I'm not saying I prefer it this way, I want open source to do well but it's just not happening at the current pace

Who cares how the script was generated. What he says is entirely factual. He cites plenty of concrete examples too.

The idea that the talent in the US surpasses the global research community is laughable. China already tops the world in artificial intelligence publications. https://www.science.org/content/article/china-tops-world-art...

China also has a population of 1.4 billion people, and an excellent education system. Pretty much all top universities are Chinese. https://www.nature.com/nature-index/institution-outputs/gene...

And let's not forget that top AI researchers from US are now fleeing to China. https://www.scmp.com/news/china/science/article/3353398/lead...

Publications != talent anymore. The top talent work at labs that keeps most of their research secret. And Microsoft AI is not in that circle

Not denying that China is a close #2 btw.

Sure, publications might not equal talent, but the fact is that China leads research in 90% of crucial technologies. So, clearly China leads in a very tangible way here https://www.nature.com/articles/d41586-025-04048-7

And specifically to AI, practically all major innovation that's been published and is used in the wild comes from Chinese companies. Before DeepSeek, everybody just assumed you needed a gigantic date centre to train models. Qwen is showing that you can get near frontier quality on your desktop. Nothing of the sort is coming out from the US.

And frankly when you look at the recent report from Stanford, it's embarrassing af for the US. Look at the chart on how much money is going into AI in US relative to China, and then at the chart showing how there's practically no difference in quality of the models. The only thing the US is ahead in is burning through capital like there's no tomorrow.

https://hai.stanford.edu/news/inside-the-ai-index-12-takeawa...

>companies growing 3x

which companies are growing, the ones mining for gold or the ones selling the shovels?

>American companies control the talent flywheel

But we already know US doesn't, the AI competition is largely Chinese talent vs Chinese talent that the Chinese gov allows to work in west, which they control plurality of global AI talent pipeline, and can cut off at any time, like the reverse has already happened for western semi talent in PRC. Leverage applies to many other sectors.

Simple law of large numbers, i.e. generating comparable STEM than RoW combined = the best talent going forward is some Chinese... with little English fluency. English fluency deprioritized from mandatory a few years ago in PRC, the smartest kids with access to most modern corpus of research in most productive academic system is going to be locked behind mandarin in future.

Western models are not getting better vs massive compute difference predicted during period where compute gap vs PRC is expanding. And better in what which ways? There's entire industrial sectors US models can't get better in va PRC for the simple reasons the industrial chains do not exist in US (or at scale in west as whole). Throwing $$$ at half the problem... is severe misallocation, but the group think in the $$$ group probably feels like everything is peak because muh valuations and fomo investments while digital companies figure out how to integrate AI to write better newsletters, meanwhile some PRC dark factory goes brrrrt. A little hyperbolic, but you get the point.

I think something to be said that PRC can cut off talent pipeline to US AI at anytime, but hasn't... nor losing shit over AGI threat complex. They see absurd amount of $$$ being dumped into western AI and ask themselves, why stop this hyper financialized capital bonfire.

> The demand for AI is currently overwhelming [...] companies growing 3x in a month while

"Yes, but have you considered number go up?"