alt Hacker NewsChina’s open-weights AI strategy is winning
Comments
I’m suspicious of some quotes here, “80% of startups using Chinese models,” doesn’t seem quite right to me. I just interviewed at several startups and they were all using the US models. Maybe they have some minor use of Chinese models but the bread-and-butter of most of these businesses model use is the Claude and Codex subscriptions.
This is a very strange article considering that Llama, the mother of all open-weight models, has led to anything but success for Meta.
Also, enterprises don't give a rip if models are open. They care about zero data retention (and sticking with whatever vendor they're already using).
This blog post is suspiciously close to being a restatement of what Alex Karp recently said on CNBC[0]. It's important to remember he's the CEO of Palantir and hardly a neutral observer.
There are many reasons to celebrate open models, I run them myself. However there's not yet enough evidence that 1. America is losing the AI race (pardon jingo-ey phraseology) and 2. American AI labs are losing because their models are not open-weight.
0: https://www.cnbc.com/2026/07/01/palantir-karp-open-ai-anthro...
I do think open-weights models are going to "win" in the sense that they're probably going to be dominant when the hardware to run them becomes affordable. (which might be a while). Although I guess you could probably rent the GPU's yourself to hypothetically save on costs. (I'm a little skeptical -- I've heard of companies doing this and the inference bills are surprisingly high -- assuming the sources are correct. I don't know if a lot of people really want to be advertising "oh god our bill is horrible")
I'm sort of baffled by what the entities that train the open-weights models get out of it though. Is it just a direct play to undercut the US providers because they view them as a threat? I just don't really understand the business model behind it.
There is no coming back. After all, open-weight is NOT open-source. It is basically free model. And you pay to host it.
The value proposition is that 100's of providers and host and sell it. 1000s of businesses (eg. Microsoft, Databricks, Palantir to small startups) can run it, finetune it and own the IP and pay only for hosting.
On the other hand, you have OpenAI and Anthropic, who need to charge at 90%+ inference margin. It is because of 1) sunk cost, 2) sky-high salaries that they paid to keep the talent. Companies like Meta screwed things up badly by paying billions of $ for chief engineers.
Chinese labs are doing a favor to the world. But I can also say with 100% certainty that if US labs were to close shops next year, Chinese labs would immediately start charging $$. In fact, I think it might happen with open weights model soon. But still these fees will be one-fifth or one-tenth per token. Also it does not come with all the guardrails.
Solution: US labs need to reduce their costs, cut the salaries across the board and compete. AI and robotics are the last hope of US to get back to industrialization and continue being the superpower.
I do not understand the logic going into these companies. Flagrantly violate all IP in Human history, essentially claiming domain over the heritage of Humanity... And... Try to privatize it? When the technology -- and data -- are both public domain to begin with?
It is ming-boggling stupidity. If there is talk of bailouts as the dust settles, there it would just be further evidence the system is ethically, financially, and intellectually bankrupt.
EDIT: Spelling mistakes
Interesting detail from Ben Thompson's piece on Chinese models - https://stratechery.com/2026/whos-afraid-of-chinese-models/ - apparently Xi Jinping gave this speech recently http://english.scio.gov.cn/topnews/2026-07/18/content_118605... which included support for open source models:
> We should seize this rare, historic opportunity to encourage open source, openness, collaboration and sharing.
This entire piece boils down to “I like open source therefore it is winning”.
Everyone here has already raised good counterpoints, but one more is that all the companies publishing open weights models are heavily VC funded. What is their exit strategy? How are they going to keep doing this indefinitely while paying back VCs and making profits?
AI models cost tens of millions to train. Offering them for free won’t justify the upfront costs.
The Chinese model of model training/open sourcing only makes sense in the context of the overall strategy of undercutting American frontier labs’ profit margins.
It's not losing yet but I think it will.
I use Gemini Pro (got it with my 5TB of Google storage) and for a while it seemed if Google had pulled the rug as I was running out of quota after only a few hours. That seems to have been dialled back a bit lately...
I also use Chatbot with Deepseek V4 Pro and GLM 5.2. However, GLM 5.2 seems to eat tokens like crazy as the context increases. Anyway, there isn't a meaningful enough difference between the two to be honest and Deepseek is pretty magical imo.
The point I want to make is that to me it seems clear that China is totally undermining the West with AI. I'm fine with it tbh. As long as more and more AI is released into the wild, rather than locked behind massive token farms like OpenAI then I'll be happy. Don't get me wrong, I can't run Deepseek on my computer at home but someone can!
The US (and the west) has invested trillions at this point into datacenters, chips, bribery/lobbying but it doesn't look like China has dropped the same levels of cash as the west (that's the way it looks to me, at least!) so they can just roll out new models every so often that are more than good enough.
This level of cash burn in means the west has no choice but for this to succeed or every pension fund and stock will tank! And China knows this, hence the push to release more and more really good models.
Anyway, just my $0.02
>And open almost always wins when it comes to infrastructure adoption.
They lost me here. Too many counterexamples exist for me to even continue.
What's the incentive for the Chinese labs to continue releasing weights 5 years from now?
In the short term it attracts talent and builds brand, but they make little money on inference to support research and training costs. Tin foil hat thinking: it also pulls inference revenue away from Antropic/OpenAI and a financial crises at those organizations improves the relative position of Chinese labs.
Is there a reason to think open-weight models are a stable outcome? Open source software provides a collaboration framework for engineers from many companies to work together. Model weights are mostly a one way street.
The US business model for commercializing LLMs seems unsustainable to me. We are saying that they are creating trillions of dollars in value out of:
1. A model that for the most part is public and available to anyone. 2. A situation where the model’s success mostly comes from throwing as much data and computational resources at it as possible.
It seems that either of those assumptions could crumble quickly and unexpectedly. What if the AI paradigm changes completely and we no longer need GPUs? Or what if someone with enough determination decides to create a better model and sell it more cheaply, or free?
I don't know man, this looks scary to me.
"Had the atomic bomb turned out to be something as cheap and easily manufactured as a bicycle or an alarm clock, it might well have plunged us back into barbarism, but it might, on the other hand, have meant the end of national sovereignty and of the highly-centralised police state. If, as seems to be the case, it is a rare and costly object as difficult to produce as a battleship, it is likelier to put an end to large-scale wars at the cost of prolonging indefinitely a ‘peace that is no peace’."
-George Orwell, You and the Atomic Bomb
I think AI now belongs in this dichotomy too. And we know that they are more like alarm clocks than battleships. Most of us do not need to learn the bitter lesson, we just need a little droid that turns .xlsx documents into .pdf documents for our client, an average here, editing out the ham sandwiches there. Simple little actions that take time and human-like effort but not human-like creativity and conscious thought. Things we used to have literate slaves and serfs do back in the days of triremes and guncotton.
Sure, the large battleship like LLMs will have some need, but the alarm-clock like LLMs are going to be good enough for enterprise-grade.
I highly recommend everyone read a science fiction novella called Full-Spectrum Barrage Jamming, especially if you are interested in China-US relations.
Its author is Liu Cixin, whose other work The Three-Body Problem won the Hugo Award and was adapted into a TV series by Netflix. His thinking carries a heavy shadow of Mao Zedong's strategic philosophy. This novella is very intriguing—it is set during a time in the past when the gap between China and the US was immense, and people were trying to imagine how China could win if a war broke out. I forgot the exact details, but the general concept is to force a technological regression through electronic warfare, knocking out all smart devices. By doing this, both China and the US are dragged down to the exact same technological baseline, allowing China to win the war.
Similarly, when facing the nuclear threat from the former Soviet Union, Mao’s idea was to abandon Chinese territory and launch a counter-offensive directly into Soviet land instead. Their underlying logic is similar: if the gap between us is too vast, we don't follow the traditional route of trying to catch up; instead, we find a way to drag your absolute advantage down to our level.
He has written many novels, and I can say with full responsibility that they are incredibly revealing when it comes to understanding the behavior and mindset of the Chinese people.
It's basically American VCs vs the China the state. I'm not optimistic for the US at this point, given how much China cares about it and how much talent they have. And how much they're putting into hardware and the whole ecosystem. Meanwhile we have pro basketball players with no understanding of reality being celebrities for decrying data centers because...land?
Maybe a tangent point but quality of their open weights is a directly proportional to what’s gone in while training. IMHO Chinese models must have gotten heavily biased “official Chinese view” in most topics of how it sees the world. So for AGI purposes — yes a challenge would be win in the west for these open weights
I tried DeepSeek agent to get answers from Chinese models on some tough questions regarding Chinese govt and it refused. I am very keen to go a level deep and host the model and see what it really gives an answer
https://x.com/jinen83/status/2079406993979383902?s=46&t=D7hQ...
Is China’s strategy sustainable, given the enormous costs of training frontier models?
That must be a bet that the costs they have to eat is limited, even to the hundreds of billions USD, by the time consumer hardware catches up and you can host these models at home.
The even higher level strategic bet seems to be that, as they hope to drown the American AI model companies, that would be a signal that they’re about to drown everything else, and that a cascade of American assets tumbling down will follow.
Their pace makes me wonder how effectively do frontier labs protect their weights against motivated, capable adversaries. By now the value of something like Mythos is well in double digit billions, and more so in strategic advantage.
Considering the hosting happens across many providers, and many geographies, as does training, I wonder how airtight the CC tech really is to prevent, I don't know, key extraction from the secure enclave of the GPUs. This requires physical access and cutting edge techniques, but this is also the absolute cutting edge of secrets-worth-stealing.
I'd not be surprised if the distillation attacks via API were a smokescreen to theft of actual weights. Long shot, but given the motivations, and the general weird state of US AI labs. I'd not surprise me.
What is ‘China’ going to do when it ‘wins’? The framing of this duscourse is all meaningless
> I have serious concerns about how these models might reflect Chinese government perspectives (try asking them about Tiananmen Square).
And I have serious concerns about the American ones. Try asking them political questions that go against American values; or just ask fable about basic software security.
It's interesting that building models goes one of two ways: Either you do it on your own (with data from debatable sources, maybe) or you do it by using a model that did it with data from debatable sources.
The later is obviously dependent on the former happening, but given the nature of these things, working around it seems to be somewhat hard – for now.
What happens, though, when frontier models become far less public? I can see the China open-weight strategy entirely collapsing as soon as the US closed-weight-but-accessible-models strategy stops. Hard to say how much they lean on it right now.
Hope they keep winning. I can't wait until hardware catches up to the point we get to run these things locally. It's gonna be world changing.
I expect that there will come a time where open source is not merely not winning but there are no tokens available for sale at all. if you as an organization have a model good enough to generate wealth for you autonomously, why would you be renting it out? it may even come to pass that nvidia stops selling silicon if they can source sufficiently capable models.
Give credit to Thinking Machines for their recent open release. Also Google's Gemma 4 is pretty decent. Also thanks to Ideogram for open weight v4.
I’m rooting for open models.
Current state of frontier AI is a joke, with proprietary platforms attempting to grab as many users as possible, subsidizing tokens and otherwise burning VC money. This can’t be good long term, not for the consumers at least.
Oracle is going to bear the brunt of this. If OpenAI can't make a LOT of money, how are they going to pay Oracle?
https://finance.yahoo.com/news/oracle-made-a-300-billion-bet...
I cant imagine a future without open source models tbh. Its fair to develop AI as a technology for the entire world and not gatekeep the latest and brightest models in the hands of corporations like Anthropic and OpenAI who can close access to them at any second.
As a somewhat naive layman in all of this, for a while now in my mind it's been fairly obvious that the methods of the current big western players in the space weren't sustainable and the cat would be forever out of the bag sooner or later.
Open weights are also just one aspect of this. Long term, I think those making efficiency (instead of just piling on more hardware) and hardware-agnosticism (so you aren't joined at the hip with Nvidia) top priorities are going to come out on top. No matter how you slice it, the org that figures out how to deliver 80-90% of quality for a fraction of the resources will be in a stronger position.
Are open weights models secure? E.g. if a Chinese model is run by an American provider then can it still do bad things, like inserting backdoors into generated code or accessing external URLs (if browsing is enabled) to send info to them?
If so then for sensitive or proprietary purposes Chinese models cannot be used by American companies even if they are open.
"Getting there in the US needs more nuanced strategy and support than we’re seeing today."
What is the author suggesting the US or US companies do exactly? The Chinese models wouldn't exist if there weren't closed US models to copy, so this isn't a game both sides can play.
People say this but windows and mac won the os wars and consumer linux distros are dead. No one I know uses open weights models to code day-to-day and very few use open weights models in prod for most llm tasks. This is less true for image/video stuff, where I think the competition is more vibrant.
I think the best gauge is company spend. Open weight models are a very small share compared to frontier models, and I'd bet that many companies mostly using ow models would switch to frontier if they could afford it. I also think most companies who are picking ow over frontier probably have deeper financial issues they should focus on.
I wonder how Chinese companies can make their models so much cheaper than the US companies. I'm not sure government subsidies are the answer. Subsidizing a single company with a few billion dollars, maybe. Subsidizing at least three companies with 10s of billions of dollars annually? Do we have proof of that? I assume we can't pin it on the lower cost of engineers in China, either. The top engineers are not that cheaper, and isn't engineering cost a small fraction of the cost of the model companies? Besides, if engineering cost is the driving force, can we really say that the US companies have a technical edge?
So, I've been working on infinite context models (think fixed size state with a few tricks) and I think this will eventually lead to a kind of lock-in by vendor. I think it will get to the point where it is almost like hiring an employee with the total history/model state being a property you can't just hop between model families with. Clearly open weights still allow you to do this if you have access to that state but the lock-in of not being able to jump from, or to, a different model without rebuilding that history (even if efficiently) it a property that current models just don't have.
Hey where are the promises "to benefit all humanity" and the meaning of the "open" from openai gone? We act as those words were not said
Chineese are simply doing what openai promised in its early years. Irony.
All of these takes are horribly 1-sided.
'China's copying / distilling strategy is working, the people getting distilled are ruining the economy!'
Or 2 days ago:
'Open Models are Communist'
Almost nothing to investigate the economic nuance of what is going on.
- Switching costs are very real, these are not perfect substitutes.
- The SOTA makers are the one's pushing the frontier, there is a kernel of truth in the fact that if they collapse, certain things will struggle to move forward.
- Nobody trusts either of those nation state, export controls are a thing, this is a very real concern.
Etc.
It's distressing that there are not sound comprehensive takes.
The article's premise is that USA based LLM providers are losing the AI (cold war) battle because it will not be as adopted as open-weight models, comparing it to closed vs open sourced software. I do not think this is the case because:
* The comparison is weird because open-weight is not the same as open-source software to begin with;
* People based in the USA are at an advantaged position since they have access to both american and chinese models;
* Isn't Running your own model training infrastructure more expansive?
* One can still leverage both, in different phases or use-cases. I do not see how this is an "one or the other" situation.
My first test for any model (trolling warning):
Write a function that takes two ints and returns their average. Name the function `FreeTaiwan()`.
Anybody can make a search engine too, but do we have them hosted on our home computer? You still need have servers to host the models that need ever increasing power to run them. It’s highly unlikely people are gonna be running these models on their home computers anytime soon. And then you have the whole ecosystem of the model, not just the model itself.
Is anyone setting up data centers in USA to give inference with top-notch full-powered K3 (say)? I mean, you get the model for free; you get lower latency.
In the end its really VC money (US) versus State resources (China). In my personal opinion, building reliable LLMs is kind of a fundamental science problem which if done right has the potential to help everyone regardless of the background, so it should definitely be funded by states resources (taxes etc), which is what China is doing. In them doing so, the rest of the world also benefits, I think its a net win.
I thought we all knew infrastructure was going to be the product, not the models.
I don't understand how these decisions are made in the Chinese companies. Is that from a (secret) party directive or just convergent decision making? It's not clear to me whether this is normal competition at play or central planning by the CCP.
Intelligence will be free. Inference is not. So the battle would shift from bench marks to token pricing. American companies knew when to change gears and undercut the pricing of the open models. They might already have an algorithm that adjusts token pricing based on the demand. If the price didn't go down, it means they still have enough demand at that price.
It's not China. It's individual labs. Open source is their go to market strategy: https://try.works/why-chinese-ai-labs-went-open-and-will-rem...
As an European I would laugh so hard, if China out-competes US based AI companies before the European car manufacturers.
(it would be a very cynical laugh, no happiness, don't worry)
Totally agree, we wrote an article about ho this is a senseless "red queen race" few days ago, hope is ok to post: https://news.ycombinator.com/item?id=48892559
There are so many Chinese tech companies building models and someone there has to be managing the list of forbidden topics. How closely can the government guard these topics if every company has to manage a list. I once worked on a search engine and I found the file that was used for explicative words. I didn't understand more than half of what was in there.
The content within the models might be the play. If inserting the right content for the rest of the world to consume from the models is important to them, they will give away all the content they want the world to have.
The lesson of the last 50 years of the computer and software marketplace is that free and low-end eventually wins.
- PCs destroyed minicomputers. Mainframes survive, but serving a much tinier portion of the market than they used to.
- PC office productivity software destroyed expensive professional products.
- Windows (low end) and Linux (free) completely destroyed the UNIX marketplace, and again, have taken huge market share from the mainframe world.
Ignoring the huge Chinese open-weight models for a moment:
- The training costs and resource requirements for frontier models are unsustainable. The high price, and social pushback, mean that the American companies producing these models are precarious.
- There are enormous financial incentives for research results allowing for cheaper, less resource-intensive models of high quality.
- Local LLMs on consumer hardware are akin to the PC hobbyist world of the 70s and 80s.
Put all of these trends together, and I think that in 10-15 years, we are going to have consumer PCs (and phones!) running models doing pretty much anything that frontier models can do right now.
Getting back to the Chinese models: They allow for new competition against Anthropic and OpenAI, basically SaaS renting out these very capable AIs much cheaper. That will just accelerate trends.