> It’s bandits all the way down, so adding a little smuggling to that doesn’t surprise me.

Implying it’s *morally* wrong for a Chinese company to bypass US sanctions is hilarious. You really say that with a straight face when even the president admits this is only protectionism?

There’s no honor among thieves. You don’t get to cry about Chinese “bandits” when Anthropic just had to pay $1 billion to settle a massive copyright infringement lawsuit. All of these models were created through the mass-scale theft of humanity’s intellectual property, personal data, and dignity.

Open always beats closed. Drain the moats. Starve the ClosedAI beast.

It's not even stealing. They paid OpenAI for the tokens. It violates the OpenAI TOS, which specifically forbids using it's outputs for training competing models (which is very ironic)

Everyone trains on queries from other models, it's called distillation

> It wouldn’t surprise me if they continued acquiring them through, well, less than legal means.

Strictly speaking, it's not illegal for them to acquire it, it's illegal for an exporter in the US to sell (even if transitively) to them.

I think the fact that DeepSeek trains on competitor queries (i.e., distillation) — along with using banned Nvidia chips — helps explain how it can achieve such low training costs (USD 6 million vs. billions) while delivering only slightly worse performance than its American counterparts. It also undermines the narrative that DeepSeek or China is posing a serious challenge to the U.S. lead in AI. The gap may be closing, but the initial reactions now seem knee-jerk.

That the discussion has being hijacked and shifted to moral superiority is really unfortunate, because that was never the point in the first place.

training on data isn’t stealing the data, in the same way learning from a textbook doesn’t mean youre stealing from it

| Kinda surprised they didn’t run into model collapse problems

Not sure why you would expect this, all the models started doing this as its much more cost effective to get data for post training don't you remember the first grok release where many times it started replies "as a model trained by openai..."

> I also read somewhere (not Wikipedia) that they trained on ChatGPT, Claude, and Gemini queries, basically feeding in the output of competitor’s LLMs as training data

All the labs permitting synthetic data do that.

They don't need to break any laws for this. Where do you think are the customers for data centers in the Middle East? Chinese companies do everything legally, paying for access to data centers that got the chips directly from the US.

Lets take the weapons embargos placed on Israel by our allies. In the NDAA must pass bill we set funds aside to procure those weapons and sell them to Israel. We don't really care about these things we have selective enforcement.

> less than legal means.

This is an absurd concept when it comes to international trade. Even intellectual property is mostly meaningless outside a state. Of course people will evade sanctions; what is the us going to do, invade singapore or malaysia?

> Kinda surprised they didn’t run into model collapse problems,

This is just model distillation.

Anyone with the expertise to build a model from scratch (which DeepSeek certainly can) can do this in a careful manner.

> but they stole their training data from other people who stole their training data from data collections that arguably stole them from content creators.

Bingo.

I have no problem with pirates pirating other pirates.

Screw OpenAI and Anthropic closed source models built from public data. The law should be that weights trained from non-owned sources should be public domain, or that any copyright holder can sue them and demand model takedown.

Google and Meta are probably the only two AI companies that have a right to license massive amounts of training data from social media and user file uploads given that their ToSes grant them these rights. But even Meta is pirating stuff.

Even if OpenAI and Anthropic continue pirating training data and keeping the results closed, China's open source strategy will win out in the end. It erodes the crust of value that is carefully guarded by the American giants. Everyone else will be integrating open models and hacking them apart, splicing them in new ways.

"The TBD group is using several third-party models as part of the training process for Avocado, distilling from rival models including Google’s Gemma, OpenAI’s gpt-oss and Qwen, a model from the Chinese tech giant Alibaba Group Holding Ltd., the people said."

LOL. Distillation doesn't count as plagiarism, or you should call Meta out on it. They're distilling the Chinese model.

Ref: https://www.moneycontrol.com/news/business/inside-meta-s-piv...

Lol @ quoting 3 companies that broke any possible copyright law as victims.

i find it weird that this comment got so much pushback. i don't think it was portraying deepseek as any more morally wrong than anyone else, or castigating anyone as morally wrong.

but maybe i gave it a gracious reading.

Singapore is where it happens.

I was just selling my RTX 4090 on Ebay recently and got a ton of bids from Chinese accounts. The winner ($2,325) had Australia set as the country on their profile, but a Chinese name on the account, and the order shipping address was to a different Chinese name (to a regular single-family house in Delaware). Most bidders straight up had China as their profile country.

So my 4090 (24 GB) is probably going to get turned into a 48/96 GB VRAM frankenstein in a Chinese chop shop. I haven't watched the full 3.5 hour documentary you linked but from the first few minutes, it seems quite interesting. And covers this exact thing.

Edit: Again, I checked the address, it was a house, not a freight forwarder warehouse. And if it was actually going to AU, the forwarder would be on the west coast in CA/WA, not east coast (had another order go to Thailand with a forwarder in SF. And Miami is the big hub for South America). For legit freight forwarding they also wouldn't have different names on the account & shipping address. As the parent comment's YT video describes, these are often just normal Chinese-Americans or international students who do this to make a bit of extra money.

Yeah, or just jump unto Alibaba and Ebay from any neighboring country to China and see for yourself how easy it would be to buy a GPU then transport yourself ~500m and now be within China with these GPUs.

One of the keys being, of course, that the Chinese government doesn't care. Yeah it might require mules bringing the GPUs into China but once they're in China, no one is breaking any laws. Of course DeepSeek is using these GPUs! It's not illegal for them to do so!

Where does nvidia manufacture those chips?

This. And it should be obvious. Drugs are banned and illegal in almost every country, yet they reach the US in vast amounts. Why would a ban on GPUs suddenly work - especially since owning a truckload of GPU is perfectly legal in most countries. Smuggling them to where the demand is, is probably easier than smuggling drugs.

This is tangential but the whole Tiananmen Square thing is kind of odd. When I visited China many people were more willing to discuss it than I had imagined. Some spoke about it unsolicited. It’s a tourist destination you have to buy tickets for. It’s rather subtle what can and cannot be discussed relating to it. Those I spoke to about it told me that most people have a good understanding of what happened, and many people speak negatively of the CCP. You just can’t do it if you have a major platform (e.g. you’re Jack Ma or you are an LLM).

Not to discount how negative free speech restrictions are, but I’m not so sure how effective that particular propaganda campaign would be.

I agree, except with the word "slightly". It can be so significant that this increased cost/friction is the very mechanism of the sanctions' effectiveness. Is it possible to police the Russian oil shadow fleet to extinction? Maybe, but even without doing so you can impose a decent haircut on their profits by issuing scary-sounding press releases and leaving it at that.

Imagine the correct answer being "what incident?"

I believe it is very much a two-way ban, depending on the specific chips [0].

[0]: https://www.aljazeera.com/economy/2025/9/17/china-bans-tech-...

> Whoa there, Bloomberg; just because the USA bans the sale of something to your country doesn't make it banned in your country.

Many Americans, including their government, seem to think that US laws apply globally.

They have extradited Ukrainian men from Poland because that Ukrainian was running a torrent website (illegal in the US, not illegal in Ukraine nor Poland).

They tried getting an Australian extradited from Sweden and the UK for supposedly hosting a website that contained information the US government considered illegally obtained.

More egregiously, they have kidnapped tens to hundreds of people from various countries, sometimes on reasons as flimsy as watch model or name, to torture (sometimes to death), because a lawyer working for the president decided that's actually legal because they're waves hands "enemy combattants".

Don’t hate the player, hate the game.

China banned them tho.

"In fact, the world would be a better place if the US would not use their influence to try to keep other countries down."

acting like china wouldn't doing the same thing to other country if they ever weld such position

every great power would do the same thing to defend their position, its not unique to the US. only because current incumbent power is we see things this way

I agree that a fair playing field for everyone would be the ideal state.

But let's not pretend China doesn't use their influence to keep other countries down as well, and let's not pretend they allow a fair playing field for foreign competitors domestically either.

The US would not have imposed these targeted sanctions if China simply wanted to fairly compete in the marketplace.

It’s debatable whether it’s a better use of US power and resources to try to stop PRC from obtaining these chips versus, say, sinking the Chinese fishing fleets actively wrecking entire ecosystems. I probably agree with you that on balance working on the later problem has a higher long term ROI.

China has promised to wage war and forcibly subjugate Taiwan, a democratic ally and critical trade partner. If China backed off Taiwan for a few decades, I think the US would drop export controls.

Yes, it’s easy, cheap and it works. Maybe they need a super big volume and all the GPUs in the same space, to guarantee minimum latency.

OpenAi, Anthropic, etc have access to specialized network topologies from high profile relationships with cloud providers. I don't think you can get anything like that off the shelf.

If the CIA shows up, Amazon is 100% going to let them rootkit those boxes. They might even have a way to do it without rebooting your servers.

sure but to do this at scale you need better control of the networking layer than they typically give you, and there will definitely be KYC at a certain point in the process

I don't think DeepSeek could do that in the volume they need

I mean, I thought the paranoia was about China making their own models with their own chips. If they're renting USA models via the cloud, that's "good".

You can’t really spend $500k on AWS without doing some form of identity disclosure.

You should really seek to actually understand an issue before you comment so arrogantly.

US authorities are ok with Chinese companies accessing GPUs in overseas DCs because those DCs will still be subject any US export controls. Right now, we don’t really care if Chinese companies are building tier-2 LLMs on US gear. If China invades Taiwan or frontier models approach AGI, we will shut down those Malaysian and Thai data centers overnight.

> And I'm almost certain nothing will happen

What realistically could happen? Nvidia is already prohibited from selling their GPUs to China, I guess if you wanted it to really stop, you'd need to prohibit Nvidia from selling GPUs in any other country but the US, and require some sort of government controlled license to be able to buy it inside the US. Neither of which sound like realistic options.

So what could anyone really do, to "solve" this "problem"?

> And I'm almost certain nothing will happen.

The Chinese government has done more for less so I wouldn't be so certain.

As someone who doesn't want to live in a country run by corporations, that first part of indication the direction of travel sounds like a good idea. Obviously, the companies should follow the direction set out by the nation willingly and not because of threats, but otherwise it doesn't sound like a dumb idea to try to unify people toward some goal rather than the goal being "individuals can get very, very rich and wealthy".

Afaik, data center grade blackwell chips have never been legal for export to china. I think this has more do to with NVIDIA than DeepSeek. For a brief moment, people thought DeepSeek had found some way to produce AI without sending boatloads of cash to NVIDIA, causing a drop in share price.

Shortly thereafter people realized they were probably just evading sanctions and ~stealing~ bootstrapping parameters from other models to reach their stated training cost. This report is just further reporting on that rumor.

> Long term consequences: China outperforms Nvidia, by producing cheaper, faster chips at a large scale, by getting inspired by the IP but using their own production lines.

Unlike your typical free market fanboy, the Chinese leadership isn't stupid. They were always planning to do that, sanctions or no.

Realistically, all sanctions can do is mess with their timelines for some temporary strategic advantage, slowing some things down and forcing reallocation of investment away from other areas into the sanctioned areas.

The US refraining from sanctions is likely the stupid move, because that lever of control will expire at some point. To not use it is to squander it.

But if there's one thing the US government and its business elite is good at, it's squandering things.

I don't know much about GPUs, but is there really any value in IP? I can learn to write HDL code all day long, but turning it into real transistors is the hard part. Code is worth nothing nowadays with AI.

I thought it was impossible for them to leapfrog without actively occupying the TSMC fab as it takes years and years just to dial-in the insane precision.

Bob is a citizen of X, the site formerly known as twitter where their are no laws

and their proprietary tea strains

and forcing them to allow opium to be sold in their country

and forcing them to give up major port cities and open up trade against their wishes

Honestly whenever China gets around to getting its served-extremely-cold revenge for all the savagery committed against it in the 19th and 20th centuries, some chips are going to be the least of everyone's problems.

the american century of humiliation begins

Where can you get 48/96GB 3090s? I know there are 24GB customized RTX 2080s (there were a few on ebay even last I checked) but I've never even heard of the 3090 being upgraded before.

The wild part is that it sounds sci-fi, but the reality is probably way less glamorous

Illegal video gamers playing 3d shooters in virtual world using contraband gpus

Young travelers have always found ways to subsidize their trips by selling some of their personal possessions while there. I heard Levi’s were so popular during the Cold War that customs started counting your clothes going in and out of Russia, which still wouldn’t stop someone swapping a worn out pair for new ones.

And American cigarettes were popular in Canada in the 90’s and 00’s. IIRC the Turkish tobacco blended in was banned by the Canadian state dept so theirs were garbage. Meanwhile Americans were flying into Canada in order to fly to Cuba.

I thought China also wanted them banned to dogfood domestic chips?

it's the standard USian view that their laws apply worldwide

Which they did! https://techcrunch.com/2025/09/17/china-tells-its-tech-compa...

I’d be curious to hear how that could be achieved at a technical level.

Presumably everything youre describing could be averted by simply air gapping the hardware? Or tightly controlling how data gets into and out of the system where the chips are used?

I’d bet good money. The chips have been decapped and roughly analyzed, the only information going into or out of them is from the host systems. Those are easy to convince they are anywhere. Chips don’t have wifi or gps or radios of any kind inside of them, and what the nvidia drivers send and receive to the world is easy to audit.

You give the tech too much credit.

It's really annoying to me that China is in this superposition between a Han ethnostate and a self-proclaimed multicultural multiethnic society, and it always happens to be the more convenient one for any side to make their point

Cool plot for a novel, I’d read that!

Yes. China banned NVidia. Jensen Huang said NVidia is the first one in the history banned by both sides. So Deepseek might find a way to get around Chinese government ban if the claim is true

Whilst there aren't many papers on the matter, I would guess that pretraining from scratch is a bit of a waste of money when you could simply expand the depth/width of the 'old' model and retrain only the 'new' bit.

Even if they're not doing full-from-scratch training every cycle, any serious model updates still soak up GPU hours

I mean, it makes sense in the USA's head. They want to win the AI race. What they are doing is making it so China has no choice but to develop an alternative to Nvidia chips.

Eventually China will make their own shovels to dig up the gold. I'm not sure what this would do to Nvidia, but I'm sure we will find out.

Not only am I not shocked, I am in no way offended as to who is doing the "stealing"