DeepSeek Accused of Using Banned US Chips for New AI Model Despite Export Ban

The allegations discussed in this article are based on reported claims from Reuters and have not been independently verified.

A Chinese artificial intelligence company called DeepSeek is in hot water with the Trump administration. US officials say the company broke American laws by using super-advanced computer chips that were banned from going to China. And here's the twist—DeepSeek is getting ready to release a brand new AI model any day now.

This isn't just some small tech dispute. It gets to the heart of the high-stakes competition between the United States and China over who will lead the world in artificial intelligence. Think of AI as the new space race, and computer chips are the rocket fuel.

Let me break down what's happening in plain language.

What Exactly Happened?

A senior Trump administration official, who asked not to be named, claims that DeepSeek trained its upcoming AI model—reportedly called V4 and coming out this week—using Nvidia's most powerful chips, known as "Blackwell" processors.

Here's the problem: American law completely bans these Blackwell chips from going to China. Period. No exceptions.

The official put it bluntly: "We're not shipping Blackwells to China."

According to this official, these banned chips are likely sitting in a DeepSeek data center right now in China's Inner Mongolia region. The US government also believes DeepSeek may try to scrub any digital evidence before releasing the model.

The official wouldn't say how the US learned about this — which raises even more questions.

The Chinese foreign ministry spokesperson said on February 24 that she was not aware of the specific situation. The Chinese embassy in Washington added that Beijing opposes what it calls "drawing ideological lines" and "overstretching the concept of national security" when it comes to trade and technology.

How Did the Chips Get There?

This is where things get murky.

While the US official didn't give specific details, reports suggest the chips probably entered China through smuggling networks. Here's how it supposedly works:

Nvidia chips first go to data centers in other countries where they can be inspected. Then, workers take them apart and sneak them into China under false labels, slipping past the US controls meant to keep this technology out of Chinese hands.

To be clear, none of this has been proven publicly yet. But if these reports are accurate, it would be a major embarrassment for American efforts to control who gets this sensitive technology.

The "Distillation" Technique

There's more to the story than just chips.

The US official also claims DeepSeek's new model likely used a technique called "distillation" —basically, borrowing knowledge from leading American AI models to train its own.

The official named several US companies whose models may have been used: Anthropic, Google, OpenAI, and xAI (Elon Musk's AI company). This isn't the first time these companies have made such claims against DeepSeek.

Here's an easy way to think about it: Imagine a young chef learning to cook by studying the recipes of master chefs. The young chef doesn't need to invent everything from scratch—they can learn from the best and build on that knowledge. That's what DeepSeek is accused of doing with American AI models.

OpenAI and Anthropic have previously argued this amounts to stealing their hard work. Based on what industry watchers say, DeepSeek's earlier model, called R1, was also likely developed this way.

What Makes Blackwell Different?

This is where we get into the real substance — and why this chip ban actually matters.

To understand what's at stake, you need to know what makes Blackwell different from the older H200 chips that the US has considered allowing into China.

Not Just Faster — Fundamentally Different

The Blackwell chip represents a major leap forward in design. While the previous generation was already powerful, Blackwell was built from the ground up for the age of massive AI models. It's not that Blackwell does the same things faster — it can do things the older chips simply can't.

Think of it this way: If the older H200 was a professional kitchen with excellent equipment, Blackwell is a fully automated factory. Both can cook, but one operates at a completely different scale.

The Memory Breakthrough

Here's the detail that actually matters for AI training:

Modern AI models need to hold enormous amounts of information in memory while they're training. Blackwell was designed with significantly more memory capacity than anything that came before. This means a company like DeepSeek could train larger, more sophisticated models without constantly shuffling data in and out — a process that slows everything down.

What does this mean practically? If DeepSeek truly trained V4 on Blackwell chips, they could have done in weeks what might have taken months on older hardware.

Why the US Draws the Line Here

The US isn't banning these chips arbitrarily. Export controls target Blackwell specifically because it sits at the intersection of cutting-edge AI capabilities. It's the hardware that enables the kind of breakthrough models that could reshape entire industries.

What This Means for Money and Markets

This is where the abstract geopolitical battle hits home for regular people — investors, tech workers, and anyone watching their 401k.

The Stock Market Factor

When DeepSeek's R1 model dropped in early 2025, it triggered what markets now call the "DeepSeek moment" — Nvidia stock took a serious hit, and the shockwaves rippled through the entire semiconductor industry. Investors panicked at the thought that Chinese companies might have found a way to compete without American chips.

Now V4 is coming, and investors are nervous again. The question on everyone's mind: If DeepSeek keeps advancing this fast, what happens to the companies selling the chips they're supposedly not supposed to have?

The Efficiency Paradox

Here's where it gets interesting — and counterintuitive.

Some market analysts argue that cheaper, more efficient AI models could actually increase demand for chips, not decrease it. There's actually a name for this: Jevons Paradox. It's the observation that when something becomes more efficient, people end up using more of it, not less.

Think about it this way: If AI gets cheaper to run, more companies will adopt it. More adoption means more data centers, more servers, more chips. So Nvidia could actually benefit if DeepSeek's efficiency forces the entire industry to lower costs and expand the market.

A Confusing Picture on US Export Rules

The Trump administration's stance on chip exports has been all over the place, which makes this situation even messier.

In August 2025, President Trump initially said Nvidia could sell a less powerful version of the Blackwell chip in China. Then he changed his mind, saying America's best chips should stay in American hands.

The twist? By December 2025, Trump approved limited exports of the older H200 chip to Chinese companies — but with strings attached. Companies have to pay a fee to the US government, submit to third-party testing, and prove the chips aren't hurting American supply.

This upset the "China hawks" in his own government because they worry even older chips could help China's military.

And here's the kicker: even those approved shipments have been slow to materialize. The requirements are so detailed that companies are struggling to meet them.

What People Are Saying

It's not surprising that this has split Washington into two very different camps.

The "Let's Sell to China" Camp: Some officials and industry leaders argue that selling chips to China might actually be smart. Their thinking? If Chinese companies can buy American chips, they won't work as hard to build their own. Cut off the supply, and you force China to speed up its own chip development — which could backfire in the long run.

The "No Way" Camp: China hawks warn that chips sold for businesses could easily end up helping China's military. They say this DeepSeek situation proves exactly why you can't trust Chinese companies to follow the rules.

Who's Saying What Officially

Nvidia won't comment on the allegations.

The Commerce Department and DeepSeek haven't responded to requests for comment.

At a regular briefing, a Chinese foreign ministry spokesperson said she wasn't aware of the situation and repeated China's standard position on the matter.

What Happens Next

The US official wouldn't say how this news might affect whether DeepSeek can buy H200 chips in the future. So the company's access to American technology remains up in the air.

DeepSeek, based in Hangzhou, made waves early last year when its AI models performed nearly as well as the best American ones. That scared Washington because it suggested China might be catching up faster than expected.

A tech news site called The Information had previously reported that DeepSeek smuggled chips into China. The Reuters report is the first time a US government official has confirmed the chips are likely being used at DeepSeek's Inner Mongolia facility.

With DeepSeek's V4 model expected any day now, everyone will be watching closely — not just to see how smart the new AI is, but for any clues that might prove or disprove the US government's claims.

My Final Thoughts

Here's the uncomfortable truth. This situation raises some really tough questions that don't have easy answers.

For starters, US export controls might not be working as well as policymakers hoped. If Chinese companies can smuggle in the very chips that are supposed to be banned, then the restrictions start to look more like a speed bump than a wall.

What really complicates things is the timing. This news breaks right as officials are deciding whether to allow exports of the older H200 chips. The China hawks now have fresh ammunition to say, "See? We told you so. You can't trust them."

And then there's another issue people aren't talking about enough. While Washington is focused on stopping physical chips at the border, Chinese companies are making rapid progress by borrowing knowledge from American AI models themselves. The US might need to think harder about protecting its software and intellectual property, not just its hardware.

For DeepSeek, the pressure is on. When they release V4 this week, the global tech community will be combing through it for any digital clues about where it came from. The company has to show off an impressive product without handing the US government proof that would trigger sanctions or legal trouble.

Here's the irony: By trying so hard to cut China off from American chips, the US might actually be speeding up Chinese innovation — in smuggling, in building their own chips, and in finding clever workarounds like distillation.

Whether this slows China down or simply forces it to innovate faster is still unclear. But one thing seems clear — this rivalry isn't cooling off anytime soon.