Nvidia and DeepSeek: How Engineers' Alleged 'Complicity' Sidesteps US Sanctions

Remember the shock when China's DeepSeek rolled out its V3 and R1 models, which matched GPT-4o in intelligence but cost tens of times less to train? Everyone wondered how it pulled that off with a severe shortage of top-tier Nvidia chips. The answer may have turned up, and US regulators won't like it. John Moolenaar, chair of the US House Select Committee on China, stated outright that Nvidia did not merely sell "permitted" hardware but actively helped DeepSeek design and optimize its architecture. It is the classic case of business finding a gap in a fence that politicians spent years building.

The situation is ironic. While Washington methodically banned shipments of H100 and B200 chips to China, Nvidia engineers, by all appearances, were working "in the field," helping Chinese colleagues squeeze the maximum out of what they had on hand. In this context, "co-design" means deep software optimization for the CUDA architecture. That matters enormously: you can have a mountain of graphics cards, but without the right software they remain expensive space heaters. DeepSeek showed the world that a Mixture-of-Experts (MoE) architecture and smart load distribution can make up for a shortage of compute. Now it turns out that technical expertise from Silicon Valley may have stood behind this triumph of Chinese engineering.

Why did Nvidia take such a risk? The answer is simple: money and influence. China remains a huge market, and if the company cannot sell its best chips there, it tries to sell its ecosystem. By hooking Chinese developers on CUDA through consulting and joint work, Nvidia secures dominance for decades, even if the hardware itself is weaker than American equivalents. It is a classic "soft power" strategy that this time collided with the hard reality of geopolitics. For Jensen Huang, this is walking a knife's edge: he is trying to hold on to a global monopoly without angering the White House.

The consequences of this scandal could be tectonic. Until now, sanctions covered only physical goods, such as teraflops and memory bandwidth. Now regulators may move to control the "export of brains." We are entering an era in which an American engineer could be forbidden from getting on a call with a Chinese colleague to discuss code. It sounds like a dystopia, but in the logic of a trade war it is the next logical step. DeepSeek has already proved that algorithmic efficiency matters more than transistor count, which means the fight for minds is becoming more important than the fight over TSMC's fabs.

The funniest part of this story is that DeepSeek's success ended up hitting Nvidia itself. When the market realized that building powerful AI does not require buying tens of thousands of the latest GPUs, Nvidia's shares wobbled. In effect, by helping DeepSeek, the company with its own hands created a case that casts doubt on the need for endless purchases of its most expensive hardware. It is the height of irony in high tech: helping a competitor prove that your main product can be used far more economically than you have been telling investors.

Nvidia is now under the microscope. Any confirmation that its technical support went beyond standard instructions will lead to new hearings in Congress. For the industry, it is an important signal: the era of free exchange of AI ideas is closing for good. Every commit on GitHub and every line of documentation may now be treated as a national security issue. We are watching a "digital iron curtain" go up around artificial intelligence, and Nvidia, it seems, has ended up on both sides of it at once.

Bottom line: Washington has realized that banning chips is pointless unless engineers are also banned from talking to each other. Should we expect "consulting sanctions"?