The Chip That Changes Everything
Here's what I keep coming back to: the United States spent years building a wall around its most advanced AI chips, convinced that keeping Nvidia's silicon out of Chinese hands was the only way to preserve America's technological edge. Then, in one announcement on December 8, 2025, President Trump tore a hole straight through it.
Nvidia can now sell its H200 processors — the second-most powerful AI chips on the planet — to approved customers in China, with a 25% surcharge attached. And I have to be honest: I'm still trying to decide whether this is the most pragmatic move the administration has made or a generational blunder dressed up in tariff math.
Let me walk you through why this is so difficult to call.
Why the Old Strategy Wasn't Working
The Biden administration's export controls on AI hardware were never as airtight as they seemed. On paper, they denied China access to the H100 and H200, forcing Nvidia to create a crippled variant called the H20 that complied with the letter of the law. In practice, China's AI sector barely flinched.
DeepSeek, Alibaba, and a growing list of Chinese labs kept producing world-class models. Huawei made rapid progress on homegrown alternatives. The restrictions didn't stop China's AI momentum — they just redirected it, fueling an indigenous chip industry that's now far more dangerous to American dominance than Nvidia imports ever would have been.
Here's the uncomfortable truth the old policy crowd doesn't want to admit: export controls turned Nvidia's competitors into heroes and turned the company itself into a geopolitical football.
What the Biden Policy Actually Achieved
- China built its own AI chip ecosystem (Huawei's Ascend line, among others)
- DeepSeek proved you can train frontier models on less advanced hardware
- American chip companies lost billions in potential China revenue
- The H20 — designed specifically for China — was itself banned by Beijing
The 25% Question
The surcharge angle is the part I find most interesting. On the surface, it's a clever compromise: let the chips flow, but tax them heavily enough that the US still extracts value. The administration frames it as the best of both worlds — Nvidia gets its market, the Treasury gets its cut, and American technology remains the global gold standard because China is still buying it.
But here's where my skepticism kicks in. A 25% surcharge on a product that costs tens of thousands of dollars per unit is real money, but it's not a deterrent. If Nvidia's H200 gives a Chinese AI lab a six-month lead on training their next model, they'll pay 25% extra without hesitating. The surcharge becomes a tax on desperation, and desperation in AI is running at all-time highs.
Some reports suggest individual shipments could be capped at 75,000 units. That's a volume limit designed to prevent bulk transfers, but it raises a bigger question: if you can cap shipments, why not just maintain the ban? The answer, I suspect, is that the White House looked at the landscape and decided that if American chips weren't going to power China's AI future, someone else's would.
Who Wins and Who Loses
- Nvidia: Massive win. The China market was a black hole on their balance sheet. This unlocks billions.
- Huawei: Headwind. Their Ascend chips just became less attractive overnight.
- Chinese AI labs: Mixed. Easier access to top hardware, but at a premium price with political strings attached.
- American allies: Confused. The signal to Taiwan, Japan, and South Korea just got muddy.
My Take: The Long Game Nobody Is Talking About
I think there's a deeper logic here that doesn't fit neatly into either the "America First" or the "free trade" narrative. The export control regime wasn't just failing — it was accelerating the one outcome it was designed to prevent: a decoupled, China-led AI supply chain. By letting Nvidia back in, the US buys something more valuable than a quarter bump on a GPU sale. It buys continued dependence.
As long as China's best AI models are trained on Nvidia hardware, Nvidia's software stack (CUDA, cuDNN, TensorRT) remains the de facto standard. And once a Chinese lab is locked into CUDA, swapping to domestic hardware becomes a multi-year migration project. The export controls were trying to starve the Chinese AI ecosystem. This new approach? It's trying to own it.
Which brings me to the three things I'll be watching:
- How China responds. They could impose their own restrictions on Nvidia chip imports as leverage.
- How Huawei reacts. This is an existential threat to their AI chip ambitions. Expect a major push on software ecosystem independence.
- Whether the surcharge escalates. If the US sees this as working, 25% becomes 35%, then 50%. The question is where the tipping point is.
I don't have a clean verdict on whether this decision was smart or suicidal. Maybe it's both. Maybe it's neither. What I do know is that the era of pretending you can keep AI hardware in a box is over. The genie is out, the market is global, and the US just chose to compete on it rather than contain it. Whether that's a strategic masterstroke or a slow-motion giveaway of our biggest technological advantage is something the next decade will decide.
For now, I'll be watching the shipments — and the models trained on them — with both fascination and a knot in my stomach.
Comments