US-China Chip War 3.0: NVIDIA's Impossible Balancing Act
New export restrictions just dropped, and NVIDIA is caught in the middle. They're losing $15B+ in China revenue while Beijing accelerates domestic chip development. Nobody wins this game.
The Whack-a-Mole That Never Ends
The latest round of US export controls on AI chips to China dropped last week, and the pattern is depressingly familiar: Washington restricts, NVIDIA designs a compliant chip, China buys through intermediaries, Washington restricts again.
We’re now on the third generation of this cycle. The original H100 ban led to the H800 (China-specific). The H800 restrictions led to the H20 (further crippled). Now the H20 is being restricted too, essentially blocking NVIDIA from selling any competitive AI chip to Chinese customers.
NVIDIA’s China revenue, once 25% of total sales, has cratered. The company estimates $15 billion in lost revenue over the past 18 months. Jensen Huang publicly called the restrictions “a disaster” at GTC. He’s not wrong, but he’s also not acknowledging the full picture.
What’s Actually at Stake
This isn’t really about chips. It’s about who controls the infrastructure for the most important technology of the century.
The US position: AI chips are dual-use technology. Chinese military and surveillance applications are a national security threat. Restricting access slows China’s AI progress and maintains US technological superiority.
The China position: The US is weaponizing technology supply chains to contain China’s economic development. Export controls prove the need for complete semiconductor self-sufficiency.
The NVIDIA position: We just want to sell chips to everyone and make money. Please stop making this political.
All three positions have merit. None of them are fully honest.
China’s Response Is Working (Partially)
The restrictions were supposed to cripple China’s AI capabilities. Here’s what actually happened:
Huawei’s Ascend 910C is now the default AI training chip in China. It’s not as fast as NVIDIA’s best, but it’s 70-80% as capable and improving rapidly. China’s top AI labs (Baidu, Alibaba, ByteDance) have all migrated significant workloads to domestic chips.
SMIC’s 7nm process is producing chips in volume, despite being theoretically impossible without ASML’s EUV machines. The yields aren’t great, but they’re good enough.
Stockpiling worked: Before the latest restrictions, Chinese companies acquired an estimated 100,000+ H100s through various channels. That’s enough compute for several years of training runs.
The net effect: China’s AI development is slowed by 12-18 months, not stopped. And the restrictions have accelerated domestic chip investment by over $50 billion.
The Unintended Consequences
Export controls are having effects that nobody in Washington planned for:
- NVIDIA loses, TSMC loses too: Every dollar of lost China revenue is a dollar that doesn’t flow to R&D and manufacturing expansion
- The rest of the world gets nervous: Middle Eastern and Southeast Asian countries are now worried about buying US tech that might get restricted later
- China’s chip ecosystem gets stronger: Necessity is the mother of invention. The restrictions are the best thing that ever happened to Huawei’s chip division
- The AI development gap between US allies narrows: If China can build 80% of NVIDIA’s capability domestically, so can Europe, Japan, and India
What to Watch
- NVIDIA’s Q2 earnings and China revenue guidance — how bad is the damage?
- Huawei’s Ascend 920 (expected late 2026) — will it close the gap further?
- Whether the next US administration maintains, tightens, or loosens restrictions
- The ASML situation — the Netherlands is under pressure from both Washington and Beijing
The chip war has no winning endgame. The US can slow China’s AI progress but can’t stop it. China can build domestic alternatives but can’t match cutting-edge performance. NVIDIA loses revenue regardless. The only guaranteed outcome is that the global tech ecosystem fragments — and everyone pays the price.