Jensen Huang is back in China, but the leather jacket isn't casting its usual spell. For years, Nvidia treated the Chinese market like a guaranteed ATM, accounting for roughly 25% of its data center revenue. Now, that faucet has run dry. Despite a "peace treaty" of sorts where the U.S. government finally approved exports of the H200 chip, Nvidia's latest earnings report dropped a bombshell: revenue from China-specific AI chips has hit a literal zero.
It’s a bizarre standoff. Washington gave the green light, yet the chips aren't moving. Huang's visit ahead of the Lunar New Year wasn't just a social call; it was a desperate attempt to figure out why the world's most valuable chipmaker is being ghosted by its biggest customers.
The H200 Approval That Nobody Wants
In late 2025, it looked like Nvidia had won. After months of lobbying, the Trump administration authorized shipments of the H200 GPU to China. There was a catch, of course—a 25% "tax" or royalty paid back to the U.S. government. Nvidia was willing to pay it. They even prepared 82,000 units for immediate delivery.
Then, nothing happened.
Nvidia CFO Colette Kress confirmed this week that the company hasn't generated a single dime from these approved sales. The problem isn't the U.S. anymore; it’s Beijing. Chinese regulators have reportedly told tech giants like Alibaba and Tencent to "pause" H200 purchases. They're worried about a double-cross where the U.S. pulls the plug again after the infrastructure is already built around Nvidia’s proprietary CUDA software.
Local Rivals are No Longer Just Jokes
If you haven't looked at Huawei lately, you're missing the real story. For years, Western analysts laughed at Chinese "homegrown" chips as inferior, power-hungry bricks. They aren't laughing now. The Huawei Ascend 910C and the newer 920 are actually beating Nvidia's "neutered" export-compliant chips in raw performance.
- Raw Power: The Ascend 910C offers more than double the floating-point performance of Nvidia’s previous "China-legal" H20 chip.
- Production Scale: Huawei is on track to produce nearly 1.6 million AI dies in 2026.
- The Software Moat: Chinese developers are rewritten their code for Huawei’s CANN architecture because they don't have a choice. Once that switch happens, Nvidia's CUDA advantage evaporates.
Honestly, it’s a classic case of "be careful what you wish for." By cutting China off from the best chips, the U.S. forced Beijing to build a self-sufficient ecosystem. Now that the ecosystem exists, they don't need Jensen Huang as much as they used to.
The Geopolitical Tax That Bites
Even if a Chinese company wants an H200, the math doesn't work. The U.S. government’s 25% cut makes these chips prohibitively expensive. When you add the new 2026 tariffs on high-end electronics, an Nvidia chip in Shanghai costs significantly more than a domestic Huawei or Moore Threads alternative.
It’s not just about the money. Every shipment is now subject to "end-use monitoring." Imagine being a Chinese CEO and having to tell a U.S. inspector exactly what your AI is thinking about every Tuesday. It’s a non-starter for national security.
Why the Rest of the World Still Matters
Don't weep for Nvidia just yet. While the China business is in a tailspin, the rest of the world is on fire. Their Q4 revenue hit $68.1 billion—a 73% jump. Microsoft, Meta, and Amazon are still throwing billions at the Blackwell and upcoming Rubin architectures.
But the "China Air Pocket" is a warning. If global demand for AI infrastructure eventually plateaus, Nvidia won't have the Chinese market to catch its fall. Huang is trying to sell a "sovereign AI" narrative—the idea that every country needs its own AI "factory." It’s a smart pivot, but it doesn't replace the 25% hole in the balance sheet that China used to fill.
What You Should Watch Next
If you're tracking this, stop looking at Nvidia's stock price and start looking at Chinese data center procurement. If Alibaba or ByteDance officially starts deploying the H200 in the next three months, Huang’s mission was a success. If they don't, it means the decoupling of the global AI industry is final.
Nvidia is basically saying "no China revenue" in its next-quarter forecast. That's a huge bet on the rest of the world’s insatiable appetite for AI. If that appetite flickers, Jensen Huang's leather jacket will start feeling a lot heavier.
