business 6 min read

Nvidia Ends H200 Sales to China After Huang Steps Down

Nvidia is closing the last major loophole allowing advanced chip sales to China after Jensen Huang's departure. Here's who wins, who loses, and what it means for the global semiconductor order.

  • AI Hardware
  • NVIDIA
  • China Semiconductors
  • Jensen Huang
  • Export Controls

The Last Loophole Closes

Nvidia is no longer selling its H200 chips in China. Acting CEO Colette Kress confirmed the move on Tuesday during the company’s first earnings call since Jensen Huang stepped down last week, ending one of the most contested chapters in the long-running battle over American chip exports to Beijing.

The H200 pathway existed because of a deal Huang personally brokered with the Trump administration. Under the arrangement, chips were shipped through the United States, subject to a 25 percent tariff, but allowed to flow to Chinese buyers. It was a compromise that angered China hawks on both sides of the aisle and drew consistent criticism from lawmakers who wanted a harder line.

Huang defended the strategy publicly and privately. In April, speaking with podcaster Dwarkesh Patel, he argued that conceding the Chinese market to American rivals would be a disservice to U.S. competitiveness. The logic was straightforward: if Nvidia didn’t sell there, someone else would — likely AMD, Intel, or eventually Chinese manufacturers stepping into the gap.

That logic no longer applies inside Nvidia’s own walls.

A Leadership Shift With Immediate Consequences

Huang’s departure appears tied to more than just policy pressure. Reports indicate part of the reasoning behind his exit involved an affair with the CEO of a company under investigation for alleged chip-smuggling activity linked to China. Whether or not that personal dimension directly caused the leadership change, the outcome is the same: a new regime at the top has moved quickly to signal alignment with national security hawks.

Within days of Kress taking the helm, Nvidia announced the creation of a board-level committee tasked with overseeing national security policy, export control compliance, and global trade risks. The committee’s formation alone sends a message — and it’s a message aimed squarely at Washington.

It also explains the speed of the decision to terminate the H200 channel. Under previous leadership, that route had political cover. Under new leadership, maintaining it looks like recklessness.

The Rules Tighten Across the Board

The H200 termination was only the first move. On Monday, new export control rules took effect barring advanced U.S. chips from foreign data centers that serve Chinese customers remotely. That closes what many saw as the next-biggest loophole: the ability for Chinese AI labs to train models on American hardware hosted in Singapore, Malaysia, Thailand, Indonesia, or other third countries.

Nvidia had been navigating that gray area carefully. Now it doesn’t have to.

Under the new framework, any large data center seeking to purchase Nvidia’s latest accelerators must certify that it does not serve Chinese customers. If it can’t provide that certification — and many regional cloud providers can’t easily guarantee it — the company faces a stark choice: lose the customer or risk violating U.S. law.

Dan Fisher-Owens, an export control attorney, told TechCrunch the impact could extend well beyond Nvidia’s direct sales. Facilities unable to certify are likely to shift toward alternative chip options, further eroding American dominance in the region.

The Revenue Question Is Murky

Jefferies forecast the combined effect of these restrictions could trim between $44 billion and $62 billion from Nvidia’s data center revenue next year. That is a staggering number — roughly equivalent to the annual revenue of some Fortune 100 companies.

But Kress pushed back during the call, pointing to a massive backlog of unfulfilled demand for advanced products like the Vera Rubin platform. If existing customers already committed to orders can absorb the shortfall, she argued, top-line revenue may remain largely unaffected in the near term.

Longer-term, the outlook is far less certain. Critics have estimated that up to 20 percent of Nvidia’s revenue traces back to China-linked demand. Annual revenue at the company’s Singapore subsidiary hit $104 billion in 2025 before Nvidia stopped disclosing those figures — a number that almost certainly includes significant Chinese customer spillover.

Revenue protection in the short run does not mean market protection. Customers who cannot access Nvidia hardware today will find alternatives tomorrow. Once those relationships form, they stick.

China Was Already Walking Away

Here is a detail the headline story often misses: China was not eagerly buying the H200 anyway. Even before Nvidia decided to stop shipping them, the Chinese government was largely refusing acceptance of the chips. The preference, according to Jacob Feldgoise at Georgetown’s Center for Security and Emerging Technology, is to generate demand for domestic chipmakers — even when major AI compute users would prefer Nvidia’s still-superior equipment.

So the loss of the H200 channel hurts less than the raw numbers suggest. The real damage is forward-looking. Chinese labs have already demonstrated the ability to train frontier models on prohibited hardware — evidence of Grace Blackwell chip usage appears across leading Chinese AI systems. But those training runs likely took place at overseas facilities. The new rules now target exactly that scenario.

If Chinese developers want to keep accessing world-class Nvidia silicon, they will need to do so through channels that the United States is systematically closing.

Who Wins and Who Loses

Washington wins. The China hawks who demanded action get exactly what they asked for — a cleaner wall around advanced computing power. The board-level committee gives the company a permanent institutional mechanism to ensure future leadership cannot reopen the kind of loopholes Huang cultivated.

Nvidia wins in the short term by protecting itself legally and politically. But it loses strategically. Every Chinese lab forced off the Nvidia ecosystem becomes a potential customer for AMD,Intel, or China’s own rising chipmakers like Huawei and Cambricon. Once those relationships crystallize, recovering market share will require superior product and aggressive pricing — neither of which is guaranteed.

Chinese AI companies lose access to the best tools available. They have shown ingenuity in circumventing restrictions, but the cost of doing so rises every time a new loophole closes.

Third-country data centers face the sharpest dilemma. Facilities in Singapore, Malaysia, and elsewhere built their business models partly on serving Chinese developers with American hardware. Under the new rules, those models require certification they may not be able to provide without losing their biggest customers.

What Happens Next

The trajectory is clear. The United States is moving toward a comprehensive decoupling framework for advanced semiconductors, and Nvidia is accelerating to stay ahead of it rather than lag behind. The H200 termination and the data center certification requirement are not isolated moves — they are symptoms of a broader architectural shift.

What remains uncertain is how quickly other American chipmakers follow Nvidia’s lead. AMD and Intel face different political dynamics and customer bases. But the direction of policy makes compliance increasingly mandatory rather than optional.

For the global AI ecosystem, the implication is starker. A bifurcated hardware landscape is emerging: one cluster built around American design and manufacturing, another growing around Chinese alternatives. The two sides will diverge in capability, tooling, and standards over time. The question is not whether that split happens but how fast.

Nvidia’s new leadership has chosen a side. The rest of the industry will have to choose too.