Nvidia H200 China: US Approved It, Beijing Blocked It, Zero Delivered

The Trump administration approved Nvidia H200 AI chip sales to China on May 14, 2026. As of May 21, not a single H200 has been delivered to any Chinese company. Beijing is blocking its own companies from taking delivery. And Jensen Huang, Nvidia's CEO, has said publicly that Nvidia has "largely conceded" China's AI chip market to Huawei.

That sequence — US approval, Chinese rejection, CEO admission — tells the actual story of where US-China AI chip policy stands after the Trump-Xi summit and all the diplomatic optimism that preceded it.

What the US Actually Approved

The Trump administration cleared H200 exports to approximately 10 Chinese companies on May 14, 2026. Direct buyers: Alibaba, Tencent, ByteDance, and JD.com. Authorised distributors: Lenovo and Foxconn.

This reversed a Biden-era restriction placed on H200 chips in January 2025, which had blocked China from accessing the GPU generation above H100. The reversal came after months of Nvidia lobbying and Jensen Huang's direct engagement with the Trump administration, including his presence at the Trump-Xi summit.

The terms attached to the US approval:

• Purchase cap: 75,000 units per approved company
• Domestic volume limit: H200 China shipments cannot exceed 50% of US domestic sales volume
• Third-party inspection: Every China-bound shipment must undergo verification in a US laboratory before export
• Usage restriction: Chips certified for use only within China — not in Chinese companies' overseas operations or data centres
• 25% revenue share: Trump negotiated a 25% fee on every H200 sale flowing to the US Treasury, with all chips transiting through US territory for inspection

On paper: US side approved, terms attached, exports cleared.

Why China Blocked Its Own Companies

The US usage restriction and Beijing's own requirements are mutually exclusive.

The US requires: H200 chips sold to Chinese companies must be used only within China — not exported or used in overseas data centres.

Beijing's position: Chinese companies should only take delivery of chips they can use in their overseas operations. Chinese regulators have flagged supply chain security concerns and specifically objected to the 25% US revenue share arrangement as a sovereignty issue.

The result: a regulatory deadlock. US approval requires China-only use. Chinese approval requires overseas-use capability. Any shipment that satisfies one condition violates the other. No shipment has cleared both simultaneously. Zero H200s delivered.

Trump addressed this directly, saying China is blocking the purchases. Beijing has not publicly confirmed or denied the characterisation but has not moved to authorise any acquisitions.

Jensen Huang Concedes the China Market

The most significant statement came from Jensen Huang on May 21, 2026. Nvidia has "largely conceded" China's AI chip market to Huawei.

This is a CEO admitting, publicly, that the world's most valuable chip company has accepted loss of the world's second-largest economy's AI accelerator market. The context: Chinese tech companies — Tencent, Alibaba, and ByteDance's parent — are accelerating development of domestic chip alternatives. Alibaba's T-Head subsidiary has been building AI accelerators. Huawei's Ascend series is the current domestic alternative to Nvidia GPUs, with significant adoption among Chinese model labs and cloud providers despite performance gaps relative to H100/H200.

Nvidia's China AI chip revenue was approximately $7 billion in fiscal year 2024 before restrictions took effect. The H100 restriction pushed Chinese buyers toward Huawei and domestic alternatives. The H200 approval arriving too late — with conditions China won't accept — is completing that market shift. Companies that built infrastructure around Huawei Ascend over the past 18 months are not going to dismantle it for an H200 shipment that may never arrive under terms that satisfy both governments.

Congressional Opposition

The US approval was not uncontested. Opposition from both parties:

Democrats: Rep. Gregory Meeks warned the H200 clearance "degrades national security." Critics noted the Justice Department had just completed a $160 million bust of an H100/H200 smuggling ring — evidence that export controls were being circumvented even before this relaxation.

Republicans: Matt Pottinger, former Trump administration deputy national security adviser, said "selling H200 to China will supercharge Beijing's military modernisation." The H200 is used in AI training and inference that has direct military applications. The Justice Department itself had noted the H200 is "integral to modern military applications."

The 25% revenue share arrangement — essentially a tariff on each chip sale flowing to the Treasury — is the Trump administration's mechanism for claiming economic benefit from the approval. Critics argue it does not address the national security objection and is primarily a revenue mechanism dressed as a control.

What This Means in Practice

For the near term: no H200s reach Chinese buyers under the current framework. The regulatory deadlock is structural, not a bureaucratic delay. Unless one side changes its conditions, the approval remains a paper permission with no physical shipments.

For Huawei and Chinese domestic chip producers: every month the deadlock persists is a month Chinese AI companies deepen their investment in domestic alternatives. The switching cost of moving back to Nvidia GPUs rises with each infrastructure cycle.

For Nvidia: China's AI chip market is effectively closed for the H200 generation. The H100 was already restricted. The H200 is approved-in-theory but undeliverable. The B200 and Blackwell Ultra — the next generation — will face the same or tighter export control debates.

For US-China AI infrastructure: the gap between Chinese AI capability (running on Huawei Ascend and domestically developed chips) and US AI infrastructure (running on Nvidia H100/H200/B200) will persist and likely widen over the next 12-18 months.

Key Takeaways

• US approval: Trump administration cleared H200 exports to ~10 Chinese companies (Alibaba, Tencent, ByteDance, JD.com, Lenovo, Foxconn) on May 14, 2026; reversed Biden-era H200 restriction
• Terms: 75,000 units per company cap; 50% US domestic volume limit; third-party inspection of every shipment; China-only use restriction; 25% revenue share to US Treasury
• Zero deliveries: Not one H200 has been delivered; China is blocking its own companies from purchasing
• The deadlock: US requires China-only use; Beijing requires overseas-use capability — conditions are mutually exclusive; no shipment can satisfy both
• Jensen Huang admission: Nvidia has "largely conceded" China's AI chip market to Huawei as of May 21, 2026
• Market shift: Chinese companies (Alibaba T-Head, Tencent, ByteDance) accelerating domestic chip development; Huawei Ascend is the incumbent alternative
• Congressional opposition: Democrats and some Republicans opposed the approval on national security grounds; DOJ noted H200 is "integral to military applications"

For the broader US-China semiconductor context, read TSMC: $1.5 Trillion Chip Market by 2030, AI at 55%. For how Samsung's strike resolution affects the AI memory supply chain that feeds these same GPUs, read Samsung Strike Averted: Deal Reached May 20, HBM Supply Secure.