The Chip Wall Backfires: US Export Controls Lock Nvidia Out of China

Ten Firms, 75,000 Chips Each, Almost Nothing Shipped

On December 8, 2025, the Trump administration approved the export of Nvidia’s H200 chip to China — a processor that offers over six times more processing power than the previously approved H20, according to the Council on Foreign Relations. By May 2026, the Commerce Department had cleared roughly ten Chinese firms — including Alibaba, Tencent, ByteDance, and JD.com — to buy H200 units, with each approved buyer permitted up to 75,000 chips, per reporting from EM360Tech.

On paper, that looked like a reopening of Nvidia’s largest addressable AI market outside the United States. In practice, it wasn’t. On July 14, 2026, a Commerce Department export-controls official told Congress that H200 shipments to China had been “very few” despite the roughly $10 billion in approved licenses, CNBC reported. EM360Tech separately cited Reuters reporting that zero deliveries had been administered under the new licenses as of that report.

This isn’t Nvidia’s first collapse in China. The company earned $12 to $15 billion in 2024 selling its China-compliant H20 chip before the ground shifted again — in April 2025, Commerce declared the H20 “noncompliant” with export rules, and Nvidia took a $5.5 billion write-off on the lost sales, according to Brookings. China still accounts for roughly 13% of Nvidia’s total global sales, but almost none of that now comes from frontier AI compute — the H200 pipeline that regulators approved has stalled well short of commercial reality.

Why Beijing Said No

The stall isn’t a Washington problem — it’s a Beijing one. On July 9, 2026, the South China Morning Post reported that Beijing selectively cleared only Alibaba, ByteDance, and DeepSeek to even consider buying H200 chips — and required each company to justify why domestic alternatives couldn’t meet its needs first. Zhou Chao of the Chinese think tank Anbound told the South China Morning Post the policy’s purpose was clear: “The core consideration behind this approach is most likely to secure a window of opportunity for domestic AI chips to grow.” Shanghai export-control lawyer Shi Shenchang offered a narrower read — that domestic chips are unlikely to close China’s computing-power gap in the near term, which is likely why Beijing left any opening at all.

Washington tightened its own side of the wall in parallel. On June 1, 2026, the Bureau of Industry and Security clarified that export-license requirements for advanced AI chips — including Nvidia’s Blackwell GPUs — apply to any company headquartered in China, regardless of where its subsidiaries sit, Al Jazeera reported. Former State Department official Chris McGuire said Chinese companies had exploited the gap between the Trump administration’s May 2025 repeal of the Biden-era AI Diffusion Framework and this clarification, telling Al Jazeera that “Chinese companies have been buying these chips, very likely at scale.” The result is a policy sandwich: Washington now requires a license before an H200 can leave the country, and Beijing requires its own justification before a Chinese buyer can accept one.

The Huawei Question: Filling the Gap or Filling Headlines

The obvious beneficiary is Huawei, and the headline numbers support that story. Huawei’s AI chip revenue is projected to reach $12 billion in 2026, Tom’s Hardware reported, driven by mass production of the Ascend 950PR chip that began in March 2026.

But the performance case for Huawei closing the gap is weaker than the revenue case suggests. Researchers at DeepSeek — the Chinese lab with every incentive to prefer domestic hardware — found that Huawei’s Ascend 910C performs only about 60% as well as Nvidia’s H100 despite comparable specifications on paper, according to the Council on Foreign Relations. CFR estimates the best US AI chips are currently about five times more powerful than Huawei’s best offerings, a gap it projects could widen to seventeen times by mid-2027 — partly because SMIC, China’s leading foundry, remains stuck at 7-nanometer manufacturing due to export controls on lithography equipment. Even under an aggressive scenario of 800,000 Huawei chips produced in 2025, CFR calculates Huawei would still account for only about 5% of Nvidia’s aggregate computing power.

The longer-run risk for Washington’s strategy is architectural, not just about node size. Brookings points to Huawei’s “LogicFolding” chip-stacking technique, which could potentially match a TSMC 1.4-nanometer-class chip by 2031 without requiring the export-controlled extreme ultraviolet lithography tools that underpin current US leverage. Add in the demand side — Chinese models such as DeepSeek’s run at roughly one-sixth the price per token of comparable offerings from OpenAI, Anthropic, and Google, per Brookings — and Chinese buyers have less urgency to fight over scarce, license-gated Nvidia allocations in the first place. The net effect: export controls have succeeded narrowly, in denying Nvidia’s balance sheet a China AI-compute line, while the wider strategic question — whether China’s aggregate compute capacity is meaningfully constrained — remains genuinely contested between the two camps of analysts.

What Global Tech Buyers and Investors Should Do

1. Stop modeling a China recovery into US AI-silicon revenue

Nvidia’s $12-15 billion 2024 China haul is gone, and the $10 billion in H200 licenses approved since December 2025 has converted to close to zero realized shipments as of the July 2026 congressional testimony. Analysts and enterprise planners building 2026-2027 forecasts should treat approved-license value as a regulatory ceiling, not a revenue projection — the gap between “cleared to sell” and “actually sold” has now held for eight months and shows no sign of closing on its own.

2. Separate “blocked from buying Nvidia” from “blocked from computing”

Huawei’s roughly $12 billion 2026 Ascend revenue forecast and DeepSeek’s low-cost model economics show that Chinese AI development hasn’t stalled just because Nvidia access has — it has simply gotten more state-directed, more price-sensitive, and (per CFR’s performance data) meaningfully slower per chip. Don’t conflate market-access denial with capability denial; track both metrics separately when assessing how far behind Chinese labs actually are.

3. Treat export-license approval as necessary, not sufficient, for delivery

Beijing’s requirement that approved buyers first justify why Huawei’s Ascend chips can’t do the job functions as a second veto point layered on top of Washington’s licensing regime. Compliance and supply-chain teams tracking chip availability for cross-border AI projects need to monitor both governments’ postures — a US export license is now only half the approval chain, not the finish line.

4. Watch packaging and architecture innovation, not just process-node headlines

Brookings’ reporting on Huawei’s LogicFolding stacking technique is a reminder that lithography-based export controls have a shelf life shaped by packaging innovation, not just EUV machine access. Investors and policy analysts should track advanced-packaging patent filings and foundry announcements out of China as a leading indicator of whether the current five-to-seventeen-times performance gap CFR describes is closing faster than official production figures suggest.

The Regulatory Question

What the H200 standoff reveals is that export controls can win the narrow fight they’re built for while leaving the broader strategic contest wide open. Washington’s licensing regime has achieved its most measurable goal: Nvidia’s China AI-chip revenue has gone from a multibillion-dollar business line to functionally zero, and the company has taken a $5.5 billion write-off along the way. That is a real, quantifiable policy success by the metric export-control advocates use most often.

But the deeper question — whether China’s aggregate AI compute capacity is meaningfully constrained — is not settled by license approvals or boycott headlines. Huawei’s revenue growth, DeepSeek’s pricing advantage, and Brookings’ warning about post-lithography packaging techniques all point toward a slower, cheaper, more domestically-directed Chinese AI buildout rather than a halted one. CFR’s performance data suggests the near-term gap remains wide, but the trajectory — narrowing from twenty-times to seventeen-times to, eventually, parity — is a multi-year bet, not a settled outcome. For now, the chip wall has done exactly what it was built to do to Nvidia’s balance sheet. What it has done to China’s underlying AI trajectory is still an open, and genuinely contested, question.

Frequently Asked Questions

Sources & Further Reading

• Ball game’s over — the US is out of the AI chip market in China — Brookings
• US says ban on AI chip shipments applies to Chinese firms outside China — Al Jazeera
• Why China is finally letting its AI firms buy the Nvidia H200 — South China Morning Post
• Huawei braces for $12 billion in AI chip revenue this year as Nvidia’s China market share hits zero — Tom’s Hardware
• US licenses Nvidia AI chip exports to China as trade rivalry continues — EM360Tech
• China’s AI chip deficit: why Huawei can’t catch Nvidia and US export controls should remain — Council on Foreign Relations
• US trade official says ‘very few’ Nvidia H200 AI chips have been shipped to China — CNBC