Nvidia H200 China Deal Stalls After Trump-Xi Summit: Why Beijing Won’t Let Its Tech Giants Take Delivery

When President Donald Trump flew to Beijing in December 2025 and brought Nvidia CEO Jensen Huang along at the last minute, the tech world braced for a breakthrough. Trump told reporters afterward that “something could happen” on chip exports. Two days later, nothing had. No Nvidia H200 chips have shipped to China since Trump first authorised the sales, and US Trade Representative Jamieson Greer later confirmed to Bloomberg that semiconductor controls weren’t even on the bilateral summit agenda.

The summit theatre, however, masked a far more consequential development—one that reveals a deliberate deadlock engineered by Beijing rather than blocked by Washington. The H200 isn’t stuck because the US won’t let it leave. The chips aren’t moving because China won’t let its own companies accept them.

The Licences Are Approved—But No One Is Taking Delivery

Washington has already cleared the path. Roughly 10 Chinese firms, including Alibaba, Tencent, ByteDance, and JD.com, hold approved US export licences for up to 75,000 units each. Lenovo and Foxconn are authorised as distributors. On paper, the pipeline is open.

Yet not a single H200 has reached Chinese soil since Trump first authorised the sales in December 2025. The bottleneck isn’t regulatory in the United States—it’s entirely domestic in China. Beijing has instructed its tech giants to limit their use of Nvidia chips to overseas operations while prioritising domestic chip manufacturing. The result is a policy standoff that no summit photo op could resolve.

Two Frameworks, One Irreconcilable Deadlock

Understanding why the H200 remains in limbo requires parsing two competing regulatory frameworks that are mutually exclusive by design.

US Export Rules Require Domestic Use

Under US law, all H200 chips ordered by Chinese clients must be used only in China. The export licences are conditional on geographic deployment within Chinese borders. This ensures that advanced semiconductor technology doesn’t get re-exported to third-party adversaries or used in ways that could circumvent US national security controls.

Beijing’s Directive Bans Domestic Use

Beijing, meanwhile, has instructed Chinese tech companies to limit Nvidia chip usage to overseas operations while supporting the domestic semiconductor ecosystem. The State Council has also ordered a supply-chain security review aimed at cutting dependence on US semiconductors. Commerce Secretary Howard Lutnick stated at a Senate hearing last month that Chinese firms are trying to keep their investment focused on domestic suppliers, including Huawei.

The two requirements cannot coexist. Chips cleared for export cannot legally be deployed where Beijing wants them deployed—outside China—and Beijing won’t authorise the domestic use that US licences require. According to analysts at Implicator, the policy contradiction is not an oversight. It is the point.

Why Beijing Prefers a Stalemate

By refusing to authorise domestic deployment, Beijing forces Chinese tech giants into a difficult choice: either forego Nvidia’s latest hardware entirely, or redirect investment toward domestic alternatives. Because the US demands domestic use, and Beijing forbids it, the only viable outcome is that Chinese firms accelerate their shift away from Nvidia. The deadlock effectively functions as a forced migration strategy.

What DeepSeek and Tencent Reveal About China’s New Strategy

While diplomats traded pleasantries in Beijing, the ground was shifting beneath them. Several data points emerged around the summit that matter far more for the long-term trajectory of the semiconductor race than any parting comment from Trump.

DeepSeek Optimises for Huawei Processors

DeepSeek confirmed that its latest model had been optimised to run on Huawei processors. This follows the April 2025 launch of DeepSeek V4, which adapted its architecture for Huawei’s Ascend chips—the first major Chinese frontier model to do so in training, not just inference. Training optimisation is critical because it represents a step beyond simply running pre-trained models on alternative hardware; it signals that Chinese AI developers are now designing their models from the ground up for domestic chips.

Tencent Sees GPU Supply Rising Through 2026

Tencent’s chief strategy officer stated publicly that Chinese GPU supply would increase progressively through 2026. This isn’t aspirational—it’s a supply-chain forecast from one of the country’s largest technology conglomerates. It suggests that domestic chip production, particularly from Huawei and others, is scaling faster than external observers may have assumed.

Alibaba Achieves Scaled Mass Production of Proprietary GPUs

An Alibaba executive confirmed that its T-Head proprietary GPUs had achieved scaled mass production. This is a significant milestone because T-Head is not a general-purpose GPU manufacturer; it is Alibaba’s in-house chip design arm, initially focused on AI inference accelerators for the company’s cloud infrastructure. Achieving scaled production means Alibaba can now deploy its own silicon at volume, reducing its dependence on Nvidia even without taking delivery of the H200s sitting in export limbo.

The Shift Is No Longer Experimental—It’s Operational

What the summit week confirmed is that China’s semiconductor pivot has moved beyond the experimental phase. It is now operational.

Just a year ago, the notion that Chinese hyperscalers could deploy proprietary GPUs at scale while simultaneously training frontier models on Huawei hardware seemed aspirational at best. Today, Tencent is forecasting GPU supply growth, DeepSeek is training on Ascend chips, and Alibaba is mass-producing its own silicon. The H200 deadlock has functioned as an accelerator for domestic alternatives—exactly as Beijing intended.

Why This Matters for Global Tech Investors

For business professionals tracking AI infrastructure, the implications are clear. The H200 standoff is not a temporary hiccup that will be resolved by the next trade negotiation. It is a structural shift designed to create permanent separation between Chinese AI development and US semiconductor supply chains.

• Nvidia loses a major market: China represented approximately 20-25% of Nvidia’s data centre revenue before export controls tightened. That revenue is being redirected to competitors like Huawei and domestic chipmakers.
• Chinese AI development decouples from US hardware: Models like DeepSeek V4 are being built on domestic chips from the training stage, not retrofitted for inference after the fact. This creates long-term dependency on Chinese supply chains.
• The US export control regime achieves the opposite of its intent: By restricting access to advanced chips, Washington hoped to slow Chinese AI progress. Instead, it has forced a faster pivot to domestic alternatives, creating a parallel AI ecosystem that may ultimately compete with—rather than rely on—US technology.

What Happens Next to the H200s?

The 75,000-unit licences held by Alibaba, Tencent, ByteDance, and JD.com are not cancelled. They simply sit unused—a testament to a policy standoff that neither side seems eager to resolve.

Could the chips eventually ship? Possibly, but only if either the US relaxes its domestic-use requirement or Beijing reverses its directive against domestic Nvidia deployment. Neither scenario appears likely in the near term. The US Commerce Department shows no sign of loosening geographic restrictions, and Beijing’s State Council supply-chain review is likely to reinforce, not weaken, the push toward semiconductor self-sufficiency.

For now, the H200 China deal survives only on paper. In practice, it is a policy contradiction that both Washington and Beijing find useful—Washington because it can claim it authorised the exports, and Beijing because it can force its tech giants to invest in domestic alternatives without taking the political blame for cutting off access to Nvidia.

The Bottom Line for Tech Leaders

The Trump-Xi summit was a diplomatic spectacle that changed nothing about chip exports. What changed was the public confirmation that China’s AI ecosystem is no longer dependent on Nvidia hardware to train and deploy frontier models.

• The H200 is not shipping because Beijing won’t let it, not because Washington blocks it.
• Licences are approved and ready—but the policy contradiction is intentional.
• DeepSeek, Tencent, and Alibaba are all scaling domestic alternatives that reduce their reliance on US chips.
• The shift is structural, not temporary, and it will reshape global AI supply chains through 2026 and beyond.

For anyone following AI infrastructure investment, the summit was a distraction. The real story happened in the data centres and design labs where Chinese engineers are building the next generation of AI models on chips that Washington cannot control.