Our Take
The stalemate is structural, not diplomatic: US rules require domestic-only use; Beijing demands overseas-only deployment. Huawei gains time to close the performance gap while Nvidia's China revenue collapses to near zero.
Why it matters
This is no longer a trade negotiation—it is supply-chain policy. The question of which AI hardware architecture dominates the world's second-largest AI market is being decided by government mandate, not technical merit.
Do this week
Enterprise AI teams: audit your China-facing inference workloads now to measure whether Huawei Ascend performance meets your latency and throughput targets before 2026 mandates eliminate H200 as an option.
The mechanics of the stall
President Trump visited Beijing in early 2026, brought Jensen Huang to the summit at the last minute, and departed after telling reporters that "something could happen" on chip exports. Nothing has. Not a single Nvidia H200 has shipped to China since Trump authorised the sales in December 2025, despite roughly 10 Chinese firms (per the source: Alibaba, Tencent, ByteDance, and JD.com) holding approved US export licences for up to 75,000 units each.
The blockage is not in Washington. US Trade Representative Jamieson Greer told Bloomberg that semiconductor controls were not even on the bilateral agenda. The obstruction is in Beijing.
The contradiction is deliberate. US export rules require that all H200 chips ordered by Chinese clients be used only within China. Beijing, meanwhile, has instructed Chinese tech companies to limit their use of Nvidia chips to overseas operations while concentrating investment in domestic suppliers, including Huawei. Two mutually exclusive requirements. Commerce Secretary Howard Lutnick confirmed at a Senate hearing that Chinese firms are being steered to domestic alternatives, and Beijing's State Council has ordered a supply-chain security review aimed at cutting dependence on US semiconductors.
Huawei's window is closing fast
The summit week produced data points that matter far more than Trump's parting comment. DeepSeek confirmed its latest model had been optimised to run on Huawei processors. Tencent's chief strategy officer stated Chinese GPU supply would increase progressively through 2026. An Alibaba executive said its T-Head proprietary GPUs had achieved scaled mass production.
This follows April's launch of DeepSeek V4, adapted for Huawei's Ascend chips during training, not just inference. The first major Chinese frontier model to do so. What the summit week confirmed is that the shift from experiment to supply-chain policy is complete.
Nvidia's China revenue has fallen to roughly 5% in recent quarters, down from above 20% before export controls tightened (per company guidance). The company assumes zero China revenue for the current quarter. Huang's last-minute inclusion in the Trump delegation suggested urgency. The outcome suggests the limits of CEO diplomacy when the obstruction is structural.
Beijing is now steering AI platforms toward Huawei's compute stack not as a temporary hedge but as a structural bet that the performance gap will close fast enough that lock-in is manageable. DeepSeek V4's results suggest it may be right, at least for inference workloads.
What this means for AI infrastructure decisions
The H200 deal is approved, licensed, and frozen. Huawei is filling the space it leaves. For teams building inference-heavy systems destined for China or East Asia, the window to evaluate Huawei Ascend silicon is now. Performance benchmarks between H200 and Ascend on real workloads (not vendor claims) should drive your architecture choice this quarter, not next year. Beijing's mandate removes optionality; your supplier lock-in will be enforced by policy, not cost.
Teams deploying in North America or Europe should expect Nvidia margin pressure as the company compensates for China revenue collapse. Multi-year pricing locks are worth negotiating now before competition from Huawei's export variants tightens terms.