AI Safety Talks Advance at the Trump-Xi Summit as Chip Export Controls Stall in Congress
The Washington summit between Donald Trump and Xi Jinping has produced the outline of a standing AI-safety channel between the two governments, with one deliberate omission: chip export controls stayed off the AI agenda. US Trade Representative Jamieson Greer said the controls were not part of the discussions around the new safety mechanism, and the channel carries no signed agreement and no formal name. The split between diplomatic machinery for AI risk and the enforcement file that governs the hardware is the defining outcome of this week's meeting.
Senate Democrats used Xi's state visit to press for floor votes on legislation tightening Beijing's access to advanced accelerators and chipmaking equipment. Minority Leader Chuck Schumer has pushed Majority Leader John Thune to schedule votes on three bipartisan bills, and he has criticised Trump's decision late last year to license exports of Nvidia's H200 chips to China on the grounds that the sales give Beijing an edge in the global AI race.
That legislative push has collided with the summit's diplomatic logic. China hawks in Congress have grown frustrated with the administration's reluctance to act on tech restrictions without a deal, and the vote count for the three bills remains uncertain. The practical lever over accelerator flows still sits with the White House rather than the Senate floor.
What the summit settled
The safety mechanism would sit alongside the existing trade and technology talks instead of inside them, giving the two governments a dedicated line for shared safety standards. That would be a first for the relationship. Nothing has been signed, and the arrangement rests on stated intent, with no binding obligation attached.
A channel without an agreement carries no reporting duties, no verification mechanism and no escalation protocol. Until something is signed, it works as a venue for talking, and either government can walk away from it at no cost.
The two leaders have been here before. An earlier round of talks this year covered shared AI guardrails and ended without an agreement, and Greer said afterwards that chip export controls had not been discussed at all.
For Beijing, the channel's value lies in the appearance of parity, since it treats China as a peer in managing frontier risk. For Washington, the appeal is dialogue without concession, because a safety conversation costs nothing on the export ledger. Both readings leave the same question open, which is what the channel can deliver when the two governments disagree about the underlying threat.
Why chip export controls stayed off the table
The agenda gap was visible before the leaders met. Beijing pushed for semiconductor export controls and access to advanced AI chips to be discussed, while Washington kept both off the AI-safety track. The rest of the agenda covered AI safety protocols, possible US sanctions tied to Iran, rare-earth supply chains and Taiwan.
Chips are the one asset Washington has refused to trade inside a diplomatic package, which is why the two tracks were kept apart. That separation leaves the export regime as the sole constraint on China's accelerator supply, and it puts the enforcement record under more scrutiny than the safety dialogue.
Beijing applies the same logic in reverse. It has blocked Western acquisitions in the region, including a move to stop Meta's purchase of the Singapore-based Manus, which shows both capitals now treat technology access as a security instrument rather than a commercial question.
Why Huawei's numbers matter
Huawei's accelerator stack is the counter-evidence to the claim that export controls have failed. The company's latest SuperPoD architecture links far fewer processors than originally planned, and each chip delivers roughly half the compute of Nvidia's equivalents. Huawei chairman Eric Xu has acknowledged that the company may not manufacture enough chips to meet domestic demand.
Xu has also suggested that Chinese AI developers may not be advanced enough to encounter the rogue-model safety risks that dominate frontier-lab debate in the United States. That assessment undercuts the premise that the safety channel addresses an immediate shared danger, and it sits awkwardly beside the alarm in Washington over Chinese capability gains.
The read-across is narrower than either camp claims. The controls have not stopped China's accelerator programme, and Huawei is still shipping a domestic alternative at scale. They have slowed it: fewer linked processors, lower per-chip compute, and a supply constraint the company's own chairman concedes. That is a measurable lag rather than a blockade, and it is the strongest evidence the export regime has produced so far.
The enforcement gap
Licensing has proved a weaker instrument than the political fight over it suggests. Washington has approved H200 sales to roughly ten Chinese firms, yet very few units have shipped, which leaves the concession on paper and the capability gap intact. The H200 is Nvidia's second-most-advanced accelerator, which is why the decision to license it drew objections from senior Democrats instead of a routine procedural complaint.
Nvidia's more advanced B200 chips have reached China through third countries despite the controls, routes the administration has tried to shut down. Enforcement therefore has two fronts: the licensing desk, where sales are authorised, and the transshipment networks, where the hardware actually moves.
Trump paused for one year an expanded Commerce Department blacklist of companies barred from buying US technology goods, semiconductor manufacturing equipment among them, as part of the deal he struck with Xi in South Korea. The pause removed the main enforcement tool aimed at the equipment layer while leaving the licensing regime in place, a combination that lets sales proceed while blocking the tools China would need to build substitutes. That one-year window expires later this year, which puts the blacklist back on the calendar beside the APEC and G20 meetings.
The administration accuses Chinese developers of extracting capabilities from American models at industrial scale. Anthropic chief executive Dario Amodei has argued for stronger export controls, a crackdown on chip smuggling, and measures to stop that distillation, presenting all three as necessary to protect an American lead.
The dispute bears directly on Nvidia, whose China revenue depends on licensing decisions the White House can reverse. Congressional action that tightens the rules would constrain the company's access to that market, while inaction leaves the question to the administration's discretion.
Tariffs and weapons sales give the White House separate leverage. Trump has described future arms sales as a negotiating chip, and the administration is weighing a delay of further sales until after the APEC and G20 summits later this year, where the two leaders are expected to meet again. On trade, Trump has given no sign of abandoning a base tariff above 10 percent or of surrendering his discretion to restrict imports selectively.
Allied strain complicates the picture further. Tariffs target China's reliance on American buyers, while export controls target a deeper dependency on advanced chips and manufacturing equipment. That leverage runs both ways, since the controls also cost US suppliers revenue and push Beijing toward the domestic substitutes Huawei is already selling.
| Summit agenda item | Where it stands |
|---|---|
| AI safety channel | Discussed; no signed agreement, no formal name |
| Chip export controls | Excluded from the AI dialogue, per USTR Greer |
| H200 sales to China | Licensed to about ten Chinese firms; few units shipped |
| Commerce blacklist expansion | Paused for one year under the South Korea deal |
| Tariffs | Base rate above 10 percent retained; no pledge to change |
| Rare earths, Iran sanctions, Taiwan | On the agenda alongside AI safety |
Why this matters
Washington and Beijing can agree to talk about AI safety while leaving the hardware competition untouched, which means the export regime remains the only real constraint on China's accelerator stack. For chipmakers, cloud operators and investors, the signals to track are the licensing decisions and the transshipment routes around them. The blacklist pause and the APEC and G20 meetings give two dates against which to measure whether those controls tighten or loosen.
Related Articles
- AI Chip Controls and Safety Pledges Converge as Lawmakers Press Trump Before Xi Visit
- US and China Launch High-Level AI Regulation Dialogue to Set Safety Standards
- Johnson Bets on a Single Summit as AI Safety Legislation Stalls
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.