U.S.–China Trade Truce Talks Hit Chip Export Controls at Beijing Summit
Quick summary
Semiconductor rules stay central to any deal. Hardware procurement scenarios for developers through 2026.
Read next
- Trump Splits China Chip Policy: H200 Allowed, Blackwell Banned March 2026Trump administration allows Nvidia H200 exports to China while banning Blackwell B100/B200/B300. What this split policy means for cloud compute access and developers.
- Trump-Xi Summit April 2026: Blackwell Ban Holds, Taiwan Skipped, 1-Year TruceTrump-Xi Beijing summit ended with an uneasy truce. China failed to get Blackwell export controls lifted. Taiwan never came up. The deal expires in 12 months.
The US and China announced a trade truce on May 12, 2026, cutting US tariffs on Chinese goods from approximately 145% to 30% and Chinese tariffs on US goods from 125% to 10%. President Trump and President Xi are meeting in Beijing on May 14-15 — the first US presidential state visit to China since 2017. The specific technology-critical items on the negotiating table: AI chip export controls (particularly H200 and future-generation Nvidia GPU licenses), China's rare earth export restrictions imposed in retaliation for the H20 ban, and the antitrust investigations Chinese regulators had opened against US chip companies including Nvidia, Qualcomm, and Intel.
This is not a resolved trade conflict. It is a pause with specific items under active negotiation at the summit. Whether the tariff reduction leads to a lasting arrangement or is reversed in three months depends on what comes out of Beijing on May 14-15.
What Changed on May 12
The tariff reduction is reciprocal and immediate. US tariffs on Chinese goods dropped from approximately 145% (the cumulative rate after multiple escalation rounds through April 2026) to 30%. Chinese tariffs on US goods dropped from 125% to 10%.
The May 12 announcement followed two days of preparatory talks in Seoul, South Korea between US Trade Representative Catherine White and Chinese Vice Premier Liu He. The Seoul sessions focused specifically on technology: semiconductor export controls, rare earth export controls, and the resolution of Chinese antitrust investigations into US chip companies.
The concrete pre-summit agreements:
- Antitrust investigations suspended: China's SAMR (State Administration for Market Regulation) agreed to suspend its antitrust investigations into Nvidia, Qualcomm, and Intel pending the summit outcome. These investigations were widely understood as retaliatory measures rather than genuine competition concerns — the suspension is a goodwill gesture.
- Rare earth export controls discussed: China had imposed export licensing requirements on gallium, germanium, antimony, and graphite in response to the H20 ban. Partial relaxation is on the table at the summit.
- H200 export control case-by-case review: The H20 chip ban (imposed April 2025) remains in force. However, the export control posture for H200s is being discussed at the summit — moving from "presumption of denial" to "case-by-case" review for Chinese commercial customers would be a meaningful shift without formally lifting the ban.
What the Beijing Summit Is Actually Negotiating
The summit agenda, as described by US and Chinese government sources, covers four tech-critical items:
1. Semiconductor export control reciprocity: The US wants China to reduce its rare earth export restrictions. China wants the US to relax H200 export controls for commercial AI applications in China. The two countries are effectively trading hostages — critical mineral supply chains vs. frontier AI hardware access.
2. Taiwan: Taiwan is not on the formal agenda but is the underlying pressure driving China's willingness to negotiate. Chinese rhetoric about Taiwan has escalated since the Iran conflict drew US carrier groups to the Gulf. A summit framing that reduces US-China tension broadly benefits China's Taiwan posture by reducing the probability of simultaneous military crises.
3. AI governance framework: A bilateral AI safety dialogue was announced as a summit deliverable. The specific content — mutual notification of capability threshold crossings, information sharing on AI incidents — is thin on detail but represents acknowledgment that frontier AI development is a topic requiring intergovernmental coordination.
4. Energy and data center cooperation: The massive US and Chinese data center buildouts are both drawing on the same global supply chains for power infrastructure components (transformers, switchgear, cooling systems). The supply chain tension is mutual. A coordination discussion on power infrastructure components is reported as a secondary agenda item.
The China Domestic AI Context: Why Beijing Has Leverage
The US chip export control strategy assumed that restricting China's access to frontier AI hardware would create a permanent capability gap. The reality in May 2026 is more complicated.
Four Chinese open-weights models were released in a 12-day window in early May: GLM-5.1 from Z.ai, MiniMax M2.7, Kimi K2.6 from Moonshot AI, and DeepSeek V4. All four are competitive with Western frontier models on standard agentic engineering benchmarks, at under one-third of Claude Opus pricing for inference. Chinese domestic AI chip market share is projected to hit 50% in 2026.
The chip restrictions accelerated domestic development rather than preventing it. Chinese companies optimised models for efficient inference on domestically available hardware — Huawei Ascend 910B, Cambricon, and others — because they had no choice. The inference cost advantage these models now carry is a direct consequence of that optimisation pressure.
Beijing's leverage in the semiconductor negotiation: China can demonstrate that the H200 ban has not prevented frontier AI capability development. The ban's primary ongoing cost is training efficiency — Chinese labs train on chips that are 30-40% less efficient per watt than Nvidia's current generation. That cost is real but not decisive. The US's bargaining chip is reducing that training overhead, not blocking a capability that is already present.
Developer Impact of the Export Control Discussion
For developers building on AI APIs and infrastructure:
H200 availability in China: If the summit produces a case-by-case licensing regime for H200 exports to Chinese commercial customers, inference capacity for Chinese AI providers increases. Models deployed from Chinese providers (DeepSeek, Kimi, MiniMax) may become cheaper and faster to access — relevant if you're pricing non-US deployments against Western providers.
Nvidia revenue upside: Nvidia's Q1 FY2027 guidance of $78 billion assumed no China H200 recovery. Even partial H200 license approvals would represent upside to that guidance. The market is watching the summit for any signals on this specific point.
Rare earth supply chains: Gallium, germanium, and antimony are critical materials for compound semiconductors (GaAs, GaN) used in RF chips, power electronics, and some optical components. Relaxation of Chinese export restrictions reduces input cost risk for semiconductor manufacturers outside China — relevant for long-term data center power electronics supply chains.
AI safety dialogue: The bilateral AI safety framework is unlikely to have near-term operational effects. It is the first step in a process that, if sustained, could eventually produce mutual notification agreements for capability threshold crossings. For developers: no immediate impact, but it is worth noting that the world's two largest AI-developing nations are attempting a governance conversation.
What Happens If the Summit Fails
The tariff reduction is currently scheduled to remain in force for 90 days regardless of summit outcome — it is a ceasefire, not a peace treaty. If the Beijing summit produces no durable agreement, tariffs revert to 145% and 125% in mid-August.
In that scenario: Nvidia's H20 ban remains in force, Chinese retaliatory antitrust investigations resume, rare earth export restrictions continue, and US-China AI infrastructure development continues on separate, decoupled tracks.
The decoupling trajectory — separate AI ecosystems, separate hardware supply chains, separate frontier model development — has been the default path since 2022. The summit is an attempt to negotiate exceptions and guardrails within that trajectory, not to reverse it.
Key Takeaways
- May 12 trade truce: US tariffs on China 145% → 30%; Chinese tariffs on US goods 125% → 10%; 90-day ceasefire; Beijing summit May 14-15
- Tech items on the table: H200 export control case-by-case review, China rare earth export restriction relaxation, SAMR antitrust investigations into Nvidia/Qualcomm/Intel suspended pending summit
- China's leverage: 4 open-weights models in 12 days; Chinese domestic AI chip market at 50% share; H200 ban accelerated rather than blocked domestic AI capability development
- Nvidia upside scenario: Q1 FY2027 $78B guidance assumed zero China H200 recovery; partial H200 licenses = material upside; watching summit for signals
- 90-day window: If no durable deal by mid-August, tariffs revert; antitrust investigations resume; decoupled trajectory resumes
- Developer impact: Cheaper inference from Chinese providers if H200 access improves; rare earth relaxation reduces long-term data center power electronics supply chain risk
For the semiconductor supply chain context, read TSMC Q1 2026: Record Profit, HBM4 Sold Out, OpenAI Titan Chip Tape-Out. For Nvidia's revenue exposure to the China ban, read the guidance analysis in context of the Anthropic compute expansion: Anthropic Leases SpaceX Colossus 1: 220K GPUs.
FAQ
Frequently Asked Questions
What happened with US-China trade talks on May 12, 2026?
The US and China announced a trade truce on May 12, 2026, cutting US tariffs on Chinese goods from approximately 145% to 30% and Chinese tariffs on US goods from 125% to 10%. The announcement followed two days of preparatory talks in Seoul focused on technology trade. President Trump and President Xi are meeting in Beijing on May 14-15 in the first US presidential state visit to China since 2017. The truce is a 90-day ceasefire — if the Beijing summit produces no durable agreement, tariffs revert to previous levels in mid-August.
What semiconductor export control changes could come out of the Beijing summit?
The primary semiconductor item at the Beijing summit is the H200 export control posture. Currently, H200 exports to China operate under "presumption of denial." Moving to "case-by-case" commercial review is the most likely outcome of a successful summit — this would not lift the ban but would create a pathway for Chinese commercial AI companies to receive H200 GPUs with individual licenses. China is also offering relaxation of its rare earth export restrictions on gallium, germanium, antimony, and graphite in exchange. The H20 chip ban (for H20 specifically) is a separate and less likely item to be addressed.
Why does the US-China trade truce matter for AI and developers?
Three developer-relevant effects. First, Nvidia's Q1 FY2027 guidance of $78 billion assumed zero H200 China recovery — any H200 license approvals represent upside to that guidance and potentially lower long-term GPU pricing as supply increases. Second, Chinese AI inference providers (DeepSeek, Kimi, MiniMax) could expand capacity if H200 access improves, making their already-low pricing more competitive for non-US deployments. Third, rare earth export restriction relaxation reduces input cost risk for compound semiconductors used in power electronics and RF components — relevant for data center supply chains over a 2-5 year horizon.
Did US chip export controls slow down China's AI development?
Not as much as intended. Four Chinese open-weights AI models — GLM-5.1, MiniMax M2.7, Kimi K2.6, and DeepSeek V4 — were released in a 12-day window in early May 2026, all competitive with Western frontier models on engineering benchmarks at under one-third of Claude Opus inference pricing. Chinese domestic AI chip market share is projected to hit 50% in 2026. The export controls increased training costs (domestic chips run 30-40% less efficiently per watt than Nvidia's current generation) but accelerated domestic hardware and model optimization rather than blocking capability development. Beijing's negotiating position reflects this reality.
Free Weekly Briefing
The AI & Dev Briefing
One honest email a week — what actually matters in AI and software engineering. No noise, no sponsored content. Read by developers across 30+ countries.
No spam. Unsubscribe anytime.
More on Geopolitics
All posts →Trump Splits China Chip Policy: H200 Allowed, Blackwell Banned March 2026
Trump administration allows Nvidia H200 exports to China while banning Blackwell B100/B200/B300. What this split policy means for cloud compute access and developers.
Trump-Xi Summit April 2026: Blackwell Ban Holds, Taiwan Skipped, 1-Year Truce
Trump-Xi Beijing summit ended with an uneasy truce. China failed to get Blackwell export controls lifted. Taiwan never came up. The deal expires in 12 months.
QUAD New Delhi May 26 Outcomes: Tech, Energy, Defence and Developer Impact
A practical read of QUAD New Delhi outcomes and what they mean for cloud regions, supply chains, data residency, and enterprise vendor strategy.
Iran Deal Makes Trump-Xi Summit 65% Likely — GPU Prices Next
Iran nuclear deal shifts Trump-Xi summit probability from 40% to 65%. A 145%→25% tariff rollback is now a real scenario. What it means for GPU and server pricing.
Free Tool
Will AI replace your job?
4 questions. Get a personalised developer risk score based on your stack, role, and what you actually build day to day.
Check Your AI Risk Score →Written by
Software Engineer based in Delhi, India. Writes about AI models, semiconductor supply chains, and tech geopolitics — covering the intersection of infrastructure and global events. 952+ posts cited by ChatGPT, Perplexity, and Gemini. Read in 167 countries.
