Trump’s China rivalry and “AI race” delusion may endanger US, experts say

The US and China have begun discussions on an AI-safety notification mechanism ahead of meetings between Donald Trump and Xi Jinping, potentially creating a channel to alert each other about dangerous or unpredictable AI systems. Cooperation remains uncertain as the US expands AI-related export controls, Congress considers further restrictions on Chinese chip and hardware access, and the Justice Department has accused six leading Chinese AI companies of stealing from US frontier labs. The talks could shape global AI governance, but escalating technology rivalry limits the likelihood of durable coordination.

Analysis

The investable issue is not AI-safety cooperation itself, but whether it creates a narrow diplomatic off-ramp that slows the ratchet of semiconductor restrictions. A credible technical channel would modestly reduce near-term tail risk for China-exposed AI supply chains—ASML, TSM, AMAT, LRCX and NVDA—but would not alter the strategic logic of limiting leading-edge compute. Markets should therefore treat any de-escalation as a reduction in probability of incremental restrictions over the next 1-3 months, not as a reopening of China’s access to frontier accelerators.

The more likely second-order outcome is accelerated technological bifurcation: Chinese cloud and model developers substitute toward domestic compute, networking and memory ecosystems, while US suppliers face a structurally lower China revenue ceiling over 6-18 months. This favors relative beneficiaries of non-China AI capex—TSM and SK Hynix—over equipment names with more policy-sensitive China shipment exposure, particularly if licensing rules expand from GPUs to mature-node tools, HBM-related equipment, or chip-design software. A safety framework could also become a verification pretext: any reported model-security incident may be used politically to justify tighter export enforcement rather than looser rules.

Consensus may overprice a headline détente if it prompts an immediate rally in China-sensitive semis. The key falsifier for the bearish China-revenue view would be a formal licensing expansion, explicit carve-outs for compliant accelerator shipments, or evidence in quarterly results that China-derived revenue is stabilizing without margin concessions. Conversely, a new entity-list action, restrictions on inference-capable chips, or a further step-down in NVDA/AMAT/LRCX China guidance would validate the structural divergence thesis.

Key Decisions for Investors

• Do not chase any summit-driven semiconductor relief rally; use a 5-8% move in China-sensitive equipment names AMAT or LRCX to establish a 3-6 month relative short versus long TSM. Thesis: foundry demand diversification and advanced-node scarcity are more durable than equipment China shipment visibility. Exit if US policy formally broadens approved leading-edge chip exports or AMAT/LRCX raise China revenue guidance.
• Maintain core long NVDA only hedged through 3-6 month put spreads or a partial SOXX hedge around policy dates. NVDA's non-China hyperscaler demand can absorb lost volume, but valuation leaves limited room for a renewed export-control shock; a 10-15% policy-driven drawdown is plausible if inference-chip restrictions broaden.
• Watch-list SMIC and Hua Hong Semiconductor as potential tactical beneficiaries of domestic substitution, but require confirmation through rising utilization, pricing, and local AI-chip order commentary before entry. Diplomatic language alone is insufficient because mature-node oversupply can overwhelm the substitution narrative.
• Set event alerts for post-meeting communiqués and subsequent Commerce Department actions within 30-60 days. A concrete notification protocol without export-control concessions is neutral-to-negative for China-exposed hardware; an explicit technology carve-out would warrant covering the AMAT/LRCX relative short.