China’s AI chip breakthrough meets Europe’s Nvidia pressure—while Beijing tests 100GW microwave power

China is moving from a prolonged AI-chip standoff toward a more permissive stance as the United States approves Nvidia’s H200 for export to China in early 2026, and the latest reporting suggests Beijing is “finally” allowing its AI firms to buy the part. The shift follows months of unusual friction over cutting-edge accelerators that underpin frontier model training and inference. While the articles do not detail the exact licensing mechanics, the core development is that a previously constrained supply channel is opening for a high-end Nvidia GPU. That matters because it signals a partial normalization of commercial access even as export controls remain a strategic lever. Geopolitically, the episode sits at the intersection of US technology containment and China’s push for AI self-sufficiency. Washington’s decision to approve the H200—rare enough to be highlighted—creates a wedge: it can reduce China’s immediate compute bottlenecks while still preserving broader restrictions on the most advanced tiers. Beijing, for its part, benefits from faster scaling of domestic AI capabilities, which can translate into competitive advantage across surveillance, industrial automation, and defense-adjacent applications. Europe’s parallel scrutiny of Nvidia by the French competition authority adds another layer: even if chips flow, market power and pricing practices are becoming a political-economic battleground. On the market side, the H200 approval and China’s uptake are bullish for Nvidia’s data-center revenue mix, but the magnitude is likely tempered by ongoing export-control boundaries and China’s ability to substitute with domestic or alternative accelerators. In Europe, Cerebras’ plan to add 200MW of AI compute capacity by end-2027 intensifies competitive pressure on Nvidia’s ecosystem, potentially shifting demand toward specialized architectures and cloud/colocation operators. The microwave-weapon disclosure—China’s high-power microwave (HPM) arsenal reportedly reaching up to 100 gigawatts—adds a security premium to resilience spending, with knock-on effects for defense electronics, electromagnetic compatibility testing, and critical-infrastructure hardening. Investors should expect cross-currents: AI-chip sentiment supported by supply access, while regulatory and competitive dynamics cap upside. Next, watch for licensing granularity (which SKUs, which end-users, and what compliance conditions) and for any US follow-on tightening or clarifications that could reverse the “opening” narrative. For Europe, the near-completion of the French Nvidia probe is a near-term catalyst: outcomes could include remedies, fines, or behavioral commitments that affect pricing and distribution. For the compute buildout, track Cerebras’ customer announcements, power-purchase agreements, and colocation partnerships tied to the 200MW target. Finally, the HPM research milestone should be monitored for follow-on tests, procurement signals, and any export-control spillovers into dual-use microwave components that could affect both defense and civilian RF supply chains.

Geopolitical Implications

• 01

  Partial easing of chip restrictions can accelerate China’s AI scaling while preserving US leverage through selectivity on top-end tiers.

• 02

  European regulatory enforcement is becoming a parallel front in the AI hardware contest, potentially constraining Nvidia’s monetization of market power.

• 03

  Directed-energy progress (HPM) suggests future pressure on communications, sensors, and critical infrastructure—raising the strategic value of EMC hardening and defense electronics.

• 04

  The combination of compute access and security R&D underscores a broader trend: AI industrial policy and defense technology are converging into a single strategic competition.

Key Signals

• —Details of which H200 SKUs and end-users are licensed for China, and whether conditions tighten or loosen after this reported opening.
• —Timing and substance of the French Nvidia probe outcome, including any remedies or fines that affect distribution and pricing.
• —Cerebras’ named customers, power contracts, and colocation partnerships supporting the 200MW target by end-2027.
• —Follow-on HPM demonstrations, procurement signals, and dual-use export-control spillovers into microwave/RF components.