US weighs advanced Nvidia chip sales to China

US Commerce Secretary Howard Lutnick said President Donald Trump will decide whether Nvidia can sell its advanced H200 AI chips to China.

US officials are discussing if restrictions on Nvidia’s AI chip exports, first imposed in 2022, should be relaxed.

Lutnick noted that Trump is consulting multiple advisers before making a decision.

Allowing these sales would ease current limits that aim to prevent China and its military from accessing top US technology.

Some US lawmakers oppose any move to let Nvidia sell advanced processors to China, citing national security risks.

Nvidia CEO Jensen Huang is seeking approval to access the Chinese market, after Beijing’s earlier block on the company’s less advanced H20 chips.

Lutnick said there are differing views within the administration and industry regarding the sale.

No final decision or timeline has been announced.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

Policy shift could help Nvidia regain China sales amid chip sovereignty push

Analysts estimate Nvidia had about $30 billion in potential China sales at risk after export restrictions expanded in April 2025 1.
In 2024, China was 13% of Nvidia revenue, or $17.1 billion 2.
H200 approval would not guarantee gains. Beijing earlier discouraged Chinese firms from buying Nvidia chips for national security reasons, and pushed domestic options 2.
The H200 offers more memory plus bandwidth than the restricted H20, important for AI training and inference workloads at Chinese hyperscalers (large cloud and internet platforms that operate massive data centers) 2. It is unclear if downgrades (reduced-performance versions to meet export rules) are needed and whether Beijing would accept chips it labels a security risk 2.

China internet firms plan $70 billion-plus data center spend in 2026

Platforms plan over $70 billion in 2026, with data center power up 30% to 30 gigawatts in 2025 3.
Analysts expect liquid cooling (coolant-based heat removal for servers) to reach 45% penetration by 2030 on a 42% Compound Annual Growth Rate (CAGR) as AI racks exceed 100 kW (kilowatts) 4.
Chinese hyperscalers build data centers across Asia, the Middle East and Latin America to grow global AI application revenue, creating openings for suppliers 3. Non-IT infrastructure (power/cooling/networking equipment outside servers and chips) could reach RMB (renminbi) 800 billion by 2030 4.
Networking is moving to 400G and 800G (400/800 gigabit-per-second links) as China activates the world’s first 1,200G backbone (a 1.2 terabit-per-second long-haul network) to cut latency for AI inference 5.