Nvidia shifts chip production on China import delays: sources

Nvidia has shifted Taiwan Semiconductor Manufacturing Company (TSMC) capacity from H200 chips intended for China to its newer Vera Rubin processors after US export controls and possible Chinese import limits stalled the approval process, two people with knowledge said.

The move implies Nvidia no longer expects significant near-term H200 sales in China following months of uncertainty over US approvals and Beijing signals, two people said.

H200 is an older AI processor positioned as compliant with US export controls, while Vera Rubin is the firm’s latest architecture, sought by US tech groups such as OpenAI and Google.

Nvidia heavily lobbied Washington and Beijing and began stepping up H200 production after President Donald Trump indicated in December he would permit sales, and the FT reported it had expected over 1 million orders.

The FT reported about 250,000 H200 chips had been produced, citing a person with knowledge, and Nvidia’s CFO Colette Kress said on a recent earnings call that small approvals had not generated revenue and future imports remained uncertain.

🔗 Source: Financial Times

🧠 Food for thought

Implications, context, and why it matters.

The China chip gamble has already cost Nvidia billions

Nvidia booked a $4.5 billion charge tied to surplus H20 inventory and related purchase commitments 1.
The prior quarter already put pressure on results, since Nvidia said sales of its export-compliant H20 chips were “insignificant” 2.
Next quarter’s outlook leaves out China data center revenue, so the guidance assumes no China data center compute sales 2.
A “stalled approval process” ties to uncertainty over US export licenses for China-bound H200 chips plus possible Chinese import limits. A legal analysis also links the delays to a disputed US policy that would tie licenses to payments to the US government, described as 25% of revenues from H200 China sales 3.

Freed capacity could widen the AI hardware gap between the US and China

Shifting Taiwan Semiconductor Manufacturing Company (TSMC) capacity from the older H200 to Nvidia’s newer Vera Rubin platform helps customers likely to adopt Rubin, including US tech groups such as OpenAI and Google 4.
This shift could stretch the US lead as China loses access to even older chips while local replacements slip further behind 5.
Estimates put America’s top AI chips at five times China’s performance today, with the advantage projected to reach seventeen times by 2027 5.
Nvidia keeps leverage over frontier AI since it can steer scarce leading-edge foundry capacity toward its newest architectures 6.