What is the future of Nvidia’s products for the China market?

  • April 13, 2025

Investing.com -- Analysts continue to expect meaningful restrictions on Nvidia’s H20 chip in China, despite a recent media report suggesting the U.S. Commerce Department may allow shipments to proceed.

The H20, a performance-limited version of Nvidia’s flagship H100 GPU, was designed to comply with U.S. export controls by staying below a specific teraflop-per-square-millimeter performance threshold.

Nvidia (NASDAQ: NVDA ) has described the H20 as addressing "a very competitive space," but Morgan Stanley noted its performance is roughly 75% below that of the H100, which was barred from export to China last year.

When you force the best AI hardware company in the world to compete at what is now 85% or more below the state of the art performance level, pricing will be competitive,” analysts at Morgan Stanley wrote.

They estimate gross margins on the H20 are in the 50% range, well below Nvidia’s corporate average in the 70s, and said H20 strength contributed to weaker gross margins over the past two quarters.

A report on NPR.com suggested that although H20 was initially targeted for restriction, the U.S. government may now allow continued shipments. Morgan Stanley said the report "has not been confirmed" and maintained that restrictions remain its base case.

Morgan Stanley pointed to U.S. export bans on high-end chips, which would drive Chinese firms not buy the H20, while newer Blackwell GPUs offer as much as 25 times the performance,

That leaves Chinese companies with limited choices: using GPUs in foreign cloud platforms (where permitted), sourcing restricted chips through secondary markets (which is not allowed), or buying compliant chips like the H20 or AMD’s MI308.

The analysts said the public cloud is not a viable solution for all Chinese users and that back-channel procurement of high-end chips is likely overestimated, due to legal and financial risks.

Which leaves companies with option 3, analyst say. And while buying performance that far behind the state of the art isn’t optimal, it’s the only option available, and companies have shown good results.