US may allow Nvidia H200 chip exports to China, report says

Share:

Amid an ongoing fierce rivalry over the artificial intelligence space, the United States is mulling over the idea of allowing Nvidia to sell its H200 AI GPUs to China.
According to a new report from Semafor, the Commerce Department plans to greenlight the chip, which is roughly 18 months behind Nvidia’s cutting-edge models.
The development will mark a significant pivot from the Biden administration’s hardline export bans and will come as a calculated compromise meant to reopen the world’s biggest AI market while managing national-security concerns.
Why is Washington shifting on Nvidia chips?
The reported plan represents a strategic rethink inside the White House, driven by a growing realization that strict prohibitions may be backfiring.
By allowing exports of the H200, Commerce Secretary Howard Lutnick aims to strike a pragmatic middle ground.
It can provide US companies access to Chinese revenue while ensuring Beijing remains dependent on American technology standards rather than developing its own.
The policy logic hinges on the idea that total bans have inadvertently accelerated China’s domestic innovation.
The development comes days after Chinese tech company Moore Threads made a blockbuster, which raised roughly ¥8 billion ($1.1 billion) with shares soaring more than 500% on the first day.
Despite US restrictions, Chinese tech giants like Alibaba and Huawei have successfully built competitive AI models and hardware, filling the vacuum left by American companies.
Sources familiar with the internal discussions told Semafor that some administration officials now view previous controls as a failure that handed market share to rivals without meaningfully slowing China’s AI progress.
What this mean for Nvidia and the AI race?
For Nvidia, the decision could unlock billions in pent-up revenue.
China has historically been one of the chipmaker’s most voracious customers, accounting for up to 25% of its data center sales before the bans.
Reopening the door with the H200 would allow Nvidia to reclaim ground lost to Huawei’s Ascend series, which had begun to dominate domestic Chinese data centers in Nvidia’s absence.
From a technical standpoint, the H200 is a “goldilocks” chip for this political moment.
It offers significantly more memory bandwidth than the crippled H20 chip, which Beijing effectively banned its companies from buying due to its poor performance.
But it remains a generation behind the hardware powering US AI labs.
By shipping the H200, Washington hopes to entice Chinese firms back into the Nvidia software ecosystem (CUDA), reinforcing a long-term dependency that serves US strategic interests.
However, the move essentially concedes that the “containment” strategy has limits.
China’s rejection of the weaker H20 chip earlier this year was a wakeup call, proving that Beijing would rather build its own imperfect supply chain than buy hobbled American goods.
The post US may allow Nvidia H200 chip exports to China, report says appeared first on Invezz
US may allow Nvidia H200 chip exports to China, report says

Share:

Amid an ongoing fierce rivalry over the artificial intelligence space, the United States is mulling over the idea of allowing Nvidia to sell its H200 AI GPUs to China.
According to a new report from Semafor, the Commerce Department plans to greenlight the chip, which is roughly 18 months behind Nvidia’s cutting-edge models.
The development will mark a significant pivot from the Biden administration’s hardline export bans and will come as a calculated compromise meant to reopen the world’s biggest AI market while managing national-security concerns.
Why is Washington shifting on Nvidia chips?
The reported plan represents a strategic rethink inside the White House, driven by a growing realization that strict prohibitions may be backfiring.
By allowing exports of the H200, Commerce Secretary Howard Lutnick aims to strike a pragmatic middle ground.
It can provide US companies access to Chinese revenue while ensuring Beijing remains dependent on American technology standards rather than developing its own.
The policy logic hinges on the idea that total bans have inadvertently accelerated China’s domestic innovation.
The development comes days after Chinese tech company Moore Threads made a blockbuster, which raised roughly ¥8 billion ($1.1 billion) with shares soaring more than 500% on the first day.
Despite US restrictions, Chinese tech giants like Alibaba and Huawei have successfully built competitive AI models and hardware, filling the vacuum left by American companies.
Sources familiar with the internal discussions told Semafor that some administration officials now view previous controls as a failure that handed market share to rivals without meaningfully slowing China’s AI progress.
What this mean for Nvidia and the AI race?
For Nvidia, the decision could unlock billions in pent-up revenue.
China has historically been one of the chipmaker’s most voracious customers, accounting for up to 25% of its data center sales before the bans.
Reopening the door with the H200 would allow Nvidia to reclaim ground lost to Huawei’s Ascend series, which had begun to dominate domestic Chinese data centers in Nvidia’s absence.
From a technical standpoint, the H200 is a “goldilocks” chip for this political moment.
It offers significantly more memory bandwidth than the crippled H20 chip, which Beijing effectively banned its companies from buying due to its poor performance.
But it remains a generation behind the hardware powering US AI labs.
By shipping the H200, Washington hopes to entice Chinese firms back into the Nvidia software ecosystem (CUDA), reinforcing a long-term dependency that serves US strategic interests.
However, the move essentially concedes that the “containment” strategy has limits.
China’s rejection of the weaker H20 chip earlier this year was a wakeup call, proving that Beijing would rather build its own imperfect supply chain than buy hobbled American goods.
The post US may allow Nvidia H200 chip exports to China, report says appeared first on Invezz



