Nvidia is preparing to resume large-scale shipments of advanced AI chips to China, marking a significant shift in the global semiconductor standoff that has defined the past three years.
According to sources cited by Reuters, Nvidia plans to begin shipping its H200 AI accelerators to China by mid-February, just ahead of the Lunar New Year. The initial deliveries are expected to come from existing inventory and could total 40,000 to 80,000 H200 GPUs, packaged as 5,000 to 10,000 modules, depending on final approvals.
If completed, this would be the first time since 2022 that Chinese customers gain legal access to an AI processor of this class.
Why The H200
The H200 is based on Nvidia’s Hopper architecture and represents a major leap over the China-specific H20 variant that was engineered to comply with earlier U.S. export limits. For Chinese cloud and AI firms, the difference is substantial: the H200 delivers roughly six times the performance of the H20 in key AI workloads.
Although Nvidia has already moved on to its newer Blackwell platform, with Rubin next in line, the H200 remains a highly capable accelerator for training and inference at scale. Its return to relevance highlights how geopolitics can revive “previous-generation” silicon overnight.
A Policy Reversal With Conditions
The planned shipments follow a policy reversal by the Trump administration, which earlier this month said it would allow H200 exports to China under strict conditions. Those include:
- Government approval in both Washington and Beijing
- Vetted end customers
- A 25% revenue fee paid to the U.S. Treasury
The move represents a sharp departure from the Biden-era restrictions that banned advanced AI chip exports to China on national security grounds.
Reuters reported that the U.S. has already launched an inter-agency review of license applications, signaling that the new policy is moving beyond rhetoric into execution.
Uncertainty Remains In Beijing
Despite Nvidia’s preparations, nothing is final yet. Sources say Beijing has not approved any purchases so far, and Chinese officials are still debating how to respond.
One proposal under discussion would require Chinese buyers to bundle each H200 purchase with domestically produced chips, an attempt to balance short-term AI needs with long-term industrial policy.
Emergency meetings were reportedly held earlier this month as officials weighed whether allowing imports could slow China’s domestic chip development, which still trails Nvidia at the high end.
Who Benefits If Shipments Proceed
For Chinese tech giants such as Alibaba and ByteDance, access to the H200 would provide a powerful boost to AI training capacity at a time when demand for compute continues to explode.
For Nvidia, the deal allows it to monetize surplus Hopper inventory without diverting production from Blackwell and Rubin, which remain prioritized for U.S. and allied markets.
In a statement to Reuters, Nvidia said licensed H200 sales to China would not affect its ability to supply U.S. customers, emphasizing that it is “continuously managing” its global supply chain.
A Sign The Chip War Is Changing
While export controls remain firmly in place, the H200 deal shows that the AI chip conflict is entering a more nuanced phase. Instead of blanket bans, access is increasingly shaped by licensing, fees, and political negotiation.
If approved, the shipments would redraw the lines of the chip war, not by ending it, but by proving that economic pressure, inventory realities, and AI demand can bend even the toughest restrictions.
As always, the real test will come after February: whether this is a one-off exception, or the beginning of a broader recalibration of AI chip trade between the world’s two largest technology powers.
