US export controls are putting its AI leadership at risk and driving “half of the world’s AI talent” to its rivals, NVIDIA CEO Jensen Huang warned, blasting a new policy that had led to an $8 billion sales hit.

During an Q1 FY26 earnings call, Huang said NVIDIA had written off $4.5 billion in H20 chips destined for China and would lose out on further sales following the new export restrictions, which he said were based on the “always questionable” and now “clearly wrong” assumption that China could not make its own AI chips.

In an unusually fiery opening, Huang said the “AI race” was about “which stack the world runs on” and warned “shielding Chinese chipmakers from US competition only strengthens them abroad and weakens America's position… US global infrastructure leadership is at stake.”

While cautious to maintain praise for President Trump, Huang added the new licensing requirements made it “impossible” for NVIDIA to offer a chip “for any productive use” in China, essentially locking it out of the $50 billion market.

Inference surges cover China losses

Despite the export troubles, NVIDIA still grew its revenue by 12% to $44 billion for the quarter thanks to the continued rapid growth of its flagship Blackwell chip and a “surge in inference demand” driven by reasoning models.

CFO Colette Kress said the “sharp jump” in demand could be seen in the “leap in token generation” for OpenAI, Microsoft and Google, with Redmond alone recording a five times year-on-year increase and processing over 100 trillion tokens in Q1.

See also: NVIDIA: Customer failed to listen, toasted their copper cables

During Huang’s keynote at the GTC event back in March, he also claimed that the amount of computation needed for inference was “100 times more” then it used to be thanks to the new models.

Putting NVIDIA at the centre of that, Kress claimed it had already upped Blackwell’s performance by 1.5x in the last month and expected further inference improvements across its lifecycle.

The Dynamo inference framework running on Blackwell NVL72 rack system also improved AI inference throughput by 30x for new reasoning models, she added.

Sovereign AI 

Alongside inference surges, Huang said the sovereign AI sector was a “new growth engine” for NVIDIA, as he claimed investments countries were making into AI infrastructure were equivalent to those made during the rise of electricity and the internet, echoing comments he made in 2023.

Citing recently announced projects in Saudi Arabia and Taiwan, he said NVIDIA had seen countries “racing to build national AI platforms” as “every nation now sees AI as core to the next industrial revolution, a new industry that produces intelligence and essential infrastructure for every economy.”

As a result, Huang said NVIDIA has “line of sight” to sales for AI factory projects requiring “tens of gigawatts” of its AI infrastructure in the near future.

Though he did not specify how many of these were sovereign AI projects, he advised more governments in Asia, the Middle East and Europe were “building national AI factories to empower startups, industries and societies.”

The link has been copied!