Nvidia announced Tuesday it will take a $5.5 billion charge after the U.S. government restricted exports of its H20 AI chip to China — a major blow to the chipmaker’s operations in one of its largest markets.
The H20 chip, designed to comply with earlier U.S. export controls, had become Nvidia’s most advanced AI chip available in China, widely adopted by tech giants including Tencent, Alibaba, and ByteDance. However, U.S. officials have now determined that the H20’s high-speed memory and connectivity capabilities could potentially aid in building Chinese supercomputers, triggering indefinite export restrictions.
Nvidia stated the U.S. government notified the company on April 9 about licensing requirements for the H20 and confirmed the rules would remain indefinitely as of April 14. The $5.5 billion in charges reflects costs tied to inventory, purchase commitments, and reserves related to the restricted chip.
Despite H20 being less powerful than Nvidia’s global chips for training AI, it is highly effective for AI inference — a growing area in the AI chip market. Its increasing popularity among Chinese firms drew scrutiny, especially following reports that companies like Tencent may have deployed the chip in AI training facilities.
While Tencent has denied any violations, a recent report from the Institute for Progress argues otherwise, suggesting that installations of H20 chips could have breached U.S. supercomputer restrictions. Nvidia has declined to comment beyond its regulatory filing, and the U.S. Commerce Department has yet to issue a statement.
The export setback comes just days after Nvidia revealed plans to invest $500 billion in building AI servers within the U.S. over the next four years, aligning with U.S. efforts to boost domestic tech manufacturing.
