• NVIDIA has received U.S. government licenses to supply its H20 AI chips to several major customers in China, a significant reversal from an April 2025 ban.
  • The licenses come after the company was forced to write off $5.5 billion in inventory due to the initial export restrictions, highlighting the profound impact of geopolitical regulation.
  • Despite the approvals, supply of the H20 remains tight, with demand from Chinese tech giants far outstripping NVIDIA's current production capacity for the near term.

A Policy Reversal Unlocks Critical Supply

NVIDIA has successfully obtained licenses from the U.S. government to resume shipments of its H20 artificial intelligence chips to several key customers in China, according to an executive speaking at a Goldman Sachs conference. The development signals a notable shift in U.S. export policy, which had previously blocked these sales in April 2025, a move that forced the chipmaker to take a massive $5.5 billion inventory write-down.

The H20 is the most advanced AI chip the U.S. currently permits for export to China, making it a critical piece of infrastructure for Chinese companies engaged in large-scale AI training and deployment. The approval of these specific licenses, granted in July, suggests a strategic willingness by U.S. officials to allow controlled exports of this sensitive technology while maintaining leverage over the pace of China's AI advancement.

Surging Demand Meets Constrained Supply

The greenlight has unleashed pent-up demand from China's tech sector. Shortly after the policy reversal, NVIDIA placed a large order with its manufacturing partner TSMC for approximately 300,000 H20 chips, an order valued at nearly $4 billion, according to people familiar with the matter. This indicates that major Chinese cloud and AI companies are moving quickly to secure their supply of the high-performance semiconductors.

However, efforts to rapidly restock are hitting a snag. NVIDIA has not yet resumed full-scale production of the H20 for the near term, creating a significant supply bottleneck. This means that even with licenses in hand, the company's ability to fulfill the surge of orders from Chinese customers will be constrained in the coming quarters, potentially slowing AI project rollouts in the region.

A spokesperson for NVIDIA declined to comment on specific customer orders or production timelines.

Navigating a Complex Regulatory Landscape

The situation remains fluid. The Cyberspace Administration of China had previously temporarily paused new domestic purchases of the chips, citing the need for security reviews—a move NVIDIA has rejected as unfounded. This creates a dual-layer of regulatory uncertainty for the company, which must now navigate approval from both U.S. export authorities and Chinese cybersecurity officials.

This development is the latest in a series of cycles involving approval, ban, and resumed export of advanced semiconductor technology to China. It reflects a broader U.S. strategy of "managed decoupling," where certain trade is permitted under strict conditions. In some cases, these conditions have included requirements that a portion of the GPUs sold to China be delivered to the U.S. government, though it is unclear if such a stipulation is attached to these latest H20 licenses.

For now, the licenses provide a crucial, though limited, lifeline for both NVIDIA's bottom line and China's ambitions to remain competitive in the global AI race, where access to cutting-edge hardware is paramount.