U.S. approves case-by-case exports of Nvidia H200 GPUs to China under new testing and supply conditions

The U.S. government has formally approved the export of Nvidia’s H200 GPUs to China, but only under a more restrictive framework that adds verification steps and new limits on volume, according to reporting by Data Center Dynamics published January 14, 2026. The decision marks a notable shift in the long-running tug-of-war between national-security concerns and the commercial reality that U.S. chipmakers derive significant demand from Chinese customers building AI infrastructure.

Under the updated approach, exports are not broadly reopened in a blanket way. Instead, licenses will be handled on a case-by-case basis, and the chips must be tested in third-party laboratories to confirm their technical capabilities before export. That testing requirement is designed to strengthen enforcement of performance thresholds and reduce ambiguity about what hardware is leaving the country—an issue that has repeatedly surfaced as chip designs evolve and as firms introduce slightly modified variants to meet regulatory lines.

The policy also introduces supply and volume constraints. Export approvals will be conditioned on “sufficient supply” for the U.S. market, and shipments to China are capped relative to the number of units sold to U.S. customers. In addition, Chinese purchasers must demonstrate adequate security procedures and attest that the chips will not be used for military purposes, placing compliance burdens on both exporters and buyers.

For Nvidia, the change potentially restores access to a high-value market segment at a time when demand for top-tier AI accelerators remains intense globally. But the compliance steps—testing, licensing, supply certifications and end-use restrictions—could add friction and time to the sales cycle. For Chinese firms, the rule signals that some access is possible, but only within a constrained and closely monitored channel that can be tightened again if geopolitical conditions change.

The broader significance is that AI competition increasingly runs through export-control design. Policymakers want to slow the transfer of advanced capability to military or intelligence use while avoiding collateral damage to domestic industry and jobs. The new regime attempts to strike that balance by allowing limited shipments of a defined product tier while keeping more powerful processors off limits, and by requiring technical and procedural checks intended to reduce diversion risk. How smoothly the system works—and whether it meaningfully changes AI compute availability in China—will depend on licensing throughput, enforcement rigor and supply conditions on both sides of the Pacific.