- The shift from “presumption of denial” to “case-by-case review” for NVIDIA H200-equivalent chips gives the Commerce Department discretion to approve exports that were previously blocked — a significant loosening for some buyers while maintaining restrictions on China.
- The 25 percent tariff on imported AI chips not sourced from the US supply chain incentivises global buyers to procure American products and penalises alternatives — commercial policy dressed as security measure.
- Nvidia’s Jensen Huang attended the Trump-Xi summit in Beijing on May 14, signalling that the company is seeking to restore some Chinese commercial access under the new case-by-case framework.
Two regulatory instruments released in January 2026 have reshaped the global market for advanced AI chips. The Bureau of Industry and Security issued a final rule changing the license review posture for NVIDIA H200- and AMD MI325X-equivalent chips from “presumption of denial” — the default applied to China-bound exports since 2023 — to “case-by-case review,” meaning individual export applications will be evaluated on their specific end-user and end-use merits rather than rejected automatically. Simultaneously, a presidential proclamation imposed a 25 percent tariff on imported AI chips not sourced from the American technology supply chain.
The two instruments pull in somewhat different directions. The BIS rule change potentially opens markets that were previously closed; the tariff penalises foreign-made chips sold in or through the US. Together, they reflect an administration navigating between commercial pressure from Nvidia — whose revenue depends on large addressable markets — and national security pressure to prevent China from accessing hardware that would accelerate its AI and military modernisation programmes.
Who benefits and who doesn’t
The case-by-case framework creates a tiered access system. US allies in Europe, Japan, South Korea, and Australia will likely see approvals processed quickly, since the BIS framework already distinguished between Tier 1, Tier 2, and restricted-country buyers in its chip export architecture. The Middle East presents a more complex case: Saudi Arabia and the UAE have both sought Nvidia hardware for sovereign AI programmes, and their requests had been complicated by proximity to China-linked technology networks. The case-by-case approach gives the administration flexibility to manage those relationships individually.
China remains effectively excluded from direct access to H100- and H200-class hardware, regardless of the label change. The case-by-case standard still applies US security guidelines that would, in practice, result in denial for most Chinese end-users. What the change may enable is a modest expansion of approvals for Hong Kong-based entities, joint ventures with non-Chinese headquarters, and research institutions that the previous blanket approach had caught in the same net as state-linked buyers.
Nvidia at the summit
Jensen Huang’s presence at the Trump-Xi summit in Beijing on May 14 — alongside Tesla’s Elon Musk, as part of a delegation of major American business leaders — was one of the most visible signs that chip export policy is now a direct subject of bilateral diplomacy. Huang did not announce new sales agreements, and the summit readout did not specify outcomes on hardware access. The signal of his attendance, however, was broadly read as an indication that the administration regards some level of commercial re-engagement with China on AI chips as part of whatever deal emerges from the Beijing meetings.