The mechanism. On January 15, 2026, the Bureau of Industry and Security published a rule flipping the license-review posture for advanced computing exports to China and Macau — including Nvidia's H200 and AMD's MI325X — from a presumption of denial to case-by-case review, conditioned on revenue-sharing (Nvidia at 25%, AMD at 15%) and other terms. That policy is a dial, not a switch. BIS can tighten the "case-by-case" conditions, expand the entity list, or slow-walk approvals at any point — and because the framework is discretionary rather than statutory, every quarter's China GPU revenue is effectively rented from Washington, not owned. The market has priced this as an AMD-vs-Nvidia "who wins China" story. It's actually a "who has more to lose" story, and the answer inverts the market-share narrative.
Who cashes in if licensing stays open or widens. NVDA still captures the lion's share of any incremental China allowance in pure dollar terms — Nvidia's China-directed datacenter GPU revenue has historically run in the billions per quarter, dwarfing AMD's equivalent line, so every basis point of license approval is worth more to Nvidia's income statement even after the 25% government cut. AVGO benefits regardless of which GPU vendor wins China, since it supplies custom AI ASICs and networking silicon (Tomahawk/Jericho switch chips) to hyperscalers building out capacity domestically as a hedge against export uncertainty. TSM cashes in on volume either way — it fabricates the leading-edge dies for both Nvidia and AMD's China-bound and domestic-bound parts, collecting foundry economics independent of which flag the finished chip ships under.