Who Cashes In
AVGO is the clearest winner the rule did not intend to create. Broadcom designs custom AI ASICs — Google's TPUs, Microsoft's Maia, Meta's MTIA — for hyperscalers that build data centers inside Tier 1 countries. Those chips are not sold into restricted markets; they are consumed domestically by the cloud giants themselves. The export control apparatus does not touch them. Broadcom reported roughly $20 billion in AI chip revenue in fiscal 2025 and is targeting $100 billion in addressable AI ASIC revenue by 2027. No compliance overhead, no export variant discount, no $4.5 billion charge.
TSM holds the manufacturing chokepoint. TSMC fabricates approximately 90 percent of the world's most advanced chips at 3nm and below. Because Taiwan sits in Tier 1 under the diffusion rule, TSMC's Arizona and Taiwan fabs are the only facilities on earth legally capable of producing frontier AI silicon for export to Tier 1 customers without restriction. Every approved chip — NVDA, AVGO, AMD — runs through TSMC's lines.
NVDA itself is not a loser, but its shape changes. Nvidia took a $4.5 billion charge in 2025 tied to H20 inventory and export obligation write-downs after the chip was abruptly relicensed. Its full-performance Blackwell products flow freely to Tier 1 buyers; the export-capped H20 and B20 variants serve Tier 2 at lower margins and average selling prices. Tier 1 hyperscaler demand remains enormous, and Nvidia's Tier 1 revenue is growing faster than the export-variant business shrinks.
AMD benefits modestly in the same Tier 1 channel. Its MI300X series faces no ceiling with U.S. cloud customers, and its position as the credible GPU alternative in unrestricted markets gives it room to pick up workloads where Nvidia supply is constrained.
Who Is Exposed
NVDA carries the most direct regulatory drag of any name in this stack. Each new policy revision — the H20 license requirement in April 2025, modifications in December 2025 — generates write-down risk and earnings unpredictability on its Tier 2 and Tier 3 variant business. The company is structurally dependent on BIS goodwill.
INTC has the worst of both worlds. Its Gaudi AI accelerators sit in the same restricted export class as Nvidia's data center GPUs but without Nvidia's dominant Tier 1 revenue base to absorb the hit. Intel lacks a hyperscaler ASIC partnership comparable to Broadcom's, meaning it cannot route around the restrictions the same way.
SMCI assembles AI servers for export and is heavily exposed to Tier 2 and Tier 3 order flow. When licensing tightens or country allocations shift, Super Micro's system shipments stall before the chip companies even feel the impact.
What to Watch
The BIS country-tier caps reset in 2027. Watch for the renegotiation — any country upgraded from Tier 2 to Tier 1 is an immediate unlock for unrestricted Nvidia and AMD GPU sales. Watch Broadcom's ASIC pipeline announcements: each new hyperscaler design win is, structurally, a bet placed outside the export control system entirely. And watch TSMC's Arizona fab ramp — the moment advanced-node U.S. domestic production hits commercial scale, the geopolitical risk embedded in the Taiwan manufacturing concentration begins to price differently.