The AI-fueled rally in semiconductors just ran into a geopolitical buzzsaw. NVIDIA Corp NVDA and Advanced Micro Devices Inc AMD disclosed this week that the U.S. government has imposed new license requirements for selling their AI-focused chips — Nvidia’s H20 and AMD's MI308 — to China, Hong Kong, Macau and several D:5 countries.

Nvidia responded with a $5.5 billion inventory charge, effectively bracing for a complete block in shipments. AMD booked an $800 million hit of its own. JPMorgan analyst Harlan Sur estimates that these curbs will shave off 8%-10% of both companies’ earnings this year.

The China Revenue Stream? Drying Up Fast

Nvidia's China data center business previously contributed a low-teens percentage to its overall data center revenue. With H20 shipments now in limbo, Sur expects that number to fall to "a de minimis level."

AMD, meanwhile, faces an estimated $1.5–$1.8 billion revenue impact on the back of the $800 million inventory charge. That's roughly 10% of its expected data center sales this year, noted Sur. The fallout isn't confined to the second quarter – the guidance hit spans all of 2025.

Winners, Losers — And The Silicon Side Effects

Sur believes there's still heavy AI demand in China — but now Chinese firms may accelerate a pivot to domestic alternatives like Huawei's Ascend chips or custom ASICs. In a surprising silver lining, custom AI ASIC projects remain unaffected by the new rules.

That's good news for players like Broadcom Inc AVGO, which continues to ramp up shipments for its third major custom AI customer. As it stands, merchant solutions like Nvidia and AMD's GPUs are feeling the heat, while bespoke chips enjoy relative regulatory immunity — for now.

With new AI diffusion rules starting mid-May and a looming Section 232 probe on imported chips, this may be just the opening chapter of a longer regulatory saga for U.S. semiconductor giants.

Read Next: