Nvidia Loses $170B as China Blocks H200 — Chip Wreck Live

Earlier this week we wrote about the H200 export greenlight as a Dow-50,000-making event — see Nvidia H200 Hits China — Dow Just Cracked 50,000 First Time. Here is what just happened: the deal that lifted the index by record amounts on May 14 was undone in 48 hours by a single Beijing decision. Same chips, same buyers, same Commerce paperwork — zero shipments, $170B gone.

Trump confirmed the refusal aboard Air Force One on the return flight, saying China had “chosen not to” act on the import permissions. The tone was unusually subdued for a Trump readout — no “great meeting” superlatives, no promises of imminent follow-up, no framework language. Jensen Huang, who had been added to the delegation at the last minute, left Beijing with nothing.

By the close in New York the damage was binary across the chip complex: NVDA -4.4%, INTC -6%, MU -6.6%, AMD -5.7%, and Cerebras (which had popped 68% on its Thursday IPO into the summit optimism) -10%. The SMH and SOX both rolled below their 21-day moving averages. The AI rally that was printing record indices 24 hours earlier is now on the defensive.

The single most underappreciated number coming out of this summit is the collapse of Nvidia's share of the China AI chip market from ~95% to ~8%. Domestic Chinese chips — primarily Huawei's Ascend line, with supporting roles from Cambricon and Biren — captured over 55% of Q1 2026 domestic AI accelerator demand. That is not a marginal substitution. That is structural.

Once a market hits that kind of domestic share, the political logic of reopening to a foreign supplier flips. Beijing no longer needs Nvidia to keep its hyperscalers competitive on training compute. It needs to give Huawei the demand runway to harden its roadmap, hit volume yields on its 3nm-equivalent process node, and lock in the next two generations of Chinese data center capex. Letting H200s in now would slow that flywheel.

There is also the negotiating-leverage layer. Beijing watched the US push through rare-earth controls, semiconductor equipment restrictions on ASML tooling, and tariff layers on EVs and solar — and it walked into this summit knowing that refusing the H200 is the single highest-cost optical move it can deny Trump. The chip refusal is leverage being saved for the next round.

Heading in: consensus FY26 revenue had baked in a meaningful re-opening of China contribution via the H200 license. That assumption is now wrong. The buy-side is in the process of stripping China revenue from FY26 models through the weekend and into Monday-Tuesday positioning.

What to watch in the print:

• China revenue line: the explicit number, and any qualitative language on FY26 China contribution. Soft = gap-down extension.
• Rest-of-world data center growth: if ex-China data center is still printing 40%+ year-on-year, Huang can re-frame the China loss as “manageable”. Below 30% YoY and the bear case extends.
• FY26 capex commentary from hyperscaler customers: Microsoft, Google, Meta, Amazon. Any qualitative pull-back signal there is gasoline on the fire.
• Huang call tone: the difference between “China is a temporary headwind” and “China is a structural reset” is worth 5-7% on the stock.

Implied move on the May 23 weekly options is wider than the trailing-4 earnings reactions. That tells you the option market is pricing for asymmetric downside. Respect that pricing.