Nvidia ships first H200 chips to China, but expects no data-centre revenue from them

Nvidia began shipping its first H200 AI processors to China under a new US licensing scheme, but they accounted for less than 1% of its $89B quarterly data center revenue. For the current quarter, Nvidia’s $108B forecast assumes $0 data center computing revenue from China. The update suggests limited near-term relief from China restrictions despite the first permitted shipments.

Analysis

The economic signal is not the shipment itself; it is that management is implicitly treating China as an option, not a driver. That matters because the current quarter already assumes zero, so any headline-driven selloff is likely to be about narrative risk rather than earnings math. In the near term, this reduces the odds of a true estimate reset, but it also reinforces a multiple ceiling: investors will keep discounting the possibility that export rules can tighten again with little warning.

Second-order, the more important consequence is competitive and strategic. Limited H200 access keeps Chinese AI builders dependent on legacy Nvidia software and ecosystem standards, which delays domestic substitution, but it also raises the probability that buyers continue hedging into local accelerators and non-U.S. supply chains over 6-18 months. That is a slow-burn share risk, not a quarter-to-quarter revenue issue, and it is more relevant for long-duration holders than for traders.

The contrarian read is that the market may overreact to a tiny revenue stream being re-opened under a fragile license regime. If the stock sells off hard, the right question is whether core data-center demand or gross margin expectations changed; if not, this is mostly policy noise. The real tail risk is a broader export-control expansion to higher-end memory, networking, or manufacturing tools, which would hit the entire AI complex, not just NVDA, and would show up first in guidance commentary and supply-chain lead times.

Key Decisions for Investors

• Fade any 2-4% headline-driven dip in NVDA over the next 1-3 sessions; add via common or 1-2 month call spreads rather than chasing strength. Risk/reward is favorable because China is effectively a zeroed-out line item in guidance, so the selloff would be mostly sentiment. Falsify this trade if next data-center guidance is revised lower or gross margin steps down.
• Run a relative-value long NVDA / short SOXX or SMH for 1-3 months if export-control headlines keep pressuring semis broadly. NVDA should be less exposed than the basket to incremental China restrictions, so a policy scare should compress the group more than the leader. Exit if the U.S. expands controls to additional accelerator classes or if NVDA’s core AI growth slows.
• Do not short NVDA outright on this news alone; instead, if you want policy downside exposure, prefer a basket short against the AI complex rather than a single-name bet. The downside case is a broader licensing rollback, which is a low-probability but high-impact event over weeks to months. Falsify if the licensing regime proves durable across multiple shipments and management reiterates unchanged full-year DC growth.
• Set an alert for the next earnings call and any quantified China commentary. If management moves China from an explicit zero assumption to even low-single-digit revenue contribution, that is a modest positive for estimates and could lift the stock 3-5% as the market adds back option value.