Sources: Nvidia ordered 300K H20 chips from TSMC last week, adding to its 600K to 700K inventory, amid strong China demand after the US cleared H20 chip sales
Nvidia (NVDA.O) placed orders for 300,000 H20 chipsets with contract manufacturer TSMC (2330.TW) last week, two sources said …
Context & Ripple Effects
Chinese cloud and internet groups had already been increasing H20 orders as demand for DeepSeek models grew, according to earlier reporting on the order ramp. This order turns that customer pull into a sizable manufacturing commitment and expands Nvidia's available China-oriented supply.
The coverage also frames H20 as exposed to policy and customer-acceptance risk: Nvidia had challenged tighter export rules, while later reporting described a supplier production halt after Chinese guidance. That makes the inventory build consequential beyond a routine foundry order.
First-order effects
- Nvidia increases its H20 inventory from roughly 600,000–700,000 units by ordering another 300,000 chipsets from TSMC, giving it more supply to meet demand following U.S. clearance of sales.
- TSMC receives additional H20 production demand, while Nvidia's China customers gain a larger prospective pool of the compliant accelerator.
Second-order effects
- The order reinforces demand for foundry capacity and the surrounding AI-chip supply chain; Nvidia must convert inventory into permitted, accepted shipments to avoid carrying excess China-specific stock.
- Chinese buyers that had been ramping H20 orders can plan more deployments around Nvidia hardware, while domestic alternatives face a more immediate availability benchmark.
Third-order effects
- The episode illustrates how export-policy decisions can rapidly redirect AI-chip capacity toward or away from a single market, making product planning and inventory management more geopolitical.
- If this pattern persists, China-focused AI compute will remain a distinct, policy-constrained supply segment rather than simply a share of Nvidia's global accelerator market.
The trend: AI-compute supply is increasingly being allocated through a combination of foundry capacity, export permissions, and country-specific customer demand.