Nvidia restarts China chip builds
Nvidia said it’s restarting manufacturing of a China‑compliant AI chip variant and is reportedly preparing Groq‑compatible chips that can be sold in China — moves aimed at restoring product availability in a constrained market. The notices surfaced alongside broader GTC supply talks and were reported this week, changing procurement timing for multi‑region customers. (reuters.com, reuters.com)
The specific model tied to the manufacturing change is the H200, which CEO Jensen Huang confirmed at GTC on March 17, 2026 after saying Nvidia had received U.S. export licenses and taken orders for shipments to China. (cnbc.com) Financial Times reporting in early March said Nvidia had previously halted China‑bound H200 output and shifted TSMC capacity toward its new Vera Rubin family, a reallocation that influenced short‑term availability. (ts2.tech) Two Reuters sources told reporters Nvidia is developing a Groq‑derived inference accelerator variant intended for sale in China, and those sources said the China‑compatible Groq product is expected to be available in May 2026. (msn.com) Nvidia licensed Groq technology late last year in a deal Reuters characterized at about $17 billion, and the company showcased Groq‑based products and a Groq 3 lineup at its March GTC alongside the Vera Rubin platform announcement. (money.usnews.com) Company comments at GTC also noted production had been “fired up” in recent weeks and that U.S. authorizations would allow only limited initial shipments into China, changing timing for multi‑region procurement cycles. (finance.yahoo.com) At GTC Nvidia projected the addressable market for AI infrastructure could reach at least $1 trillion through 2027, a forecast the company used to justify accelerating inference‑focused products and reclaiming capacity for China orders. (businessinsider.com)
