NVIDIA expands Vera Rubin AI chip production as global chip race intensifies

NVIDIA has moved its Vera Rubin AI‑chip platform into mass production, with the NVL72 system rolling out to OpenAI and other cloud partners from the third quarter of 2026. The company cites a global manufacturing network of over 350 sites in 30 countries and more than 300 partners, aiming for up to 1 000 racks per day. Early benchmarks show a ten‑fold increase in tokens‑per‑megawatt compared with earlier generations, highlighting the platform’s power‑efficiency focus.

At the same time, Taiwanese contract manufacturer Wistron opened a $650‑$700 million plant in Fort Worth, Texas, to produce NVIDIA’s GB300 Grace Blackwell Ultra super‑chip and future Vera Rubin chips. The facility, visited by NVIDIA CEO Jensen Huang, marks the first U.S. mass‑production site for an NVIDIA‑designed AI chip and is expected to create 500‑1 000 jobs while scaling to tens of thousands of boards per month.

In Shanghai’s World AI Conference, Chinese firms displayed their own AI‑chip prototypes, underscoring a race to narrow the gap with U.S. leaders despite longstanding export restrictions on advanced semiconductor equipment. Analysts note that while Chinese designs lag in performance, the domestic demand surge and government investment keep pressure on the global supply chain.

Together, these developments illustrate a sharpening competition for AI‑compute capacity, with manufacturers expanding on‑shore production, improving energy efficiency, and governments seeking strategic autonomy.