NVIDIA 2026-07-25
Industry Signal Impact: Major Conf: 95%

NVIDIA and Wistron Launch US-Based GB300 Production, Shifting AI Manufacturing Landscape

Summary

NVIDIA partnered with Wistron to open a $700M factory in Texas, achieving L6 integration of the GB300 Grace Blackwell Ultra system. The first US-made unit contains 1.5M parts, weighs 2 tons, and costs $4M. This milestone accelerates US AI capex localization from 5% to 30%, reshaping the global AI supply chain.

Key Takeaways

On July 21, 2026, Wistron inaugurated its D1 factory in Fort Worth, Texas, a $700M facility spanning 324,000 sq ft. NVIDIA CEO Jensen Huang signed the first GB300 Grace Blackwell Ultra Superchip produced there, marking the highest L6 integration level from chip to rack. The GB300 features 72 Blackwell Ultra GPUs and 36 Grace ARM CPUs connected via fifth-gen NVLink at 130 TB/s, delivering 1,440 PFLOPS FP4. The upcoming Vera Rubin with Rubin GPU and 88-core Vera CPU will triple performance. This factory is the cornerstone of NVIDIA's US AI manufacturing strategy, aiming to raise US AI capex localization from 5% to 30% by 2028, using 'recursive manufacturing' where NVIDIA AI designs the factory that builds NVIDIA hardware.

Why It Matters

NVIDIA's US manufacturing push is a strategic lock-in play. The L6 integration of GB300 NVL72 with proprietary NVLink and Grace CPU creates a tightly coupled ecosystem, making it hard for hyperscalers to swap in alternative GPUs or interconnects. US assembly may increase costs and face yield issues, while critical components like HBM3e still rely on Asian supply chains. The 'recursive manufacturing' approach further deepens toolchain dependency, raising barriers for competitors. Hidden costs include labor, power, and rapid depreciation with Vera Rubin upgrades.

PRO Decision

【Vendors】AMD and Intel should accelerate US-based manufacturing partnerships for their AI accelerators, emphasizing open standards like OCP and UEC to counter NVIDIA's proprietary lock-in. They must offer modular L6 integration alternatives through similar OEM collaborations. 【Enterprises】CIOs should audit GB300 NVL72 for lock-in risks, exploring PCIe or CXL-based open ecosystems. Contracts must include upgradeability and component interchangeability clauses. Monitor US factory delivery and cost transparency. 【Investors】Recognize that while NVIDIA's move strengthens its moat, it raises capex and operational costs. Track factory utilization, yields, and competitor responses. Over-reliance on US manufacturing may reduce supply chain flexibility; factor in geopolitical risk premiums.

Source: 36氪
View Original →

Get 3-5 key AI infrastructure signals weekly →

💬 Comments (0)