Overview
- OpenAI has lined up more than 30 gigawatts of capacity across Nvidia, AMD, Broadcom and Oracle, with deployments beginning in the second half of 2026 and ramping through 2029.
- Broadcom’s newly formalized agreement calls for co‑developing and supplying 10 gigawatts of custom AI accelerators with Ethernet networking under a cash‑based contract, with deliveries starting in H2 2026.
- AMD secured roughly 6 gigawatts of data‑center GPUs for OpenAI and landed a major Oracle order, underscoring OpenAI’s push to diversify suppliers and drive cost competition.
- Chip stocks rallied on the supply announcements but fell after China tightened customs checks on advanced U.S. semiconductors and Micron moved to exit China’s data‑center market, spurring further supply‑chain shifts out of China.
- Nvidia remains the dominant provider, with Microsoft Azure deploying production‑scale GB300 NVL72 systems and HSBC upgrading the stock on expectations for a larger AI‑GPU market.