Oracle与OpenAI已取消德克萨斯州Abilene Stargate旗舰园区的扩建计划，原定容量将从1.2GW增至2GW的目标落空。受阻于融资挑战、OpenAI需求预测频繁变动及冬季天气导致的液冷设备故障引发的关系紧张。此外，现场电力就绪需一年，届时OpenAI拟部署新一代Nvidia Rubin芯片而非Blackwell，故转向其他选址。与此同时，Nvidia为防竞品流入，已向开发商Crusoe支付1.5亿美元定金，正协助其将剩余产能租赁给Meta，但协议尚未签署。
Oracle and OpenAI have cancelled plans to expand their flagship Abilene Stargate campus in Texas, scrapping targets to grow capacity from 1.2GW to 2GW. The move follows financing challenges, shifting demand forecasts from OpenAI, and strained relations due to winter weather damaging liquid cooling equipment. With power availability delayed by a year, OpenAI prefers deploying next-gen Nvidia Rubin chips elsewhere rather than Blackwell at this site. Meanwhile, Nvidia has paid Crusoe a $150m deposit to prevent rival chip usage and is facilitating a deal to lease remaining capacity to Meta, though no agreement is finalized.
本文由智算芯能前沿网编辑整理，转载请注明出处。