OpenAI、5000億ドル規模のStargate Projectを発表 Announcing The Stargate Project
- OpenAIはOracle、SoftBank、MGXと共同で、米国内のAIインフラ構築に4年間で5000億ドルを投じる新合弁会社「Stargate」を発表した。
- 即時に1000億ドルを投入し、テキサス州で建設を開始。
- AGI実現に向けた計算基盤の確保と数十万人規模の雇用創出を目指す。
OpenAIは2025年1月、SoftBank、Oracle、アブダビの投資会社MGXと共同で新会社「The Stargate Project」を設立し、今後4年間で米国内のAIインフラに最大5000億ドルを投じると発表した。うち1000億ドルは即時に投入される。SoftBankが資金面、OpenAIが運用面の責任を主導し、SoftBankの孫正義氏が会長に就く。
初期パートナーとしてはMicrosoft、NVIDIA、Arm、Oracleが技術提供で参画する。建設はすでにテキサス州アビリーンで始まっており、米国全土の候補地でキャンパス展開を検討中とされる。OpenAIはこれにより数十万人の雇用が生まれ、AGI(汎用人工知能)実現に向けた計算資源の確保と、米国および同盟国の戦略的優位の維持につながると説明している。
背景には、フロンティアモデルの学習・推論に要する計算需要の急膨張がある。GPT-4以降のスケーリングでは、モデル規模だけでなく、推論側のチェーン・オブ・ソート計算(o1系のような推論時計算)も電力・GPUを大量に消費する傾向が強まっている。Microsoft Azureとの既存契約は維持されつつ、Stargateで新規容量を上積みする構図と見られる。
OpenAIはOracle、SoftBank、MGXと共同で、米国内のAIインフラ構築に4年間で5000億ドルを投じる新合弁会社「Stargate」を発表した。
業界ではAmazonがAnthropicへ巨額出資しTrainiumクラスタを構築、GoogleもTPU v5pで自社モデルを支える等、ハイパースケーラー間のインフラ軍拡が加速している。Stargateはこの潮流の中でも突出した規模で、米政府による国家安全保障的な後押しも示唆されており、半導体輸出規制やエネルギー供給(原子力・SMR含む)を巡る議論と連動して進む可能性がある。
OpenAI announced in January 2025 the formation of The Stargate Project, a new joint venture with SoftBank, Oracle and Abu Dhabi-based investor MGX that intends to invest up to $500 billion over the next four years to build AI infrastructure in the United States. An initial $100 billion is being deployed immediately. SoftBank will lead financial responsibilities while OpenAI takes operational responsibility, with SoftBank chairman Masayoshi Son serving as chairman of the venture.
Microsoft, NVIDIA, Arm and Oracle are named as key initial technology partners. Construction is already underway at a campus in Abilene, Texas, with additional sites across the country reportedly under evaluation. OpenAI frames the project as a vehicle to create hundreds of thousands of American jobs, secure the compute capacity needed to pursue AGI, and reinforce US and allied strategic leadership in AI.
The announcement reflects a sharp escalation in the compute demands of frontier model development. Beyond pretraining ever-larger models, the rise of inference-time reasoning systems such as OpenAI's o1 family has shifted significant load to serving infrastructure, where extended chains of thought consume far more GPU-seconds per query than conventional chat completions. Stargate appears designed to layer net-new capacity on top of OpenAI's existing relationship with Microsoft Azure rather than replace it, although the precise division of workloads has not been disclosed.
The move comes amid an intensifying infrastructure arms race among hyperscalers and model labs. Amazon has committed billions to Anthropic and is scaling Trainium-based clusters, Google continues to expand TPU v5p capacity for Gemini, Meta is building out hundreds of thousands of H100-class GPUs, and xAI brought its Memphis Colossus cluster online in record time. Even within that context, a half-trillion-dollar headline figure is unprecedented, and execution will hinge on power availability, grid interconnects and chip supply rather than capital alone.
The political backdrop is also notable. The announcement was made alongside the incoming US administration, signaling a posture in which large-scale AI buildout is treated as a matter of national competitiveness vis-à-vis China. That framing could dovetail with ongoing debates over semiconductor export controls, domestic fab incentives under the CHIPS Act, and novel energy procurement strategies — including the small modular reactor and behind-the-meter gas deals that several hyperscalers have begun to explore. How quickly Stargate can translate pledged capital into energized megawatts and racked GPUs will likely determine whether the project reshapes the competitive landscape or simply matches it.
本ページの本文・要約は AI による自動生成です。正確性は元記事 (openai.com) をご確認ください。
