OpenAI struck a $38 billion cloud deal with Amazon’s AWS, sending Amazon shares sharply higher as investors cheered a marquee AI customer choosing its infrastructure. The multiyear pact underscores how voracious demand for computing power is reshaping alliances across Big Tech.
The companies announced the agreement on November 3, 2025. OpenAI will begin running core AI workloads on AWS immediately, with all planned capacity targeted to be in place by the end of 2026 and room to expand in 2027 and beyond. Under the seven-year arrangement, OpenAI will tap hundreds of thousands of Nvidia GPUs on AWS, including next‑generation accelerators, to train models and serve products such as ChatGPT. On the news, Amazon shares rose about 5% to an all‑time high, highlighting investor confidence in AWS’s competitiveness against Microsoft and Google.
The deal follows OpenAI’s recent restructuring and updated terms with Microsoft, which ended the software giant’s exclusive cloud rights and gave the ChatGPT maker more latitude to diversify suppliers. It also adds a high‑profile customer win for Amazon even as it supports rival AI startup Anthropic on AWS. Beyond bragging rights, the pact speaks to the economics of frontier AI: companies are racing to assemble massive, reliable compute at scale, and cloud providers are vying to supply it with cutting‑edge chips and tightly networked data centers.
OpenAI’s workloads are moving onto AWS now, with the bulk of capacity scheduled over the next year. All planned capacity is targeted by end of 2026, after which the partnership allows for further expansion. Watch for phased cluster buildouts and performance milestones as AWS deploys new Nvidia hardware and networking to reduce latency for training and inference. For Amazon, the win could translate into steadier high‑margin cloud revenue and more customer interest in related services such as Amazon Bedrock, where OpenAI’s models have been appearing. For the broader market, the agreement reinforces a simple reality: the AI race will be won, in part, by whoever can secure the most capable and cost‑efficient compute—reliably, and at unprecedented scale.