OpenAI Signs US$38bn AWS Deal to Scale AI Data Centre Power

OpenAI secures a seven-year, US$38bn deal with Amazon Web Services (AWS) to access one of the world’s largest pools of cloud-based compute power, marking a decisive move toward multi-cloud infrastructure and reshaping the way frontier artificial intelligence (AI) workloads are hosted and scaled.

This strategic partnership gives OpenAI, the developer behind ChatGPT, full access to Amazon EC2 UltraClusters, purpose-built infrastructure powered by hundreds of thousands of Nvidia GPUs. The scale of this agreement positions AWS at the centre of OpenAI’s AI workloads, ranging from training next-generation models to running inference at low latency.

As data centre operators prepare for increasing pressure from AI-intensive compute demands, this deal reflects a trend toward diversified and redundant infrastructure, breaking away from the one-provider dominance that has characterised much of AI’s growth to date.

Sam Altman, Co-founder and Chief Executive Officer at OpenAI, says: “Scaling frontier AI requires massive, reliable compute. Our partnership with AWS strengthens the broad compute ecosystem that will power this next era and bring advanced AI to everyone.”

AWS brings hyperscale GPU infrastructure to OpenAI

The partnership unlocks immediate access for OpenAI to AWS clusters that support workloads using tens of millions of CPUs and hundreds of thousands of GPUs. AWS already operates infrastructure clusters with over 500,000 chips, optimised for high-throughput AI operations.

AWS infrastructure will now serve as a key backbone for OpenAI’s model training and deployment. These data centre clusters are designed to support high-volume inference, sustained training, and edge-to-core performance tuning. With OpenAI’s models scaling in size and complexity, the infrastructure must deliver consistent, low-latency throughput across globally distributed systems.

This deal also gives AWS a considerable advantage over other cloud rivals. Anthropic, another AI firm supported by AWS, has an US$8bn partnership with the platform, but OpenAI’s US$38bn agreement places it in a new league of compute-intensive clients.

Matt Garman, Chief Executive Officer at AWS, says: “As OpenAI continues to push the boundaries of what's possible, AWS’s best-in-class infrastructure will serve as a backbone for their AI ambitions. The breadth and immediate availability of optimised compute demonstrates why AWS is uniquely positioned to support OpenAI’s vast AI workloads.”

For AWS and data centre stakeholders, the implications are substantial. Massive clusters, high-performance cooling and energy management systems must keep pace with workloads that push every part of the data stack. This growth also raises operational questions around power density, chip interconnects and availability of specialised hardware such as Nvidia H100 and GB200 GPUs.

Multi-cloud strategy reshapes AI infrastructure

OpenAI’s deal with AWS signals a wider shift across the AI sector toward infrastructure diversification. The company had already entered long-term commitments with Microsoft Azure, with an incremental US$250bn. Azure remains a key supplier, but following OpenAI’s internal restructuring into a public-benefit corporation (PBC), the firm is no longer exclusively bound to Microsoft for cloud provision.

The AWS deal adds further flexibility. It allows OpenAI to deploy workloads across multiple vendors to avoid bottlenecks and maximise availability, particularly during peak compute cycles. The AI company has also signed infrastructure contracts with CoreWeave, a specialist provider offering Nvidia GPU-accelerated instances for model training. That agreement is reportedly worth US$22.4bn in total contract value.

This multi-supplier strategy serves multiple functions, distributing risk, securing availability and accelerating deployment timelines. For data centre operators, it signals a growing opportunity to cater to AI companies seeking scalable, high-density environments that can be brought online rapidly and scaled horizontally across geographies.

The move also reflects the increasing need for redundancy in the supply of critical infrastructure, especially for organisations whose workloads are compute-bound and increasingly tied to training performance at scale.

Data centres central to AI’s next scaling phase

For OpenAI, the AWS partnership is not just about volume, it is about readiness. The ability to train larger and more capable models at speed is now viewed as the primary competitive advantage in the AI industry. Infrastructure access is no longer a back-office decision but a front-line determinant of product capability.

The deal also deepens OpenAI’s presence in AWS’s service ecosystem. The company’s open-weight models are already available through Amazon Bedrock, enabling thousands of customers to deploy them across sectors like scientific research, software engineering and data modelling.

In this context, the AWS agreement represents more than just an expansion—it is a structural evolution in how AI platforms are built and delivered. From a data centre perspective, it validates the continued investment in AI-optimised infrastructure: from chip density and energy efficiency to high-throughput networking and geographically distributed redundancy.

The scale of OpenAI’s compute requirements ensures that hyperscale cloud providers must continue investing in GPU-accelerated clusters and supporting technologies. In return, AI companies like OpenAI gain the flexibility to scale, experiment and deploy globally.

With infrastructure diversification accelerating and AI workloads growing exponentially, the OpenAI-AWS agreement marks a shift from reliance on a single platform to a networked, multi-cloud architecture designed for the next generation of AI models.

In the race for AI dominance, control of the data centre stack, from chip to cooling, becomes the decisive edge.