TensorWave Signs 457 MW Data Center Deal With OpenAI
The six-year agreement covers planned U.S. capacity using AMD Instinct MI455X GPUs and AMD Helios rack-scale systems.
Updated on Nov. 16, 2026 in Data Centers

TensorWave has signed a six-year agreement to supply OpenAI with 457 megawatts of planned data-center IT capacity across multiple U.S. sites. The buildout will use AMD Instinct MI455X GPUs and sites designed for AMD Helios rack-scale systems.
Why it matters
The deal adds a large, dedicated block of AI computing capacity in the United States, a key input for training and running advanced AI models. It also shows AMD hardware being specified across both the accelerators and the rack-level system design.
The 457 MW figure refers to IT capacity across multiple U.S. data centers under a six-year term. Specific site locations, contract value and the power and cooling specifications of each facility were not disclosed.
The players
TensorWave
TensorWave is a smaller AI cloud and data-center company that is supplying the dedicated capacity under this agreement.
OpenAI
OpenAI is an artificial intelligence research company that develops large AI models such as the GPT series and is the customer for this capacity.
AMD
AMD, or Advanced Micro Devices, is a semiconductor company that makes the Instinct GPUs and Helios rack-scale systems specified for this buildout.
The details
TensorWave will provide a dedicated deployment for OpenAI using AMD Instinct MI455X GPUs, with the sites designed around AMD Helios rack-scale systems. The buildout is phased, with computing capacity expected to be delivered in stages before full deployment.
Timeline
Computing capacity deliveries are expected to begin in the first half of 2027.
Full deployment is expected to be completed in early 2028.
The Tech Race
The agreement places AMD's Instinct MI455X GPUs and Helios rack-scale systems at the center of a major U.S. AI buildout, competing for share against Nvidia-based deployments that dominate much of today's AI data-center capacity. It reflects the broader push by AI developers to secure dedicated, multi-year compute supply rather than relying only on spot cloud capacity.
For ordinary users, the buildout points toward more dedicated AI computing capacity for the models they use, though the source does not tie the deal to any change in consumer pricing or features. The timeline stretches into 2028, so any effect on day-to-day products would come gradually.
The takeaway
Large AI capacity deals are increasingly tied to specific chip and system choices, so the hardware stack behind AI services is becoming a visible competitive decision. Watching the 2027 delivery milestones will show whether the phased buildout stays on schedule.
Further reading
Explore more on Data Centers coverage for the wider U.S. buildout context.






