AI Capacity Deal Locks In 457 MW for OpenAI

Deliveries on AMD Helios systems are expected to begin in the first half of 2027, with full deployment targeted for early 2028.

Updated on Nov. 16, 2026 in Data Centers

AI Capacity Deal Locks In 457 MW for OpenAI

TensorWave has signed a six-year agreement to supply OpenAI with 457 megawatts of planned U.S. data-center IT capacity, using AMD Instinct MI455X GPUs. The arrangement is announced rather than operational, with deliveries not yet begun.

Why it matters

The deal shows AI capacity being contracted in multi-year blocks sized in megawatts rather than individual servers, tying a neocloud provider's buildout directly to a frontier AI lab's compute needs. It also extends AMD's rack-scale Helios systems into large dedicated deployments for a major AI customer.

The capacity is specified as 457 megawatts of IT load, a measure of power available to computing equipment rather than total facility draw. The accelerators are AMD Instinct MI455X GPUs deployed in sites designed for AMD Helios rack-scale systems, and the source does not report performance benchmarks for this configuration.

The players

TensorWave

A small AI cloud provider that is building dedicated GPU capacity for AI customers, with this agreement as its large-scale OpenAI deal.

OpenAI

The developer of the ChatGPT and GPT model family, which is securing dedicated computing capacity for training and running its models.

AMD

A semiconductor company whose Instinct MI-series accelerators and Helios rack-scale systems compete with Nvidia in AI data-center hardware.

The details

The 457 MW covers IT capacity, meaning the power budget for servers and accelerators rather than the cooling and facility overhead layered on top. TensorWave will house the OpenAI workload in data centers built around AMD's Helios rack-scale design, which packages many GPUs into a single integrated system. The buildout is phased, with initial capacity arriving before full deployment is complete.

Timeline

  1. Initial computing capacity deliveries are expected to begin in the first half of 2027.

  2. Full deployment is expected to be completed in early 2028.

The Tech Race

The deal places AMD's Helios rack-scale design alongside the GPU-cloud capacity contracts that AI labs are signing to secure compute ahead of demand. It shows a second-tier accelerator supplier winning a dedicated, multi-year role in a frontier lab's infrastructure plans.

The capacity will not reach users until deliveries begin in the first half of 2027, and the source does not say which OpenAI products will run on it first. For most people the effect is indirect, showing up later as the compute behind AI services that OpenAI runs.

The takeaway

Watch whether deliveries begin in the first half of 2027 as projected, since that date will show if the phased buildout is holding. The timeline to full deployment in early 2028 is the milestone to track for whether AMD Helios systems scale at this size.

What happens next

Initial computing capacity deliveries are expected to begin in the first half of 2027, the first observable milestone for whether the phased buildout is on schedule.

Further reading

For the broader buildout behind deals like this, see the Data Centers section.