TensorWave, a cloud provider that builds only on AMD hardware, said on Nov. 16 that it has signed a six-year agreement to provide OpenAI with 457 megawatts of IT capacity across multiple U.S. data center sites. In its press release, the company said the capacity will run on AMD Instinct MI455X GPUs and is purpose-built for AMD's Helios rack-scale systems. It will be operated as dedicated, single-tenant capacity for OpenAI.

According to the release, capacity begins coming online in the first half of 2027, with full deployment expected by early 2028. TensorWave called it the largest deployment in its history. The company also said it is the first neocloud publicly announced as an infrastructure provider supporting the previously announced agreement between AMD and OpenAI.

Neither company disclosed financial terms or named the sites, according to a secondary report from World Today News. The press release itself names no sites and gives no financial terms.

What Was Announced

These are the deal's main terms as TensorWave described them and independent outlets repeated them:

  • Term: six years.
  • Size: 457 MW of IT capacity, according to the TensorWave release. Pulse 2 and TokenPost reported the same figure.
  • Locations: multiple U.S. data center sites. The sites were not identified.
  • Hardware: AMD Instinct MI455X GPUs in AMD Helios rack-scale systems. TensorWave said the sites are designed for Helios and use high-performance Ethernet for scale-out networking.
  • Tenancy: dedicated, single-tenant capacity for OpenAI.
  • Schedule: first capacity in the first half of 2027 and full deployment by early 2028.

The 457 MW figure needs careful reading. TokenPost stressed that it is planned IT capacity across multiple U.S. data centers, not capacity that has already been delivered. The release describes IT capacity, which is the power available to computing equipment. It is not the total power the facilities will draw from the grid, which also covers cooling and other building loads. Neither the release nor the independent reports gave a grid-level figure.

Why the Rack Design Matters

TokenPost noted that the deal names AMD hardware for both the GPUs and the rack design. A rack-scale system such as Helios is designed as a unit of many GPUs, so naming it suggests the sites are planned around that design rather than around standalone accelerators; that is an inference, not something the sources state. TensorWave said the sites in this deal were designed around Helios and use Ethernet for scale-out networking, meaning the links that connect racks to each other.

The single-tenant structure is a notable term. TensorWave said the capacity is dedicated to OpenAI alone for the length of the contract.

The Companies' Statements

The press release includes statements from Vamsi Bopanna, senior vice president of AI at AMD, and from Piotr Tomasik, TensorWave's president and co-founder. The release also includes a statement from Uday Ruddarraju, OpenAI's chief technology officer for compute, who said the agreement will help OpenAI continue scaling AI. Tomasik said TensorWave will help deliver "a meaningful portion of the infrastructure behind OpenAI and AMD's next generation of AI compute," without quantifying that share. Bopanna called the deployment "landmark."

Who TensorWave Is

The release describes TensorWave as powered exclusively by AMD Instinct GPUs and as the first public cloud to deploy AMD Instinct MI300X infrastructure. World Today News reported that it later added MI325X and MI355X accelerators. The company says in its release that it has secured more than 2 gigawatts of long-term data center capacity. The release and Pulse 2 also said TensorWave raised a Series B round in June 2026. World Today News put that round at $350 million and and the release lists Magnetar, AMD Ventures and Nexus Venture Partners among its investors.

The 2 GW figure is TensorWave's own claim about capacity it has secured. The sources do not break it down by site or say how much of it is built and running. The sources also do not say how the 457 MW for OpenAI fits within that total.

Where It Fits in the AMD-OpenAI Partnership

TensorWave describes the agreement as part of the existing partnership between OpenAI and AMD and calls itself the first neocloud publicly named as an infrastructure provider under that arrangement. The sources do not say whether other providers will be named or what share of the partnership's total capacity this deal represents.

What Is Still Unknown

Several details that matter to utilities, local officials and nearby residents were not disclosed:

  • Sites: the release says only "multiple U.S. data center sites." Which states, counties or existing campuses are involved has not been made public.
  • Power supply: the sources do not name the utilities, interconnection arrangements or on-site generation that will serve the 457 MW of IT load.
  • Money: financial terms were not disclosed, World Today News reported.
  • Phasing by site: beyond the overall schedule of first capacity in the first half of 2027 and full deployment by early 2028, no site-by-site schedule was given.

What Happens Next

The next milestone on the record is the first block of capacity. TensorWave says deployments will begin coming online in the first half of 2027, with full deployment of the 457 MW expected by early 2028. Neither company has named a public filing, permit or site announcement linked to the agreement.

Sources