Kivorent News

Wednesday, November 18, 2026

Tech and AINews

TensorWave signs six-year deal to give OpenAI 457 MW of AMD-powered capacity

TensorWave, an AMD-backed cloud provider, will build 457 megawatts of dedicated AMD Instinct MI455X capacity for OpenAI across several US sites, starting in the first half of 2027.

Kiln, covering chips, data centers and power
, 2 min read

457 MWIT capacity TensorWave will run for OpenAI

What to know

  • TensorWave announced a six-year agreement on November 16 to provide OpenAI with 457 megawatts of IT capacity built on AMD Instinct MI455X GPUs.
  • Capacity starts coming online in the first half of 2027, with full deployment expected by early 2028, across multiple US data centers that were not named.
  • TensorWave says it is the first neocloud publicly named as a provider under OpenAI's existing partnership with AMD.

TensorWave, a cloud provider that rents out AMD graphics processors, said on November 16 that it has signed a six-year agreement to supply OpenAI with 457 megawatts of IT capacity. The capacity will run AMD Instinct MI455X GPUs in AMD's Helios rack-scale systems and will be dedicated, single-tenant capacity for OpenAI, according to TensorWave's announcement.

The company said the first capacity comes online in the first half of 2027, with full deployment expected by early 2028. TensorWave called it its largest deployment to date. Neither company disclosed what OpenAI will pay.

Where and how

The capacity will sit in multiple US data centers, which TensorWave did not name. It said the sites are being designed specifically for Helios racks, with high-speed Ethernet networking linking the racks together. TokenPost, which reported the deal the same day, also described the buildout as phased across several US sites.

The headline of the Business Wire release says "more than 450MW"; the body of the release and TensorWave's own blog give the figure as 457 MW of IT capacity, which is the number used here. IT capacity is the power that reaches the computers themselves; a site's total grid connection is larger once cooling and other overhead are added. Our explainer on data center power and water covers why.

Part of OpenAI's AMD partnership

TensorWave framed the deal as part of OpenAI and AMD's previously announced strategic partnership, and said it is the first neocloud publicly named as an infrastructure provider supporting that agreement. A neocloud is a newer cloud company that specializes in renting out AI chips rather than offering a full range of cloud services.

OpenAI's chief technology officer for compute, Uday Ruddarraju, said in the release that the agreement advances "our previously announced partnership with AMD." AMD senior vice president Vamsi Bopanna called TensorWave a strong partner in scaling AMD Instinct systems.

TensorWave is backed by AMD Ventures, Magnetar and Nexus Venture Partners and raised a Series B round in June 2026, according to the release. It says it has secured more than 2 gigawatts of long-term data center capacity and was the first public cloud to deploy AMD's earlier MI300X chips.

Context

The deal adds to a busy month for AMD in data centers. AMD made its 256-core Venice EPYC server chips generally available on November 10 and has pitched Mext software as a way to cut server memory needs during the DRAM shortage. OpenAI, for its part, is also building its own campuses, including the contested Project Camellia site in Georgia, and TSMC's board recently approved stakes of up to $500 million each in Anthropic and OpenAI.

Why it matters

The deal puts a size, a timeline and a named operator on part of OpenAI's AMD partnership: 457 MW of AMD's newest GPUs reserved for OpenAI alone for six years, delivered through a third-party cloud rather than by AMD directly. It shows OpenAI spreading its compute across chip suppliers and cloud providers, and it gives AMD's Helios racks one of their first large, publicly named deployments.

Sources

AI chips and data centersOpenAI

Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.