TensorWave has signed a six-year agreement to provide OpenAI with 457 megawatts of dedicated AI infrastructure, running AMD Instinct MI455X accelerators, across multiple US data-centre sites. The company said on 16 November that it expects the first capacity to come online in the first half of 2027 and the full deployment to be complete by early 2028. A Business Wire release the same day carries the same figures and quotes from AMD and OpenAI. No contract value was disclosed.
The 457 megawatts, and the 2027–2028 timetable, are planned capacity. They are not a report of racks already in service for OpenAI.
TensorWave will install AMD MI455X capacity for OpenAI on a six-year term
TensorWave said the infrastructure will be “purpose-built for AMD Helios rack-scale systems” and “operated as dedicated, single-tenant capacity for OpenAI,” in the wording of the Business Wire release. The company blog puts it the same way: dedicated, single-tenant capacity, with Helios racks linked by high-performance Ethernet for large-scale training and inference. Delivering that capacity, it said, “requires careful coordination across GPU systems, networking, power, cooling, and operations,” on a project “spanning hundreds of megawatts and multiple data center sites.” The release does not name those sites. TensorWave described the agreement as its largest deployment to date, and as making it “the first publicly announced neocloud infrastructure provider supporting” OpenAI and AMD’s previously announced partnership.
Vamsi Boppana, AMD’s senior vice-president for AI, said: “TensorWave has been a strong partner in scaling AMD Instinct infrastructure, and we are excited to expand our work together with this landmark deployment for OpenAI.” The Business Wire release also quoted Uday Ruddarraju, OpenAI’s chief technology officer for compute: “We’re excited to advance our previously announced partnership with AMD through this agreement with TensorWave, which will help us continue scaling AI to serve more people and organizations around the world.”
The company also set out its own history as the reason it can take the work on. TensorWave said it “was the first public cloud to deploy AMD Instinct MI300X GPUs and has continued expanding its infrastructure across successive generations of AMD hardware, including MI325X and MI355X.” It is an AMD Helios launch partner, it said, and “operates one of the world’s largest AMD-exclusive AI clouds and one of North America’s largest AMD-based AI training clusters.” Those are company claims about its existing fleet, not measurements for the new OpenAI deployment.
Piotr Tomasik, TensorWave’s president and co-founder, said: “AI infrastructure is entering a new phase, where the ability to deliver hundreds of megawatts of advanced compute reliably and at speed will separate the companies that can support frontier AI from those that cannot.” He presented the OpenAI work as taking that effort “into an entirely new level.”
TensorWave said it has “secured more than 2 gigawatts of long-term data center capacity.” That is a floor on sites it says it already has a right to, not a statement that two gigawatts are built, and not a price for the OpenAI slice. The 457 megawatts of IT capacity, in the Business Wire wording, is the only capacity figure given for this agreement.
The deal sits in a wider AMD compute build-out that TSN has already been covering. AMD has a credit-support arrangement for phase one of Fermi’s Project Matador in Texas, with a framework of up to 650 MW, and its latest quarter showed data-centre sales nearly doubling. Those posts are about other contracts and about AMD’s reported results. They are not this OpenAI agreement. World Labs, which AMD has now said it has acquired, is a model lab, not a data-centre operator.
A six-year capacity agreement is not output already online
The firm facts are the parties, the 457-megawatt figure, the six-year term, the MI455X and Helios description, the US sites, and the stated timetable of first capacity in the first half of 2027 and full deployment by early 2028. The official pages do not name the sites, do not give a contract value, and do not say that any of the 457 megawatts is in service today. “More than 2 gigawatts” is TensorWave’s claim about capacity it has secured, not a measured output figure for this deal.
That gap is the story as much as the headline number. A six-year, hundreds-of-megawatts agreement can be signed well before power is delivered, and a “largest deployment to date” can still be a plan. Readers should not convert 457 megawatts into a chip count, a revenue figure or a share of OpenAI’s wider AMD partnership. None of those conversions is in the release.
The next official facts to watch are the ordinary ones: whether TensorWave or OpenAI later names the sites, states a value, or reports that the first capacity has actually come online in the first half of 2027. Until then, this is a signed capacity agreement with no price attached, not a completed build.
Sources
- TensorWave, “TensorWave to Provide OpenAI with 457 Megawatts of AMD-Powered AI Infrastructure,” 16 November 2026
- Business Wire, “TensorWave Announces Agreement with OpenAI to Deliver More Than 450MW of AMD Powered AI Infrastructure,” 16 November 2026
Get TSN’s biggest tech stories, explained.
News and insights across AI, tech, crypto and cybersecurity. Every source linked. Unsubscribe anytime.
Get the newsletterRelated stories
- AMD Signs Credit Support Deal for Phase One of Fermi’s Project Matador in Texas, With a Framework for Up to 650 MW is the nearest TSN account of another large AMD-linked capacity plan, and a useful comparison for a megawatt figure that is still a timetable rather than output.
- AMD Q3 Results: Record $13.4bn as Data Center Sales Nearly Double
Follow @tsnmediahub on X for breaking AI news as it happens.
Follow on X



