The first slice of six gigawatts
OpenAI has locked in 457 megawatts of dedicated AI computing from TensorWave, an AMD-only cloud provider, under a six-year agreement announced on 16 November. The capacity will run on AMD Instinct MI455X graphics processors, built into AMD’s Helios rack systems at data centres across the United States.
The deal is the first publicly announced neocloud agreement inside the strategic partnership AMD and OpenAI struck in October 2025, covering six gigawatts of AMD graphics processors. TensorWave calls it its largest deployment to date, a step change from its existing clusters to hundreds of megawatts.
Capacity will arrive in phases. The first systems are expected to come online in the first half of 2027, with the full 457 megawatts deployed by early 2028. TensorWave will operate the sites as dedicated, single-tenant capacity, meaning the hardware serves OpenAI alone rather than shared cloud customers.
Helios at the centre
AMD is putting its newest hardware at the centre of the build. The Helios rack systems combine MI455X processors with AMD’s latest server chips, networking and software into one integrated design, which AMD says can deliver up to 2.9 exaflops of peak FP4 performance in a single rack.
All three companies put executives behind the announcement. AMD’s Vamsi Bopanna called it a landmark deployment for OpenAI, TensorWave president Piotr Tomasik said the deal takes his company’s work to an entirely new level, and OpenAI’s Uday Ruddarraju said expanding large-scale compute is critical to building more capable models.
A small provider, a large role
TensorWave is a small company playing a large role. Backed by AMD Ventures, it runs one of the world’s largest all-AMD graphics clouds, was the first public cloud to deploy AMD’s MI300X chips, and has now secured more than two gigawatts of long-term data-centre capacity for future deployments.
The first-neocloud claim is the point. Until now, the AMD–OpenAI partnership has been described in gigawatts and roadmaps, with no named provider building the capacity. TensorWave’s announcement puts a name, a scale and a timetable on the first slice, and invites rivals to show their own.
The company raised $350 million in a Series B round in June 2026 to fund its expansion, and the OpenAI agreement significantly enlarges its footprint. It is an AMD Helios launch partner, and the announcement stresses that the infrastructure is purpose-built for Helios, with high-performance Ethernet networking across the racks.
Why It Matters
The deal matters because it shows OpenAI’s compute supply diversifying beyond Nvidia. AMD’s six-gigawatt partnership with OpenAI dates to October 2025, and this is the first concrete neocloud slice of it to go public. If TensorWave delivers on schedule, OpenAI’s future training and inference will partly run on AMD silicon.
It also tests whether the neocloud model can scale. Specialist providers like TensorWave and CoreWeave have won business by moving faster than the big clouds on scarce accelerators. A 457-megawatt, single-tenant commitment is a far bigger bet than a pilot, and its delivery timetable will be watched across the industry.
Background
The announcement lands on the eve of Nvidia’s quarterly results, due after the market closes on 17 November, with investors watching margins and guidance. OpenAI’s embedded safety assessments, announced the same day, put outside researchers inside its training runs — a reminder that its buildout and oversight are advancing together.


0 Comments