TensorWave has signed a six-year agreement to supply OpenAI with 457 megawatts of dedicated AI infrastructure, built on AMD Instinct MI455X GPUs in AMD’s Helios rack-scale systems. The Las Vegas neocloud announced the deal on November 16. It is the first neocloud publicly named as a provider under the AMD and OpenAI partnership, and the company calls it its largest deployment to date.
The capacity is spread over several US data center sites and will run single-tenant for OpenAI. Nobody else shares the racks. Networking between racks is high-performance Ethernet. The first capacity should come online in the first half of 2027, with the full build done by early 2028.
Where it fits in the AMD and OpenAI deal
AMD and OpenAI announced their partnership in October 2025: 6 gigawatts of AMD GPUs over several product generations, starting with a 1 gigawatt MI450 deployment in the second half of 2026. AMD issued OpenAI a warrant for up to 160 million AMD shares, vesting as deployment milestones and AMD share-price targets are hit.
Six gigawatts is a lot of power, and somebody has to build the halls, secure the grid connections and run the racks. TensorWave’s 457MW is a slice of that, a bit under 8% of the headline figure. It also shows how the deal gets delivered in practice. OpenAI buys more than chips here. It rents finished, dedicated capacity from a cloud that already runs AMD hardware at scale.
Who TensorWave is
TensorWave is backed by AMD Ventures and runs one of the largest AMD-only AI clouds. It is also a launch partner for Helios. The company closed a Series B in June 2026 and says it has locked in more than 2 gigawatts of long-term data center capacity, so this contract uses well under a quarter of what it has lined up.
Piotr Tomasik, TensorWave’s president and co-founder, said AI infrastructure is “entering a new phase”, one where the providers that can bring hundreds of megawatts online quickly and reliably pull away from the rest. Vamsi Bopanna, SVP of AI at AMD, called TensorWave a strong partner in scaling Instinct infrastructure. Uday Ruddarraju, OpenAI’s CTO for compute, said the agreement moves the AMD partnership forward and helps OpenAI keep scaling.
For AMD the deal matters for a simple reason. Nvidia still dominates rented AI compute, and every dedicated MI455X hall that goes live for OpenAI becomes a reference other buyers can point to. The real test comes in 2027, when the first racks have to show up on schedule.