Chips & Compute

TensorWave Signs Six-Year OpenAI Deal for 457 MW of AMD AI Capacity

TensorWave will supply OpenAI with 457 MW of dedicated AMD MI455X and Helios capacity over six years, becoming the first named specialist AI cloud in the buildout.

TensorWave Signs Six-Year OpenAI Deal for AMD-Powered AI Capacity
TensorWave Signs Six-Year OpenAI Deal for AMD-Powered AI CapacityTho-Ge / pixabay
By Rebecca Stone5 min read

Why it matters

  • TensorWave announced a six-year deal on November 16, 2026 to deliver 457 MW of dedicated AMD capacity to OpenAI.
  • The buildout uses AMD Instinct MI455X GPUs and Helios rack-scale systems, with first capacity online in H1 2027 and full deployment by early 2028.
  • The deal extends the AMD-OpenAI partnership announced October 6, 2025, which covers 6 GW of AMD Instinct GPUs.
  • TensorWave reports more than 2 GW of secured long-term data-center capacity.
  • TensorWave says it is the first publicly announced specialist AI-cloud provider in the OpenAI-AMD collaboration.

TensorWave will deliver 457 megawatts of dedicated, AMD-powered AI capacity to OpenAI over six years, under an agreement announced on November 16, 2026. The deal covers multiple U.S. data center sites and makes TensorWave the first publicly named specialist AI cloud operator inside OpenAI's AMD infrastructure buildout.

The capacity will run on AMD Instinct MI455X GPUs deployed in Helios rack-scale systems, connected by high-performance Ethernet for large-scale training and inference. TensorWave expects to bring the first capacity online in the first half of 2027 and to complete the full deployment by early 2028.

The arrangement matters for the structure of the AI compute market. OpenAI has framed its AMD strategy around a direct relationship with the chip supplier. This contract adds a third party: an infrastructure operator that builds and runs the machines. For AMD, a named cloud operator with a committed anchor customer strengthens the case that its Instinct GPUs can carry production AI workloads at scale. For TensorWave, a company that has staked its business on AMD hardware, the contract is its largest deployment to date.

What does the deal actually deliver?

OpenAI will receive dedicated, single-tenant capacity reserved for its own workloads, not shared cloud capacity carved out of a multi-tenant pool. The infrastructure is intended to support both large-scale model training and inference, the two workloads that dominate AI operators' compute budgets.

The engineering scope is broad. Delivering 457 megawatts across several sites requires coordination across:

  • GPU systems — AMD Instinct MI455X GPUs in Helios rack-scale configurations
  • Networking — high-performance Ethernet linking the racks for distributed training and inference
  • Power, cooling and day-to-day operations at each site

TensorWave describes the project as a significant expansion of its U.S. infrastructure footprint. It also frames the build as a natural extension of its record: the company has already constructed and operated large-scale AMD GPU clusters, and the OpenAI project turns that experience into a dedicated deployment for a single customer on next-generation systems.

Who said what?

TensorWave president and co-founder Piotr Tomasik tied the contract to the company's original thesis. "TensorWave was built to prove that AMD infrastructure could operate at the scale and reliability that the world's most demanding AI companies require," he said in the announcement.

The quote carries weight because of the customer behind it. OpenAI is among the most compute-hungry organizations in the industry, and its willingness to place 457 megawatts with an AMD-focused specialist cloud is a direct test of that thesis. Tomasik's statement now has a measurable deadline attached: first capacity in the first half of 2027, full deployment by early 2028.

How does this fit OpenAI's AMD partnership?

The TensorWave agreement builds on the AMD-OpenAI strategic partnership announced on October 6, 2025. That arrangement is multi-year and multi-generation, covering 6 gigawatts of AMD Instinct GPUs. Its initial 1-gigawatt deployment of MI450-series GPUs was scheduled to begin in the second half of 2026.

The 2025 partnership also includes a technical component. AMD and OpenAI agreed to share expertise to optimize their respective product roadmaps, extending hardware and software collaboration that began with the MI300X and continued with the MI350X.

TensorWave says it is the first publicly announced specialist AI-cloud provider supporting that collaboration. The distinction matters. Until now, the partnership's public face consisted of two parties: the chip supplier and the model developer. TensorWave adds a delivery layer — an operator that will physically host, connect and run dedicated AMD capacity on OpenAI's behalf.

Why TensorWave?

TensorWave's track record with AMD hardware underpins its selection. The company states that it was the first public cloud to deploy AMD Instinct MI300X GPUs, and that it later expanded across MI325X and MI355X hardware. It now operates one of the world's largest AMD-exclusive AI clouds and one of North America's largest AMD-based AI training clusters. It is also an AMD Helios launch partner.

The OpenAI project extends that lineage. TensorWave moves from deploying individual GPU generations to building purpose-built Helios rack-scale systems — AMD's integrated rack platform — for a single tenant. The progression tracks AMD's own product arc, and it positions TensorWave as the operator in the best position to commission Helios at multi-site scale.

The company also reports that it has secured more than 2 gigawatts of long-term data-center capacity. That secured pipeline, more than four times the size of the OpenAI commitment, provides the physical foundation for the U.S. expansion the contract requires.

What happens next?

Execution is now the immediate test. TensorWave must convert its AMD operating experience and its secured power pipeline into reliable single-tenant infrastructure across several sites, on a timetable that starts service within roughly six months of the announcement and completes the build by early 2028. Multi-site deployments of this scale depend on synchronized GPU systems, networking, power and cooling — any slippage in one layer delays the whole.

For OpenAI, the arrangement adds a dedicated delivery channel for AMD compute that sits alongside its direct strategic work with the chip supplier. If the first sites come online on schedule in the first half of 2027, TensorWave becomes a proven second path for AMD capacity into one of the industry's largest AI estates. If they do not, OpenAI's 6-gigawatt AMD partnership will rest more heavily on deployments made directly with AMD itself.

Source: unite.ai

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

728 articles

Related articles

  1. OpenAI Commits to 6 Gigawatts of AMD GPUs in Multi-Year Deal
  2. OpenAI and Broadcom to deploy 10 gigawatts of custom AI accelerators
  3. AMD launches 256-core EPYC 9006 CPUs to fight Intel and Nvidia
  4. NVIDIA to invest up to $100 billion in OpenAI for 10 gigawatts of AI infrastructure
  5. AWS and OpenAI Sign $38 Billion Multi-Year Compute Deal

« Previous articleNext article »