OpenAI signs 457MW AMD compute deal with TensorWave in six-year agreement

AI cloud provider TensorWave will supply OpenAI with 457 megawatts of dedicated capacity powered by AMD Instinct MI455X chips, the first publicly named cloud partner in the OpenAI–AMD partnership.
AI cloud provider TensorWave announced on November 16 that it has signed a six-year agreement with OpenAI to supply 457 megawatts of dedicated AI computing capacity across multiple data centres in the United States, all powered by AMD's latest Instinct MI455X chips. It is the first cloud infrastructure provider to be publicly named under the OpenAI–AMD strategic partnership.
The deal: 457MW, six years, dedicated capacity
Under the agreement, TensorWave will provide 457MW of IT capacity at several U.S. data centre sites, using AMD Helios rack-scale systems built around Instinct MI455X GPUs. The capacity will run as dedicated, single-tenant infrastructure reserved for OpenAI, for training and running AI models. The first capacity is expected to come online in the first half of 2027, with full deployment targeted for early 2028. TensorWave calls it the largest deployment contract in its history. Neither the price nor the specific sites have been disclosed.
Why AMD: part of a 6GW strategic partnership
The deal sits within the strategic partnership OpenAI and AMD announced on October 6, 2025. The multi-year, multi-generation arrangement covers up to 6 gigawatts of AMD Instinct GPUs, with the first 1-gigawatt deployment of MI450-series GPUs scheduled to begin in the second half of 2026. TensorWave is the first neocloud to be publicly announced as an infrastructure provider supporting that partnership. Backed by AMD Ventures, the company runs AMD hardware exclusively and says it was the first public cloud to deploy AMD Instinct MI300X chips, later expanding to MI325X and MI355X, and now operates one of the world's largest AMD-exclusive AI clouds. TensorWave raised $350 million in a Series B round in June 2026 and says it has secured more than 2 gigawatts of long-term data centre capacity.
Helios racks: compute by the rack
The deployment will use AMD's Helios rack-scale systems, launched in July, which combine Instinct MI455X GPUs, sixth-generation EPYC 'Venice' server processors, ROCm software and Pensando networking into a single rack architecture, linked by high-performance Ethernet for large-scale training and inference.
Market context: is demand being double counted?
On the same day, OpenAI chief executive Sam Altman warned at a Semafor event that AI chip demand may be double counted: many chipmakers assume that OpenAI, Anthropic and a few others will buy most of their output, so the same demand is counted several times over in different suppliers' plans. He said only OpenAI and Anthropic are driving a substantial increase in demand, and that supply shocks and periods of instability are possible. The market reaction to the TensorWave announcement was mixed: AMD shares closed Monday at $504.28, down 5.6% from Friday, while the iShares Semiconductor ETF (SOXX) fell 5.4% the same day.
Sources: TensorWave (Business Wire announcement), DatacenterDynamics, Unite.ai and Semafor event coverage.
N環球 
💬 Comments
No comments yet. Be the first!