Updated ETGet the DI Daily by Email

Technology news for e-commerce professionals, with our take on why it matters.

Ongoing coverage: The AI infrastructure spending race

OpenAI signs six-year deal for 457MW of AMD-based capacity with TensorWave

OpenAI signs six-year deal for 457MW of AMD-based capacity with TensorWave
Image: DatacenterDynamics

The dedicated Helios racks with MI455X GPUs are expected to start running in the first half of 2027 and be fully deployed by early 2028.

Source: DatacenterDynamics

OpenAI has signed a six-year contract with TensorWave, a neocloud focused on AMD hardware, for 457MW of capacity, DatacenterDynamics reported. TensorWave will supply AMD Helios rack-scale systems built on Instinct MI455X GPUs across multiple data center locations, run as dedicated, single-tenant capacity for OpenAI. Operations are expected to begin in the first half of 2027 and be fully deployed by early 2028. DatacenterDynamics said it is TensorWave's largest deployment contract to date.

AMD claims Helios delivers 15% more peak FP4 performance and 50% more high-bandwidth memory than Nvidia's Vera Rubin NVL72, based on published specifications, and up to 30% more tokens per dollar. Uday Ruddarraju, CTO for compute at OpenAI, said the deal advances the company's previously announced partnership with AMD.

TensorWave raised $350 million in a Series B round in June 2026 and claims to have secured 2GW of long-term data center capacity across the US, according to the report.

Read the full article at DatacenterDynamics

Our take

Every AI tool your team rents runs on capacity like this, and OpenAI buying at scale from an AMD-based provider adds supply beyond Nvidia that could ease inference pricing over time.

Operations are expected to begin only in the first half of 2027, and the report gives no price for the deal. Watch whether AMD's claimed 30% tokens-per-dollar gain shows up in OpenAI's API pricing.

Was our take useful?

Key takeaways

  1. OpenAI contracted 457MW of capacity from TensorWave over six years
  2. Helios racks with MI455X GPUs are expected to begin operating in the first half of 2027
  3. AMD claims Helios gives up to 30% more tokens per dollar than Nvidia's rack

More on this topic

Illustration for The AI infrastructure spending race
Illustration: Digital Influence

Ongoing story8 postsUpdated Nov 19

The AI infrastructure spending race

Hyperscalers keep raising capital spending as demand for AI cloud capacity outruns supply.