Skip to content
Rush Commerce
AI & Automation2 min read

TensorWave-OpenAI 457 MW AMD deal: capacity starts 2027

TensorWave will run 457 MW of AMD MI455X capacity for OpenAI under a six-year deal. First capacity comes in 2027. Here is what it means for your AI costs.

TensorWave, an AMD-only GPU cloud, signed a six-year deal to supply OpenAI with 457 megawatts of AMD-powered AI capacity. It is the first neocloud publicly named as a supplier under the AMD–OpenAI partnership. It is another sign that OpenAI is moving more of its compute off Nvidia. None of this capacity is online yet, and none of it changes your OpenAI invoice this year.

What actually happened

Per TensorWave's Business Wire release on November 16:

  • 457 MW of IT capacity across multiple U.S. data center sites. The release names no locations.
  • AMD Instinct MI455X GPUs in AMD Helios rack-scale systems, with Ethernet scale-out networking.
  • Dedicated, single-tenant capacity for OpenAI.
  • Timeline: capacity starts coming online in the first half of 2027, with full deployment expected by early 2028.
  • No contract value and no financing details are disclosed.

OpenAI's Uday Ruddarraju, CTO for compute, described the deal as a step forward in the company's earlier partnership with AMD. The release does not say whether these 457 MW count toward the 6 gigawatts AMD and OpenAI announced earlier. TensorWave's own post calls it a multi-year partnership.

Why it matters for your business

Two points for an operator, neither of them urgent.

First, the AI supply chain is getting a second source. AMD hardware running frontier workloads at this scale makes Nvidia pricing less absolute. Over time, more supply tends to push per-token prices down. But "over time" here means 2027 and 2028, and the deal says nothing about API pricing.

Second, the risk moves to new companies. Your OpenAI calls will increasingly run on capacity built by GPU clouds you never signed a contract with. That is fine until a build slips. A late data center shows up for you as rate limits, slower responses or a delayed model, not as a press release.

So do not plan this year's AI budget around cheaper tokens. Plan it around what you pay today, and make sure your app can move a workload to a second model vendor if your main one starts throttling.

Key takeaways

  • TensorWave will supply OpenAI with 457 MW of dedicated AMD MI455X capacity under a six-year deal
  • First capacity comes online in the first half of 2027; full deployment is expected by early 2028
  • No contract value, site locations or link to the 6 GW AMD figure was disclosed
  • More AMD supply may lower token prices later, not this year
  • Budget on today's prices and keep a second model vendor ready

Your AI vendor's data center schedule should not decide when your app slows down. We build model-agnostic integrations with a tested fallback vendor, so a rate limit at one provider becomes a config change, not an outage. See how we build vendor-agnostic AI systems, or price your current AI workload.

Sources: TensorWave release via Business Wire, TensorWave blog.

  • #openai
  • #amd
  • #tensorwave
  • #ai-infrastructure
  • #gpu-capacity
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.