Particle.news
Download on the App Store

OpenAI Taps Cerebras in Multi-Year Deal for Up to 750 MW of Low-Latency AI Compute

The agreement brings Cerebras’ wafer-scale systems to accelerate real-time inference across OpenAI’s services.

Overview

  • Cerebras is set to deliver up to 750 megawatts of compute in phased rollouts starting this year and continuing through 2028.
  • The structure has Cerebras building or leasing data centers and selling cloud services that OpenAI will use for inference.
  • The deal is reported by multiple outlets to exceed $10 billion over its term, a figure not detailed by the companies.
  • OpenAI says the integration targets faster responses for inference and reasoning models to enable real-time interactions.
  • The partnership expands OpenAI’s multi-vendor strategy beyond Nvidia and helps Cerebras diversify beyond G42 and advance IPO and funding plans, with Sam Altman identified as an early Cerebras investor.