Particle.news
Download on the App Store

Cerebras Positions Itself as an AI Hardware Ecosystem Excluding Nvidia

An effort to build an alternative to Nvidia through deals with Amazon and OpenAI hinges on Cerebras proving it can scale fragile wafer‑level chip production.

Overview

  • Cerebras, which outlined the strategy at the Bloomberg Tech conference on June 3, said it is partnering with essentially every major AI data‑center component supplier except Nvidia.
  • The company reported a deployment agreement with Amazon Web Services to run Cerebras processors inside AWS data centers alongside AWS’s own chips.
  • Cerebras has secured a supply deal with OpenAI that media reports describe as a multi‑year commitment tied to large inference capacity needs, with one report citing a 750 MW figure.
  • The firm completed an IPO in early 2026 after withdrawing a prior filing in 2024, a shift that broadened its investor base but also raised expectations for fast commercial execution.
  • Cerebras’s wafer‑scale engine uses an entire silicon wafer as a single processor, a design that gives it performance differentiation but creates acute manufacturing and yield risks that could delay deliveries to hyperscalers.