Cerebras Refiles IPO Targeting US$40 Billion Valuation


TL;DR

  • Refiled IPO: Cerebras filed an April 2026 S-1 seeking up to US$4 billion at roughly a US$40 billion valuation, about double its US$23 billion February mark.
  • Backlog Pillar: A US$24.6 billion revenue backlog and 2025 sales of US$510 million underpin the case, but UAE-linked buyers still carry 86% of revenue.
  • OpenAI Anchor: A January OpenAI deal commits 750 megawatts through 2028 above US$10 billion, expanded in April by another US$10 billion in spending.
  • Pricing Test: Nasdaq pricing will gauge whether the cleared CFIUS review and 76% revenue growth can sustain the doubled valuation against Nvidia.

Cerebras Systems filed a refiled IPO prospectus seeking up to US$4 billion and positioning the company at around US$40 billion, challenging Nvidia in AI chip infrastructure after a 2024 withdrawal.

The April 2026 S-1 would rank among the largest AI-chip listings to date and roughly doubles a last private valuation around US$23 billion set after a US$1 billion February capital raise. Cerebras returns to the SEC with a cleared U.S. regulatory review, a multibillion-dollar revenue backlog, and an anchor commitment from OpenAI that builds on its earlier Nvidia-challenger inference launch.

Inside the Refiled Prospectus

Cerebras’s US$24.6 billion revenue backlog sat on the books at year-end 2025, with management expecting to recognize roughly 15% of it across 2026 and 2027. A backlog-to-revenue ratio of nearly 50 to 1 is the load-bearing pillar of the US$40 billion valuation case.

Two UAE-linked buyers carry 86% of 2025 revenue: the Mohamed bin Zayed University of Artificial Intelligence at 62% and G42 at 24%. An earlier 2024 prospectus had shown G42 alone delivering 87% of first-half revenue, so concentration has shifted but not eased, and a widening operating loss compounds the pushback against the doubled valuation.

OpenAI Anchor Reshapes the Customer Book

Diversification leans on a January agreement with OpenAI’s compute commitment to Cerebras for 750 megawatts of inference capacity through 2028, valued above US$10 billion. An additional US$10 billion in spending commitments followed in April. Cerebras’s prospectus also points to an AWS partnership for Cerebras inference on Amazon Bedrock as a second hyperscaler-scale demand anchor, a structural counterweight investors will want against the UAE concentration.

CEO Andrew Feldman has framed the proceeds as fuel for capacity expansion, additional data centers, and a performance roadmap targeting order-of-magnitude gains over current systems. That OpenAI contract alone is the operational test of whether Cerebras can build, power, and staff enough Wafer Scale Engine deployments through 2028 to convert backlog into recognized revenue on the schedule the prospectus assumes.