Cerebras files for an IPO again, this time with $510M in revenue and a $10B OpenAI deal in its pocket

Cerebras Systems filed its S-1 on April 17 and plans to list on Nasdaq under the ticker CBRS, TechCrunch reported. The targeted listing window is mid-May, at roughly the same $23 billion valuation Cerebras held in its 2025 Series H. The hook in the prospectus: Cerebras booked $510 million in revenue in 2025 and lists an OpenAI deal it values at more than $10 billion over its term.

What we know

• The filing. Cerebras’s S-1 with the SEC went public on April 17. Morgan Stanley is the lead underwriter, per the Motley Fool’s read of the document. The roadshow is expected to begin in late April, with a listing in mid-May.
• Revenue. CNBC notes the $510M 2025 revenue figure. The company posted a non-GAAP net loss of about $76 million (excluding one-time items), though TechCrunch flags GAAP net income of $237.8M, helped by tax-driven items.
• OpenAI is the anchor. The OpenAI deal, characterized as worth $10B-plus, is what changed Cerebras’s story since the failed 2024 IPO. CEO Andrew Feldman put it bluntly: Nvidia “didn’t want to lose the fast inference business at OpenAI, and we took that from them.”
• AWS as the second pillar. Cerebras has a separate supply agreement with Amazon to put its wafer-scale chips into AWS data centers. That’s the same week Amazon topped up its Anthropic stake by $5B and broadly leaned into custom AI silicon.
• What killed the 2024 attempt. Cerebras filed in 2024 and withdrew during a federal review of the company’s relationship with Abu Dhabi-backed G42, its largest customer at the time. Per CNBC, the review cleared, and customer concentration has shifted away from G42 in the past year.

What’s still unclear

• The actual IPO price range. Cerebras hasn’t published a price band. The $23B valuation is the private mark, not the listing target. Investor demand at roadshow is what decides whether it lists at $20B or $30B.
• Customer concentration. OpenAI is the obvious anchor, but the S-1’s full breakdown of customer-by-customer revenue is buried in the risk factors. A lopsided book is the standard pre-IPO complaint about chip companies.
• Capacity. Cerebras’s wafer-scale chips are fabbed at TSMC. With Intel courting Tesla as a 14A foundry customer and TSMC’s leading-edge nodes booked through 2027, Cerebras needs to demonstrate scaled capacity to back its OpenAI commitment. The S-1 doesn’t fully answer that.
• Margin profile vs. Nvidia. Cerebras’s pitch is faster inference per dollar than an H100 or a B200. Nvidia’s margin profile is what Cerebras has to beat on at scale. Public-market investors will read every margin line in the prospectus accordingly.

Who reported it first

CNBC and TechCrunch broke the filing on April 17 and 18. Axios published an S-1 teardown on April 20 with the $10B OpenAI line, and follow-on coverage from Reuters, the Motley Fool, and AI Insider expanded the picture over the weekend. The filing itself is on SEC EDGAR.

What this means for you

If you build on inference, the practical upside of a Cerebras IPO is that a credible Nvidia alternative gets a balance sheet to fund capacity. Wafer-scale inference is fast (claims of 1,800 tokens/sec on Llama 3.1-70B aren’t unusual in their benchmarks), but the bottleneck has been chips on the floor, not silicon design. Going public funds that.

If you’re an enterprise buyer, this matters for pricing power. The OpenAI/Cerebras deal demolished a meaningful chunk of OpenAI’s inference bill on Nvidia. AWS adopting Cerebras adds Bedrock and SageMaker as deployment endpoints over the next year. Your contracts in 2027 will look different if Nvidia, Cerebras, Google’s TPUs, and Broadcom-built Meta MTIAs are all bidding for the same workload.

My read: this filing is a story about whether the AI capex cycle can fund a second, credible inference-first chip company. If CBRS prices well, expect Groq’s, SambaNova’s, and Tenstorrent’s bankers to start the same conversation by July.