Cerebras Systems raised $5.5 billion in its IPO on Thursday, pricing 30 million shares at $185 the night before and then watching them open at $385, a 108% first-day pop. The pricing itself was already well above the initial $115–$125 range and the revised $150–$160 range, and the public market doubled it on the open. At the $185 IPO price, the AI chip designer entered trading at a $56.4 billion fully-diluted valuation.
The stock cooled only slightly through the session, changing hands above $330 mid-day. Co-founder and CEO Andrew Feldman's stake is worth nearly $1.9 billion at $185 per share; co-founder and CTO Sean Lie's stake clears $1 billion. Both figures climb sharply if Cerebras holds above $300 into the close. It is the first large tech IPO of 2026 and the largest pricing surprise the AI hardware sector has produced since Nvidia's post-2023 run — and Cerebras competes directly with Nvidia, having designed its wafer-scale chip from scratch for AI workloads.
A year ago, this outcome looked unreachable. Cerebras first filed to go public in 2024, but a large investment from Abu Dhabi-based Group 42 dragged the deal into an extended review by the Committee on Foreign Investment in the United States. Investors had a second problem with the financials: G42 accounted for nearly all of Cerebras's revenue at the time, a customer-concentration profile that public markets rarely tolerate. The company shelved the offering.
Key facts
- 01Cerebras raised $5.5B at a $185 IPO price, well above its revised $150–$160 range and original $115–$125 range.
- 02Shares opened at $385, up 108%, before settling above $330 mid-day Thursday.
- 03Fully-diluted valuation hit $56.4B at the IPO price, with CEO Andrew Feldman's stake worth $1.9B and CTO Sean Lie's at roughly $1B.
- 042025 revenue reached $510M, up 76% year-over-year, with $237.8M in net income versus a near-$500M loss the prior year.
- 05Customers now include OpenAI, G42, Saudi's MBZUAI and Amazon Web Services.
IPO plans reappeared in April after a sharply different set of numbers. Cerebras reported $510 million in 2025 revenue, up 76% year-over-year, and a swing to $237.8 million in net income from a loss of nearly half a billion dollars the prior year. The customer list also broadened. Cerebras now counts OpenAI — in what the company has described as a circular commercial arrangement — alongside G42, Saudi Arabia's Mohamed bin Zayed University of Artificial Intelligence, and Amazon Web Services.
“Cerebras priced at $185 and opened at $385, a 108% pop that pushed its fully-diluted valuation to $56.4B and made co-founder Andrew Feldman's stake worth nearly $1.9B.”— Jaeden Schafer
The pitch to public investors is inference. Training new frontier models still runs almost entirely on Nvidia GPUs, but the compute required to actually serve those models to users — answering prompts, running agents, processing tokens — is a separate and rapidly growing workload. Cerebras has positioned its wafer-scale design as a high-throughput, low-latency option for that side of the equation, and the customer roster suggests at least four serious buyers are testing the thesis at scale.
The chip itself is the company's distinguishing bet. Rather than packaging many smaller dies, Cerebras builds a single silicon wafer as one massive chip, keeping data movement on-die instead of shuttling it across interconnects. That architecture is unconventional and capital-intensive to manufacture, but it produces throughput numbers on inference workloads that smaller-die competitors have struggled to match. Public-market investors are now effectively underwriting that architectural bet against Nvidia's roadmap.
The revenue concentration risk has eased but has not disappeared. G42 remains a major customer, and the OpenAI relationship — described as a circular deal — invites the kind of scrutiny that public companies face quarterly. A handful of large buyers can produce 76% growth one year and a much different curve the next, particularly if any one of them shifts spending to a competing supplier or builds in-house silicon. The CFIUS history around G42 is resolved for IPO purposes but remains a structural feature of the customer base.
For the broader AI hardware market, the Cerebras debut sets a fresh public-market reference point for a Nvidia challenger. At $56.4 billion fully diluted on $510 million in revenue, the multiple is steep by traditional semiconductor standards but in line with how investors are now pricing pure-play AI infrastructure. Whether subsequent AI chip IPOs price off this level — or off the mid-day $330-plus print — will shape funding conditions for every private competitor still raising. Cerebras has just made the comparison concrete, and Nvidia now has a publicly-valued rival on the inference side of the workload it has dominated.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
