Cerebras Files for Nasdaq IPO With $20B OpenAI Contract and $510M in Revenue
Cerebras Systems filed its S-1 with the SEC on April 17, making its second attempt to list on the Nasdaq under ticker CBRS. Morgan Stanley, Citigroup, Barclays, and UBS are lead underwriters. The first filing, pulled in October 2025, was delayed by a U.S. national security review of UAE-linked investor G42; the company received CFIUS clearance in 2025.
The Numbers
- Revenue: $510M in 2025, up 76% from $290.3M in 2024
- Profitability: $1.38 EPS in 2025 vs. a $9.90-per-share loss a year earlier — first full year of net income
- Backlog: $24.6B in remaining performance obligations as of December 31, 2025; Cerebras expects to recognize 15% through 2026–2027
- Private valuation: $23B, established in a February 2026 funding round — nearly triple the $8.1B it commanded in September 2025
The OpenAI Anchor
The filing’s central asset is a multi-year agreement with OpenAI valued at more than $20 billion. Under the deal, Cerebras will supply 750 megawatts of compute by 2028, delivering 250MW per year across 2026, 2027, and 2028. OpenAI holds an option for an additional 1.25 gigawatts by 2030. The relationship runs deeper than a purchase order: OpenAI provided a $1B loan to Cerebras at 6% annual interest, earmarked for data center infrastructure, and received warrants to purchase up to 33.4 million Class N (non-voting) shares — exercisable only if OpenAI buys 2GW of cumulative capacity.
Cerebras acknowledges in the filing that OpenAI represents a substantial share of projected revenue for several years. A concentration risk that once centred on G42 (87% of revenue in 2024, now 24%) has shifted to a different single customer.
The Architecture Thesis
Cerebras competes on architecture. Its Wafer Scale Engine 3 places an entire neural network accelerator on a single silicon wafer rather than cutting the wafer into discrete chips and networking them together. That eliminates inter-chip communication overhead — the primary bottleneck at inference scale — and delivers faster token generation at lower per-inference cost compared to GPU clusters. The company positions itself as the inference layer for frontier AI, not a training competitor to Nvidia’s H100/H200 line.
An Amazon Web Services integration announced alongside the IPO filing puts Cerebras chips inside AWS data centers, giving hyperscaler distribution to what was previously a direct-sale infrastructure play. Amazon doesn’t add hardware vendors for science experiments; the integration signals predictable enterprise demand.
Context
The filing lands 18 months after AI hardware IPOs effectively paused. Investors pulled back in late 2024 when demand trajectories were unclear. Cerebras going public with a $24.6B backlog and a signed hyperscaler partnership reframes the question: the inference market isn’t speculative, it’s contracted. If Cerebras sustains even partial execution on the OpenAI commitment, CBRS would represent the first direct public-market play on inference-optimised silicon — a segment that, by most estimates, accounts for 60–70% of total AI compute spend.
The risk is concentration. A contract renegotiation or delivery failure with OpenAI would materially impair 2026–2028 revenue visibility. The filing is candid about this.
Key Numbers
| Metric | Value |
|---|---|
| 2025 Revenue | $510M |
| Revenue Growth | +76% YoY |
| 2025 EPS | +$1.38 (vs. -$9.90 in 2024) |
| Backlog (RPO) | $24.6B |
| OpenAI Deal | $20B, 750MW by 2028 |
| OpenAI Loan | $1B at 6% interest |
| Pre-IPO Valuation | $23B (Feb 2026 round) |
| Ticker | CBRS (Nasdaq) |