Cerebras Systems is a US AI accelerator company that designs wafer-scale processors — single chips covering an entire silicon wafer — positioned as an alternative to Nvidia GPU clusters for training very large neural networks and serving high-throughput inference workloads. Its current product is the CS-3, built around the Wafer-Scale Engine 3. The company filed for a US IPO in May 2026, its second attempt, targeting a valuation of roughly $40 billion and a raise of up to $4 billion.
Snapshot
Valuation
| Date | Amount | Round / Event | Source |
|---|---|---|---|
| 2026-05-13 | ~$56B | IPO priced at $185/share | (Source: cerebras.ai) |
| 2026-05-04 | up to $26.62B (target) | IPO filed (second attempt); 28M shares at $115–$125, raise $3.5B; FY2025 revenue $510M; ~$20B / 750 MW multi-year compute deal with OpenAI disclosed | reuters.com |
| 2026-05-01 | ~$40B (target, withdrawn) | Bloomberg leak; later refined to $26.62B in May 4 S-1 filing | bloomberg.com |
Background
Cerebras was founded in 2015 by Andrew Feldman and a founding team that included several people from SeaMicro, a company AMD had acquired. The company's central technical bet is that wafer-scale silicon eliminates the chip-to-chip interconnect overhead that limits cluster-scale training, with one wafer effectively acting as a single processor with roughly 850,000 cores and multi-terabyte-per-second on-chip bandwidth.
Customers and deployments
Cerebras serves frontier-AI inference partners, including Mistral, and government customers. Its inference pricing per million tokens has repeatedly led public benchmark tables in 2024–2025 for select Llama and Mistral configurations.
In June 2026 OpenAI said it would launch GPT-5.6 Sol on Cerebras hardware in July 2026, citing speeds of up to 750 tokens per second for latency-sensitive enterprise applications (Source: venturebeat.com).
OpenAI equity stake
A quarterly filing released after the market closed on August 12, 2026 disclosed that OpenAI exercised every vested Cerebras warrant share during July 2026, acquiring 10,033,508 Class N shares at $0.00001 each for about $100 in cash. The stake is 4.22% of the 237,564,041 shares outstanding as of August 5, 2026, with an implied value near $2.3 billion at the Class A price of about $229 on August 13 — the day OpenAI previewed Ultrafast, a Cerebras-hosted tier running GPT-5.6 Sol at up to 750 output tokens a second (Source: implicator.ai). The warrants pair the customer relationship with an equity position taken at nominal cost, so the roughly $20 billion compute commitment and the stake move together in value.
IPO and valuation
Cerebras filed for a US IPO in May 2026, its second attempt at going public. Bloomberg first reported on May 1, 2026 that the company was targeting a valuation of approximately $40 billion and a raise of up to $4 billion (Source: bloomberg.com). The S-1 filing reported on May 4 refined the terms to 28 million shares priced at $115–$125, a target raise of about $3.5 billion and a valuation of up to $26.62 billion. The filing disclosed FY2025 revenue of $510 million and a multi-year compute deal with OpenAI valued at roughly $20 billion for 750 MW of capacity (Source: reuters.com).
At the targeted valuation, Cerebras would rank below Nvidia among AI-accelerator companies by market capitalization but ahead of Groq, Tenstorrent, and other AI-chip startups. Cerebras priced the IPO on May 13, 2026 at $185 per share, valuing the company at roughly $56 billion (Source: cerebras.ai).
Competitive context
The IPO follows Google's April 30, 2026 announcement that it will sell TPU chips for installation in customers' own data centers. Together, the two developments are read as evidence that non-Nvidia AI silicon has become a credible procurement option for hyperscalers and frontier labs.
The May 2026 IPO timing also intersects with the broader AI-bubble debate: a pure-play AI hardware company seeking a roughly $40 billion valuation, in a market where hyperscaler 2026 capital expenditure is projected at $725 billion, serves as one of the nearer-term tests of how public-market investors price the AI-chip stack.
Relationships
- related: Nvidia & TSMC — AI Compute Infrastructure (incumbent), Google DeepMind (TPU chip-vendor competitor), Semiconductor Supply Chain, Inference Economics and Token Pricing, AI Bubble Debate.
- related: OpenAI (~$20B compute deal; GPT-5.6 Sol inference host), GPT-5.6 (Sol, Terra, Luna).
- deploys-in: Financial Services — AI Deployment, Defense / Military — AI Deployment.