Cerebras Systems filed an amended S-1 on May 4 to sell 28 million Class A shares at $115–125 per share, a $3.5 billion raise at a valuation up to $26.6 billion. The company will list on Nasdaq under CBRS.
The price range implies roughly a 15% step-up from Cerebras's $23 billion private valuation in February — small over three months. That's not the interesting number. The interesting number is the order book: banks are fielding approximately $10 billion in indicated orders for the $3.5 billion on offer. At nearly 3× oversubscribed, this thing is going to price above the range.
Three things every investor reading this needs to model.
First: the revenue base is real but concentrated. Cerebras posted $510 million of 2025 revenue, up from $290 million in 2024 — a 76% YoY clip on a non-trivial base. The company also disclosed $24.6 billion in remaining performance obligations, the contracted-but-unrecognized backlog. That's an enormous number for a company this size, and the bulk of it is reportedly tied to OpenAI and a small handful of UAE-affiliated entities. Concentration runs both ways: the customer base validates the technology but exposes the financials to any single counterparty's roadmap.
Second: this is a wafer-scale bet, not a GPU bet. Cerebras's product is the WSE — wafer-scale engine — that bundles silicon at the entire-wafer level rather than dicing into discrete chips. The architecture is genuinely differentiated; that's the entire pitch. It also means the company isn't competing on Nvidia's price/performance curves directly. They're betting the workloads matter more than the unit economics. For training runs at frontier scale, where memory bandwidth dominates, that bet has been working.
Third: the market is hungry for AI infrastructure exposure that isn't priced like Nvidia. NVDA is a four-trillion-dollar-plus name. Funds that can't add concentrated GPU exposure can buy CBRS. That's a meaningful chunk of the order-book demand — pure positioning, not necessarily fundamental conviction.
The bear case is the customer concentration, full stop. If OpenAI shifts its training mix toward in-house silicon (which it's been telegraphing) or toward Nvidia's next generation at scale, the $24.6B in backlog becomes a number with footnotes. Cerebras has been transparent about the concentration in its filings, but the market may be pricing the optimistic scenario.
- The February 2026 private round at $23B was led by Tiger Global and others
- $10B indicated demand against $3.5B on offer suggests pricing well above the $115–125 range — final pricing should land closer to $140
- 2024 revenue was about $290M, so 2025's $510M is a 76% YoY clip
If you're allocating to AI infrastructure as a public-equity sleeve, CBRS is going to be hard to ignore on day one. If you're a builder evaluating Cerebras hardware, the IPO doesn't change the technical story — but it does mean their pricing power and capacity commitments just got more rigid. Lock in your terms before the post-IPO hangover.