Cerebras Systems opened its IPO roadshow on Monday at a $40 billion target valuation, with up to $4 billion in proceeds, after the deal's lead banks — Morgan Stanley, Citigroup, Barclays, and UBS — booked more than $10 billion in indications of interest in advance of pricing. [1] [2] Pricing is expected mid-May. The company will list under the ticker CBRS. The IPO would be the largest U.S. AI-infrastructure listing of 2026 and the largest semiconductor IPO since Arm's 2023 debut.

The bookbuild structure is the news. An IOI book larger than the offer is a market signal: at $4 billion of paper for sale, the underwriters have $2.50 of demand for every $1.00 of supply. The 2.5-times oversubscription is on the upper end of what the AI-infrastructure cycle has produced and is what permits the pricing-band tightening that typically follows the roadshow's first week. [3] Bloomberg, which broke the $10 billion IOI level on Friday, reported that the order book is led by long-only public equity managers. [4] Sovereign-wealth participation is rumored at the $1.5 billion-plus level; the prospectus is silent on named anchors.

The OpenAI compute deal is the anchor. Cerebras disclosed a multi-year compute agreement with OpenAI worth more than $10 billion — covering up to 750 megawatts of inference capacity through 2028. [5] [6] The deal closed in January 2026 and was the precipitating event that allowed Cerebras to refile its S-1 in mid-April after a 2024 CFIUS-induced retreat. [7] OpenAI is acquiring inference capacity, not training capacity. The 750 MW figure represents one of the largest AI inference deployments in the world. The runway through 2028 means the deal is, in revenue-recognition terms, multi-year ratable revenue against a company whose 2025 booked revenue was $510 million. The order-book tells the underwriters what the public market thinks of that arithmetic. [4]

The valuation walk is steep. Cerebras's Series G in September 2025 priced the company at $8.1 billion. Its Series H in February 2026 — a $1 billion round — priced at $23 billion. The proposed $40 billion IPO valuation, if achieved, is a 74% step-up over the Series H in roughly three months and a 394% step-up over the Series G in eight. [3] [8] The price-to-2025-sales multiple at $40 billion is roughly 78 times. By comparison, the trailing public-comps print for AI-infrastructure peers is closer to 25 times forward sales, with NVIDIA itself at roughly 18 times. The arithmetic only closes if the OpenAI deal lifts forward revenue closer to a $4-5 billion run rate by 2028. The S-1 filings have not yet disclosed the year-by-year revenue ramp inside the OpenAI contract.

Cerebras's product is the wafer-scale engine — a single integrated-circuit assembly the size of a dinner plate with roughly 900,000 cores on a single die. The architecture's argument against NVIDIA's GPU is that very-large neural-network inference benefits from on-die memory locality. The buy-side question is whether the Cerebras advantage holds through three more NVIDIA generation cycles.

Berkshire Hathaway's Saturday Q1 disclosure — $397.4 billion of unallocated cash, with Greg Abel's "AI has to be additive" line dominating Monday analyst notes — sits in the same Monday tape as the Cerebras opening. [9] One side has $397 billion of cash and is not buying. The other side has $10 billion of indicated demand and is bidding 78 times sales.

The OpenAI relationship is the dependency. Public filings suggest more than 60% of contracted forward revenue is concentrated in the single customer. Cerebras's $40 billion valuation depends on the OpenAI contract, which depends on OpenAI's compute strategy, which depends on the willingness of consumer- and enterprise-markets to pay for inference at scale. The bookbuild is the buy-side's pricing of that chain. [4]

The roadshow runs through May 14. Pricing is targeted for May 14 or 15. Trading begins the following day. The CBRS ticker will be the cleanest single Mag-Seven AI-capex referendum a public-market portfolio manager has had since the cycle began.

-- THEO KAPLAN, San Francisco