Cerebras Systems opens its IPO roadshow Monday morning. The S-1, filed April 17, set an initial price band that values the company between $22 billion and $25 billion. The pricing target the lead banks are walking into the institutional pitch with on Monday is approximately $40 billion. [1] If pricing lands at the upper end of demand, Cerebras will list mid-May at a valuation that is roughly 60% above its filed band. The secondary market — the private trades that have happened since the filing — has cleared in the $26-28 billion range. That is the gap the bookbuild has to close.

The paper covered the run-up in Cerebras asks for four billion at forty billion and the secondary market clearing price is twenty-six. The piece that needs to be added Sunday is what the lead banks will be reading aloud Monday morning. Page 12 of the S-1 lists OpenAI three times in three different roles. OpenAI is the company's largest customer, with a multi-year contract worth more than $10 billion in committed revenue across up to 750 megawatts of compute capacity. OpenAI is the company's largest lender, with a $1 billion convertible note priced at the IPO valuation. OpenAI is also a warrant shareholder under a strategic-purchase agreement that vests at IPO. [2] One counterparty. Three balance-sheet roles. The same page.

The geometry is not new in venture finance — strategic-investor structures often combine equity and commercial — but at $40 billion, the concentration is pricing the OpenAI relationship at roughly 50% of total enterprise value. Anthropic is not in the prospectus. Microsoft is not in the prospectus. Google is not in the prospectus. The single name that is in the prospectus, as customer-lender-shareholder, is the same name that has been the silent equity-holder of every AI valuation in the post-October-2025 cohort. Cerebras does not have a Microsoft. It has the OpenAI of the Microsoft RPO disclosure: 45% of $625 billion, $281 billion, on page 9 of the 10-Q. [3]

What MSM has emphasized: the demand-driven repricing. TheNextWeb's Sunday writethrough led with "Cerebras targets $40 billion." [4] TechCrunch framed the bookbuild as the first blockbuster AI IPO of the year. [5] The Forge secondary tape, which has run between $26 and $28 since mid-April, is the data point that does not appear in the demand-driven story. [6] The two numbers — $40 and $27 — measure two different things. The Forge tape measures what willing buyers paid for shares with no listed venue. The roadshow target measures what investment banks believe institutional buyers will pay for shares with a NASDAQ ticker and an OpenAI counterparty disclosure. The premium of $13 billion is the value the public-market institutional bid puts on lockup, liquidity, and the OpenAI marker.

Two pieces of context complicate the institutional bid. The first is the litigation environment. Cerebras's largest manufacturing partner runs through a fab whose adjacent facility is the subject of the Tumbler Ridge tort, a wrongful-death suit filed in the Northern District of California with seven Edelson complaints consolidated. [7] The complaint is not against Cerebras. The complaint is against OpenAI. The customer-lender-warrant-holder. The investor read of the Tumbler Ridge docket on Cerebras's IPO is an indirect read: not whether Cerebras has tort exposure, but whether OpenAI's tort exposure compromises the contract value, the loan repayment risk, or the warrant economics. The S-1 mentions "ongoing litigation involving counterparties" in a single risk-factor paragraph. [2] That paragraph is the part of the document the buy-side will spend the most time on this week.

The second piece is the Florida USF criminal probe. The OpenAI Tampa data center, where a former employee has alleged falsified power-draw reporting under the Universal Service Fund subsidy regime, is currently under federal investigation. [8] The probe is independent of Cerebras. The complication is that the same OpenAI 750-megawatt commitment that drives the Cerebras revenue model is also the demand center the USF probe is investigating. The bookbuild, then, is asking institutional buyers to underwrite a counterparty whose two largest exposures — a major civil tort and a federal criminal probe — are both unresolved.

What X has emphasized: the secondary mark. The Polymarket adjacent on AI IPO valuations has held the $27-30 billion range as the median expected close. The X discourse around the S-1 since April 17 has been consistent: the $40 billion target is what the bank syndicate would like to print, the $27 billion is what the secondary tape says, and the gap is the OpenAI premium. The premium is not absurd. It is also not free. It is the price institutional buyers are being asked to pay for a single-customer concentration in a tort-and-probe environment.

The lead banks — Morgan Stanley, Goldman Sachs, JP Morgan, and BofA, in the named-on-the-cover sense — have the institutional pitch this week and pricing the Thursday or Friday after. Lockup is the standard 180 days. The expected listing is on NASDAQ under the proposed ticker WSE. [2] The float is approximately 10% of post-IPO shares outstanding. The proceeds — at $40 billion enterprise value, with a $4 billion offering, that is the same ratio Snowflake printed in 2020 — are earmarked for capacity expansion at the Wisconsin and Texas sites and for the second tranche of the Sunnyvale R&D buildout.

The honest mark is in the secondary tape. The aspirational mark is what the banks are walking into Monday with. The number that prints, mid-month, will determine whether the AI-state-power register continues to be denominated in single-counterparty premiums or whether the public market reasserts the $27 number it has been quoting on Forge for six weeks. Either answer is news. The number itself is the referendum.

The other thing the bookbuild will reveal: whether the buy-side reads page 12 the way the paper does. If it does, the $40 will not hold. If it does not, the next AI IPO will look like this one.

-- THEO KAPLAN, San Francisco