Cerebras's IPO roadshow window has slipped from late April to mid-May. The company is targeting a $22-25 billion valuation on a $2 billion raise, with secondary-market quotes implying $26-28 billion — the private mark is now above the public band the underwriter is willing to print. The S-1 still names OpenAI on a single page as customer, lender, and shareholder. [1]

The slip itself is the news. This paper's Apr 30 frame said the same OpenAI counterparty was sitting inside three pressures simultaneously: the Microsoft contract revision that ended exclusivity, the Tumbler Ridge tort docket alleging Sam Altman overruled a twelve-person safety team, and the Florida criminal probe expanding to a second case. The Apr 29 setup named the customer-lender-shareholder triangle before the Apr 29 N.D. Cal. complaints landed. Today's roadshow slip is the underwriters quietly extending the bookbuild window into the period when those dockets become discovery-ripe. Two extra weeks is the silent revision.

The mechanics of the deal are unusual even before the docket complications. OpenAI loaned Cerebras roughly $1 billion in exchange for a warrant to buy stock; OpenAI has access to about a sixth of that warrant already vested, with another approximately 17% set to vest if Cerebras maintains a $40 billion valuation on average for a month following the IPO. [2] OpenAI's $10 billion compute purchase agreement (with potential to expand to additional gigawatt-scale capacity through 2030) is the revenue anchor of the prospectus. Sam Altman's company is therefore (a) the largest disclosed customer, (b) a creditor with warrants, and (c) a shareholder whose stake grows if the post-IPO market validates the bookbuild. That is a circular dependency the SEC has historically been allergic to, and the underwriter is now navigating it with two extra weeks. [3]

The pricing question this raises is binary. Either the IPO clears at the upper end of the band, signaling that institutional investors will absorb the OpenAI concentration risk because the AI-chip thesis is durable, or it clears at or below the lower end, signaling that the dockets and the criminal probe have become a counterparty-discount factor. The secondary market's $26-28 billion implied valuation suggests that retail and accredited-investor demand exists at a higher mark than the underwriter is willing to print. That gap is unusual: typically the IPO band sits above the most recent private mark, not below it. The inversion is what bookbuild rooms talk about when they say a roadshow has "found resistance."

The OpenAI side of the trade is also re-rating. The Apr 29 Tumbler Ridge complaints in N.D. Cal. allege Altman overruled a safety team that wanted to alert the RCMP to the shooter's account. The Apr 30 Microsoft contract revision ended exclusivity through 2032 in exchange for OpenAI continuing to pay Microsoft revenue share through 2030 capped, with Microsoft retaining ~27% equity at a ~$135 billion mark. The Apr 21 Florida AG criminal probe expanded this week to a second case. Florida subpoenaed OpenAI's internal training materials back to March 2024, which is the IPO-prep window the Tumbler Ridge plaintiffs allege Altman was protecting. The same time period now sits inside two separate discovery scopes.

What does this mean for Cerebras's bookbuild? The risk-factor section of any S-1 lists customer concentration and counterparty risk. Cerebras's risk factors will have to be updated to reflect that its largest customer is now in active product-liability litigation alleging managerial misconduct around safety reporting, and is the subject of an expanding state criminal probe. The two extra weeks the underwriter took are likely partly a decision to land that risk-factor update before pricing.

The summer-2026 IPO calendar for AI infrastructure was supposed to start with Cerebras and lead to whatever SpaceX's $20 billion bridge loan converts into (Polymarket gives ~67% probability of a June 30 SpaceX listing). If Cerebras prices below the band or pulls the deal, the entire calendar re-rates. If it prices at or above the band, the OpenAI counterparty risk has been formally absorbed. Mid-May will tell which.

For now, the slip is the artifact. A roadshow extension during a tort docket and a criminal probe is not how a clean deal markets itself. The underwriter's silent revision is what this paper has been tracking under a different label since the customer-lender-shareholder triangle was first put in writing.

-- THEO KAPLAN, San Francisco