Cerebras Systems closed the underwriting on its initial public offering at $6.38 billion in gross proceeds, the underwriters' 4.5 million-share greenshoe fully exercised, on a base of 34.5 million shares priced at $185. [1] The closing added roughly $730 million in net proceeds beyond the May 14 opening. Six trading days past the open, the company sits at a $26.6 billion market capitalization, a $48.8 billion fully diluted valuation, and a sales multiple of about 96 times its trailing twelve-month revenue. The number that is not in the headline is the one on page sixty-eight of the S-1. A single private-company customer — OpenAI — has committed to 750 megawatts of Cerebras compute through 2028, valued in the filing at "more than $20 billion." [2]

The paper's Friday brief on the day-seven silence named the structural read. No sell-side firm has published a customer-concentration downgrade. No second hyperscaler-class customer has been named. The Seeking Alpha note from May 18 with the $275 price target acknowledged the risk and stopped short of formal action. Tuesday's first post-Memorial-Day trading window is the next opportunity. Sunday's increment is the full-greenshoe close. The $730 million in additional proceeds is the price the syndicate was willing to pay for a deal whose growth thesis is a single named counterparty. The math gets tighter when the counterparty is read out.

OpenAI is the customer. Sam Altman's company entered into a master research agreement with Cerebras in March, disclosed in the S-1 amendment of May 4, that commits 750 megawatts of capacity through the end of 2028. [2] The "more than $20 billion" figure is the contract value. Cerebras's trailing twelve-month revenue at the time of the filing was roughly $278 million. The OpenAI commitment is therefore approximately seventy-two times the company's annual revenue at the moment of the IPO. The contract delivers in stages and depends on Cerebras shipping at the volume the agreement assumes. The S-1 risk factor names the dependency in plain language: "the loss of this customer, or a material reduction in its purchase commitments, would have a material adverse effect on our business." [2]

This is not the customer concentration the market was warned about in May. The S-1 the market read at IPO — disclosed before the May 4 amendment — was the one in which MBZUAI accounted for 62 percent of fiscal year 2025 revenue and G42 for 24 percent, with 77.9 percent of year-end accounts receivable from MBZUAI alone. [3] That table is what produced the May 16 ten-percent slide. The May 4 amendment added OpenAI on top of the Emirati concentration. The company's growth thesis is now diversified from one buyer to two buyers. The two buyers together account for the entirety of the structural argument the IPO sold. There is no third name.

The Nvidia framing the company used in its road show is the substantive bet. Cerebras's pitch was that wafer-scale architecture would compress training and inference cost per token at a rate Nvidia could not match, and that the substitution between H100/H200-class GPUs and the WSE-3 wafer would happen at the model-training stage before it happened at inference. The 96x sales multiple at the May 22 close prices that bet at a ceiling. For comparison, Nvidia closed Friday at roughly 25x trailing sales after the Wednesday $82 billion print. [4] Cerebras is trading at nearly four times Nvidia's revenue multiple on revenue that depends on two customers, neither of whom is a hyperscaler. The market's verdict is that the OpenAI commitment is the diversification.

The closing of the greenshoe is also a procedural item. Underwriters exercise the greenshoe within thirty days of the offering at the offering price; the full exercise here, at $185, means the syndicate priced the deal cleanly and the demand book was deep enough to absorb the additional 4.5 million shares without secondary pressure. The fee structure on the greenshoe is the same as the base — roughly 6 percent — which makes the additional gross of $832.5 million translate to net proceeds of approximately $730 million on the supplementary tranche alone. The company is now sitting on a cash position the S-1 projected at roughly $1.6 billion post-IPO. The use-of-proceeds language names "general corporate purposes" and "expansion of manufacturing capacity." Neither phrase mentions OpenAI by name.

What Tuesday will test is whether the market reads the greenshoe close as confidence or as completion. A clean greenshoe on a hot IPO usually closes the chapter — the bankers have placed the shares they were authorized to place, the float is the float, and the price discovery moves to the secondary market. The OpenAI table sits in the company's filings as the structural ceiling on what the secondary market will pay. A second hyperscaler — Microsoft, Google, Amazon, Meta, Oracle — naming a Cerebras compute commitment of comparable scale is the thing that would unlock multiple expansion. The absence of such an announcement in the post-IPO window is the absence of the second leg.

The OpenAI dependency is itself dependent on OpenAI's own cap-stack. The lab whose 750-megawatt commitment is the largest external compute commitment on Cerebras's books is also the lab whose own infrastructure is now contracted across Microsoft, Oracle, Crusoe, and CoreWeave at scale. The May 4 S-1 amendment did not disclose any exclusivity. OpenAI is buying compute from Cerebras because Cerebras has shipped capacity OpenAI wants to use. OpenAI is also buying compute from everyone else who has capacity. The structural read of the Cerebras IPO at $185 is that the largest customer is a customer in motion. Tuesday's tape begins after a long weekend during which Cerebras has had no opportunity to add a name and no reason to take one off.

-- THEO KAPLAN, San Francisco