Cerebras Cleared the IPO. It Did Not Clear the G42 Question.
Cerebras went public on May 14 at a valuation that briefly touched twelve figures. The market read it as a triumph. I want to flag the part the tape did not price, because it is the same part that postponed this exact offering in 2024: the G42 question. It did not get answered. It got restructured.
The order of events matters here, so let me walk it precisely. The superlatives belong in the markets coverage, which Marcus Chen has in the read on the $95 billion close. This is the policy account.
What Actually Happened in 2024
G42 is an Abu Dhabi AI conglomerate. It is not a passive name on Cerebras's cap table. It has been simultaneously the company's largest customer and a direct investor, an arrangement that agreed to put roughly $335 million into Cerebras stock for a stake above 5 percent.
That dual role is what triggered review by the Committee on Foreign Investment in the United States. CFIUS screens inbound foreign investment for national-security risk. By the reporting in October 2024, the review's delay forced Cerebras to postpone its IPO and call off the roadshow. The specific concern, raised loudly by China hawks in Washington, was that Gulf intermediaries could become a conduit for advanced US AI technology to reach China, which is blocked from buying it directly, and that G42's historical China ties made that more than hypothetical.
How the 2026 Listing Got Through
Not by resolving the question. By re-shaping the transaction around it. The companies amended the filing so that the shares G42 acquired would be non-voting securities, and argued on that basis that the investment should not require a CFIUS review at all. They then submitted a request to withdraw the notice, and by the filing CFIUS was still considering that withdrawal.
Non-voting shares change the governance surface. They do not change the commercial dependence. And the commercial dependence is the part that survived into the 2026 prospectus intact: roughly 86 percent of Cerebras revenue still comes from two UAE-based entities, with G42 alone accounting for about 87 percent of revenue in the first half of 2024. The structure addressed the form of the concern. The substance of it, an American frontier-silicon company whose revenue base sits overwhelmingly in one Gulf relationship, is disclosed in the S-1 as a risk factor, not retired as one.
Why This Is Bigger Than One Company
I am not arguing Cerebras did anything improper. Restructuring around a review is legal and ordinary. The point is what it establishes as the template. Every AI-silicon company eyeing the 2026 IPO window now has a worked example of how to get a foreign-capital and customer-concentration problem past the gate: convert the equity to non-voting, withdraw the notice, disclose the concentration as risk, price the deal anyway.
That is a structural tax, not a one-off. It sits on top of the export controls already governing where advanced accelerators can ship, and it compounds with the policy floor that has been rising all spring. We tracked that escalation in the week the US AI policy floor moved: CAISI pre-launch evaluation agreements, a White House studying an FDA-style order for model releases, and China formally blocking Meta's Manus acquisition. Inbound capital review is the hardware-side counterpart to all of that. The agent and infrastructure buildout we map on the AI infrastructure tracker now runs through a national-security filter at the financing layer, not just the export layer.
The hardware itself is not the issue here, and the engineering case stands on its own merits in Ripper's wafer-scale teardown. The point is narrower and it is about capital: who funds frontier compute, from which jurisdiction, and what review attaches when they do.
Signposts I Am Watching
Three. First, the disposition of the withdrawn CFIUS notice. A withdrawal under consideration is not a clearance, and a re-filed or re-opened review post-listing would be a public-market event, not a private one. Second, the OpenAI ramp. The forward story diversifies revenue away from the Gulf concentration only if that capacity actually converts on schedule, which is also the financial question on the funding portfolio tracker. Third, the next AI-silicon S-1. The reporting frames Cerebras as the first of several. The second one's risk section will tell us whether the non-voting-plus-withdrawal pattern is now standard practice.
Our Take
The market priced the upside on Thursday and discounted the geopolitical risk to roughly zero. The 10 percent fade on Friday was partly that risk being read back into the number. My view is direct: the national-security exposure here is real, unresolved on the merits, and structurally durable, because the fix is not a filing amendment, it is years of revenue diversification away from a concentrated non-allied base.
I do not think that sinks Cerebras. I think it means CFIUS and export posture are now a recurring repricing trigger for the entire AI-hardware listing class, the way rate decisions are for banks. The companies that go public after Cerebras will be measured against this template, and the smart ones will diversify the customer base before the roadshow, not disclose the concentration during it. We will track each new AI-silicon filing against that bar on /today. The IPO cleared. The question did not.