Cerebras Is Asking IPO Buyers to Finance an AI Capacity Buildout

NEW YORK, May 11, 2026 - Cerebras is no longer just an AI-chip curiosity with a dramatic roadshow. It is shaping up as a public-market financing event for a capital-hungry inference buildout, and that makes the deal more interesting than the usual semiconductor-adjacent IPO. In its May 4 amended S-1, Cerebras put forward 28 million shares at $115 to $125, with Nasdaq trading under CBRS. The live question now is whether the book has already outrun those terms.

That distinction matters. IPOGrid reads this one less as a simple AI hardware float and more as a test of how much public investors will finance capacity before the income statement fully de-risks. Cerebras has already lined up substantial private and credit support. The company said in February that it closed a $1 billion Series H at an approximately $23 billion post-money valuation, and in April it announced an $850 million revolving credit facility. The same April release is worth lingering on because the arranging banks - Morgan Stanley, Citi, Barclays, UBS, Credit Agricole CIB, MUFG, Mizuho, TD Securities and SVB/First Citizens - overlap heavily with the IPO syndicate. That gives the book a more institutional shape than most AI issuers get on first launch.

The commercial story is also real. OpenAI said in January it is partnering with Cerebras to add 750 megawatts of ultra-low-latency AI compute, with capacity coming online in multiple tranches through 2028. Cerebras's own January post goes further, saying full warrant vesting would require OpenAI to exercise options that bring purchases to 2 gigawatts of AI inference capacity. On another front, AWS and Cerebras said in March that Cerebras systems would be deployed in AWS data centers and accessed through Amazon Bedrock. For an IPO buyer, those are not cosmetic logos. They are evidence that Cerebras has escaped the lab and entered the platform stack discussion.

But the filing also shows why this cannot be valued like a clean software infrastructure name. The April S-1 laid out a business that grew fast enough to command attention: 2025 revenue of $510.0 million, up from $290.3 million in 2024, split between hardware and cloud and other services. The later amended filing points to $244.9 million of 2025 pretax income versus a $479.7 million pretax loss a year earlier, plus $701.7 million of cash and cash equivalents and $824.1 million of working capital. Still, our interpretation is that investors should resist reading that swing as proof the model is already mature. The April filing also highlights a $96.1 million non-GAAP operating loss for 2025, heavy infrastructure spending including $382.7 million of 2025 property-and-equipment purchases tied primarily to Cerebras Cloud services, and large working-capital swings driven by deposits, receivables and inventory.

The concentration profile is another reason to keep the analysis hard-edged. The April filing says MBZUAI accounted for 77.9% of accounts receivable at December 31, 2025, while G42 accounted for 91.0% a year earlier. The same filing discloses a working-capital loan of approximately $1.0 billion from OpenAI and an OpenAI warrant covering up to 33,445,026 Class N shares at a nominal exercise price. None of that invalidates the demand story. It does mean the structure appears to us less like a diversified public-company revenue base and more like a rapidly scaling AI financing web wrapped around a handful of strategic counterparties.

Governance will not rescue minority holders if the cycle turns. Cerebras's amended prospectus says its Class B stock carries 20 votes per share, and that after the offering Class B holders would control roughly 99.2% of the voting power. That is a perfectly common late-stage tech move, but it changes the bargain. Public investors are effectively underwriting capacity expansion and liquidity while accepting near-total founder and insider control.

There is also the simple fact that this is a return engagement. In an October 3, 2025 withdrawal request, Cerebras told the SEC it did not intend to conduct the previously proposed offering at that time. The company is back now with a hotter AI tape, fresh financing, stronger partnerships and what appears to be a much more aggressive order book. That is exactly why the deal deserves attention today.

The bullish case is straightforward: Cerebras has authentic product differentiation, real hyperscaler and model-lab relationships, a deep bank group and enough demand to push terms upward before pricing. The reviewer’s concern is that the market may be tempted to pay for inevitability when the economics still look like those of a company racing to secure compute, customers and data-center footprint at once. Use of proceeds in the filed deal remains broad - working capital, R&D and general corporate purposes - which is exactly what a company says when it needs flexibility more than precision.

If EDGAR reflects the higher reported terms before pricing, Cerebras could become one of the clearest public tests yet of whether AI enthusiasm now supports not just premium multiples, but premium financing for unfinished hardware infrastructure stories.