Cerebras Systems IPO: $3.5B Raise Targets a $28B Valuation

Cerebras Systems, the startup that built the world’s largest computer chip for AI workloads, is targeting a $3.5 billion raise in a Nasdaq IPO scheduled for May 14, 2026 — the largest technology public offering of the year. After more than $10 billion in orders flooded in for the $3.5 billion deal, the company raised its price range from $115–$125 to $125–$135 per share, implying a valuation approaching $28 billion. For a company that nearly saw its IPO derailed by a national-security review two years ago, the turnaround is remarkable.

The Chip That Started at Wafer Scale

Cerebras was founded in 2015 around a single contrarian premise: instead of linking thousands of small GPUs to train AI models, build one enormous chip — the entire silicon wafer — and eliminate the communication bottleneck between chips entirely. The result is the Wafer-Scale Engine 3 (WSE-3), a single-die chip containing approximately 900,000 AI-optimized processing cores. For comparison, NVIDIA’s H100 packs around 80 billion transistors on a chip the size of a palm; the WSE-3 covers an entire wafer.

Cerebras ships the chip inside the CS-3, a 1.8-ton computing appliance, and also sells access via a cloud service. That recurring cloud layer is central to its financial story — gross margins on its cloud business run roughly 41%, and the company has been pulling forward enterprise contracts at an accelerating rate.

A Revenue Ramp Few Pre-IPO Companies Match

Cerebras is not a pre-revenue concept. The company has been scaling commercially for four consecutive years, with each annual period representing a step-change in scale:

Fiscal Year Revenue YoY Growth Net Income / (Loss)
FY2022 $24.6M
FY2023 $78.7M +220%
FY2024 $290.3M +269% ($485M)
FY2025 ~$510M +76% $87.9M
Source: Cerebras Systems S-1 prospectus (filed April 17, 2026) via SiliconAngle and Capital.com.

The swing from a $485 million net loss in FY2024 to a $87.9 million GAAP profit in FY2025 is the headline. The company attributes much of that improvement to non-cash warrant income related to the OpenAI partnership and expanding cloud margins. The trajectory — $25M to $510M in three years — is among the steepest revenue ramps of any technology company heading into an IPO.

Cerebras Systems Annual Revenue FY2022–FY2025 Bar chart showing Cerebras revenue growth from $24.6M in FY2022 to approximately $510M in FY2025. $0 $250M $510M FY2022 $24.6M FY2023 $78.7M FY2024 $290.3M FY2025 ~$510M
Cerebras Systems annual revenue (FY2022–FY2025). Source: S-1 prospectus, April 2026.

The OpenAI Anchor — and a Concentrated Customer Base

OpenAI is far more than a customer for Cerebras. In a structure unusual even by startup standards, OpenAI extended Cerebras a $1 billion loan secured by warrants for more than 33 million shares alongside a multi-year compute agreement worth more than $10 billion through approximately 2030. Sam Altman, OpenAI’s CEO, is listed as an angel investor in the S-1.

The relationship is lucrative — but it illustrates the company’s central risk. According to the prospectus, the two largest customers combined represented approximately 86% of FY2025 revenue. The second anchor customer is MBZUAI, an Abu Dhabi-linked AI institute. The concentration means a single contract renegotiation could materially reshape the income statement.

The CFIUS Detour

Cerebras’ path to Wall Street was not straightforward. The company first attempted to go public in late 2024, but the Committee on Foreign Investment in the United States (CFIUS) opened a review of an investment in Cerebras held by G42, an Abu Dhabi-based technology group. The review forced Cerebras to pause its IPO roadshow, restructure its shareholder base, and remove G42 from the capitalization table before refiling.

CFIUS cleared the revised structure in March 2025. The public S-1 was filed on April 17, 2026. The episode illustrates how AI semiconductor companies straddling U.S.-Gulf partnerships must navigate regulatory scrutiny — a dynamic that remains relevant given the ongoing MBZUAI customer relationship.

The Valuation Debate: 55× Revenue or a Backlog Play?

At a $130 per share midpoint — the center of the raised $125–$135 range — Cerebras would trade at roughly 55× trailing revenue, a multiple that towers above traditional chip companies. NVIDIA itself trades at around 20–25× forward revenue; Arm Holdings priced its 2023 IPO at roughly 25× trailing sales.

The bull case rests on Cerebras’ reported contract backlog of approximately $24.6 billion in committed future revenue, a figure that dwarfs the current revenue run-rate and argues that today’s valuation is not as stretched as trailing multiples suggest. If that backlog converts on schedule, the IPO price could look reasonable in hindsight.

The bear case: nearly all of that backlog flows through two customers. Any deterioration in the OpenAI or MBZUAI relationships — whether from competitive alternatives, geopolitical friction, or customer-specific funding constraints — could compress the backlog and reprice the stock sharply.

Deal Terms and Capital History

Cerebras last raised private capital in February 2026: a $1 billion Series H at a $23 billion pre-money valuation. Before that, a September 2025 Series G raised $1.1 billion at an $8.1 billion post-money valuation — meaning the implied enterprise value tripled in under a year, driven almost entirely by the deepening OpenAI partnership.

Morgan Stanley is leading the IPO, with Citigroup, Barclays, UBS, Mizuho, and TD Cowen as co-underwriters. Underwriters hold an option on up to approximately 4.2 million additional shares (the greenshoe), which could push total proceeds toward $4 billion if exercised in full. The 28 million primary shares on offer represent a relatively modest float given the company’s scale, a structure that could amplify early trading volatility.

The Bigger Picture: AI Infrastructure Comes of Age

Cerebras’ IPO arrives in an AI infrastructure cycle running near full throttle. Hyperscalers — Amazon, Microsoft, Google, and Meta — committed a combined $300 billion-plus in capital expenditure in 2025, a significant portion directed at AI compute. NVIDIA remains the dominant GPU supplier, but a cohort of challengers including Cerebras, Groq, and SambaNova are pursuing workloads where chip-to-chip latency or power efficiency matters more than raw training throughput.

Whether the public market will sustain a $28 billion valuation for an AI chip startup that derives most of its revenue from two customers remains the defining question. The $10 billion in institutional orders that flooded into the bookbuild suggests money managers are, for now, betting on yes.

Sources

Disclosure: This article was produced with AI assistance and reviewed before publication. It is for informational purposes only and is not investment advice.

Leave a Comment