- Cerebras filed its S-1 with the SEC on April 17, targeting a Nasdaq listing under ticker CBRS at a reported $35 billion valuation.
- The company posted $510 million in 2025 revenue, a 76% year-over-year surge, fueled by its wafer-scale AI chip technology.
- A $20 billion multiyear deal with OpenAI for 750 megawatts of Cerebras systems anchors the company’s forward pipeline.
- New contracts with AWS and Meta signal customer diversification away from previous G42 concentration risk.
The AI chip wars just got a serious new contender on Wall Street. Cerebras Systems, the company that bets on building entire processors out of single silicon wafers, has filed for an IPO that could value it at $35 billion. After a failed 2024 attempt derailed by a federal review, the company is back with a transformed customer book, a blockbuster OpenAI contract, and revenue growth that venture capitalists are calling the strongest in enterprise AI hardware.
The IPO Filing: Second Time’s the Charm
Why 2024 Failed and 2026 Works
Cerebras first attempted to go public in late 2024, but withdrew after the Committee on Foreign Investment in the United States (CFIUS) launched a review of its relationship with Abu Dhabi’s Group 42 (G42), then the company’s largest customer. That single-customer dependency spooked regulators and investors alike.
Fast forward to April 2026, and the picture looks radically different. The S-1 filed on April 17 reveals a diversified customer base headlined by OpenAI, Amazon Web Services, and Meta Platforms. Morgan Stanley, Citigroup, Barclays, and UBS are leading the offering as joint underwriters, with a mid-May listing window on the Nasdaq under ticker CBRS.
The company raised $1 billion in its Series H round just two months earlier at a $23 billion valuation. Now it’s targeting $35 billion in the public markets, a 52% premium that reflects the velocity of its recent deal wins.
Business Insight — The 18-month gap between the failed and revived IPO gave Cerebras time to solve its biggest weakness: customer concentration. Enterprises evaluating AI chip vendors should watch whether this diversification holds post-IPO, or whether OpenAI’s outsized deal recreates the same single-customer risk G42 once posed.
Wafer-Scale Technology: Betting Against the Chip Industry’s Playbook
One Wafer, One Chip
While NVIDIA, AMD, and every other chipmaker cut silicon wafers into hundreds of small processors, Cerebras does the opposite. Its WSE-3 (Wafer Scale Engine 3) uses an entire 300mm wafer as a single, massive chip. The result is a processor with 4 trillion transistors, roughly 50 times more than NVIDIA’s H100 GPU, purpose-built for AI training and inference workloads.
The approach eliminates the communication bottlenecks that plague multi-GPU clusters. Instead of networking thousands of individual GPUs together, a single CS-3 system can hold an entire large language model in on-chip memory, dramatically reducing latency for inference tasks.
The Inference Angle
CEO Andrew Feldman made a pointed claim in the filing: Cerebras captured “fast inference business at OpenAI” directly from NVIDIA. As AI companies shift spending from training to inference (running models for end users), Cerebras positions its architecture as uniquely suited for the low-latency, high-throughput demands of production AI services.
Business Insight — The AI industry’s center of gravity is moving from training to inference. If Cerebras truly won OpenAI’s inference workloads from NVIDIA, it signals that the “NVIDIA is the only game in town” narrative has cracks. Enterprises planning large-scale AI deployments should benchmark wafer-scale alternatives before locking into multi-year GPU procurement contracts.
The Numbers: $510M Revenue, 76% Growth, and a $20B Pipeline
Revenue Breakdown
Cerebras reported $510 million in 2025 revenue, up 76% year-over-year. Hardware sales grew 69%, while cloud and services revenue nearly doubled at 99% growth. The company posted $237.8 million in GAAP net income, though this was inflated by non-operational gains; on a non-GAAP basis, Cerebras recorded a $75.7 million net loss.
Operating cash flow came in at negative $10.1 million, nearly breakeven, suggesting the company is not burning through cash reserves at an alarming rate. R&D spending consumed 48% of revenue, a figure that reflects the capital intensity of building custom silicon at wafer scale.
The OpenAI Anchor
The headline contract is a multiyear agreement signed in January 2026 committing OpenAI to deploy 750 megawatts of Cerebras systems. Reports peg the total deal value at $20 billion over three years, a figure that would dwarf Cerebras’s current annual revenue by nearly 40x. Additionally, AWS has signed on to deploy Cerebras chips in its data centers, giving the startup access to Amazon’s massive cloud customer base.
Business Insight — A $20B pipeline against $510M in current revenue creates a compelling growth story, but investors should scrutinize delivery timelines and ramp schedules. Manufacturing wafer-scale chips at the volumes OpenAI requires is an unproven feat, and any supply chain hiccup could turn this pipeline into a liability.
Risks and the Road Ahead
What Could Go Wrong
For all the momentum, Cerebras faces real headwinds. The OpenAI contract, while transformative, risks recreating the customer concentration problem that torpedoed the 2024 IPO. NVIDIA remains dominant with a proven ecosystem, CUDA software moat, and rapidly improving inference performance with its Blackwell architecture. And manufacturing wafer-scale processors at volume has never been done at the scale OpenAI’s contract demands.
There’s also the macro question: AI chip spending is at record highs, with the “Magnificent Seven” tech companies pouring $78 billion into AI infrastructure in Q1 2026 alone. If that spending cycle cools, the entire AI hardware supply chain, including Cerebras, could face a sharp correction.
Business Insight — Cerebras represents a genuine architectural bet against NVIDIA’s GPU-centric model. For investors, the question is not whether wafer-scale technology works (it clearly does), but whether it can scale manufacturing fast enough to justify a $35B valuation before NVIDIA closes the inference performance gap.
Related
- OpenAI Takes Aim at Anthropic With Beefed-Up Codex Desktop App
- Q1 2026 Shatters Venture Funding Records as AI Boom Pushes Investment to $300B
- AI Startup Cursor in Talks to Raise $2B at $50B Valuation
- OpenAI Raises $122 Billion to Accelerate the Next Phase of AI
- Big Tech AI Spending: The $300B+ Capex Race in 2026
Sources
- TechCrunch — AI chip startup Cerebras files for IPO (April 18, 2026)
- The Motley Fool — Nvidia Rival Cerebras Files for an IPO: What Investors Should Know (April 20, 2026)
- Bloomberg — AI Chipmaker Cerebras Systems Files Publicly Again for US IPO (April 17, 2026)
AI Biz Insider · AI Business EN · aibizinsider.com
댓글 남기기