Cerebras Systems is racing toward a Nasdaq listing in the second quarter of this year, targeting $2 billion in fundraising with a valuation between $22 billion and $25 billion.
If successful, it would be the first pure-play AI chip alternative to debut on the public market since Nvidia's GPU monopoly took hold.
What Does It Sell?
Cerebras's core technology is the Wafer Scale Engine 3 (WSE-3) — literally a wafer-scale AI processor.
It's not about specs on paper; it's physically massive: the WSE-3 occupies an entire silicon wafer, making it 56 times larger than Nvidia's biggest GPU. It packs 900,000 AI compute cores and 4 trillion transistors.
Why build something this big?
GPU clusters are slow largely because of the time spent transferring data between chips — with 10 H100s, those chips constantly send data back and forth, incurring significant communication overhead. The WSE-3 solves this problem at its root, delivering inference speeds more than 20 times faster than GPU systems, with lower power consumption per unit of compute.
Is the Business Proven?
One landmark contract: OpenAI signed a compute supply deal with Cerebras worth over $10 billion, deploying 750 megawatts of Cerebras inference infrastructure — the world's largest high-speed AI inference cluster. OpenAI's inference speed on this system is 15 times faster than on its original GPUs.
A $1 billion Series H round closed in February 2026, with investors including Tiger Global, Benchmark, AMD, and Fidelity, at a $23 billion valuation.
Other customers include IBM, Meta, and Mistral AI.
A Rocky Road to IPO
This isn't Cerebras's first attempt at going public.
In October 2025, just days after announcing an $1.1 billion Series G round, it suddenly withdrew its IPO application. The reason: the U.S. CFIUS (Committee on Foreign Investment in the United States) was reviewing the company's investor structure, specifically the equity stake held by UAE-based firm G42 in Cerebras.
G42 is a top Middle Eastern tech investment firm, but it also has deep ties with Chinese tech companies — a sensitive issue in the eyes of the U.S. government.
The solution was straightforward but telling: convert all of G42's equity into non-voting shares, removing G42 from corporate governance while preserving its financial interest. CFIUS cleared the deal, and the IPO moved forward.
This episode is a microcosm of a larger reality: by 2026, AI infrastructure is not just a business issue — it's a geopolitical one. The investor structure of a U.S. AI chip company can stall an IPO for the better part of a year.
Why This Matters
Nvidia's CUDA ecosystem remains the deepest moat in the entire AI industry. All major AI frameworks support CUDA by default, and training and inference infrastructure heavily relies on H100/H200.
Cerebras is taking a different path from Nvidia — it's not trying to compete on the same turf, but differentiating on inference speed and energy efficiency.
After the IPO, several questions are worth watching:
- How will institutional investors value it? Can the $23 billion private valuation hold up in the public market?
- How will it address the training gap? Cerebras has a clear advantage in inference, but AI companies' capital expenditure is still focused on training.
- Could the OpenAI contract become a risk of over-reliance on a single customer?
As the first company to test the "non-CUDA route" on the public market, whatever the outcome, this story alone is worth following for some time.
Sources: CocoLoop, Nvidia Rival Cerebras Raises $1 Billion in Funding at $23 Billion Valuation (Bloomberg); Cerebras Systems Raises $1 Billion Series H (Cerebras.ai press release); Cerebras Targets April AI Chip IPO to Raise $2 Billion (AI CERTs News); AI chipmaker Cerebras targets Q2 2026 for IPO launch (Seeking Alpha)