$510 million in annual revenue. $95 billion in market capitalization. The ratio is either the most forward-looking bet in semiconductor history or the clearest signal yet that AI fever has detached from fundamentals.

Cerebras Systems closed its first day of Nasdaq trading Thursday at $311.07, a 68% surge from its $185 offering price that valued the AI chipmaker at roughly $95 billion. The offering raised $5.55 billion across 30 million shares — the largest US tech IPO since Uber in 2019. Underwriters Morgan Stanley, Citigroup, Barclays, and UBS hold options on 4.5 million additional shares that could push total proceeds to $6.38 billion.

The valuation demands a double take. At roughly 187 times trailing revenue, Cerebras trades at a multiple that dwarfs most tech debuts. Alibaba generated $5.5 billion in revenue before its 2014 IPO and closed day one at $231 billion. Cerebras brought in less than a tenth of that — and landed at $95 billion.

A different kind of chip

Cerebras’ wager, a decade in the making, is that bigger is better. Its wafer-scale engine uses an entire silicon wafer rather than the small die cuts standard fabrication produces. The WSE-3 packs 4 trillion transistors and 900,000 AI cores onto a single 46,000-square-millimeter die on TSMC’s 5-nanometer process — 58 times larger than Nvidia’s B200, according to the company’s registration statement.

The architectural advantage centers on inference. By keeping compute and memory on one wafer, Cerebras eliminates the inter-chip communication latency that bogs down GPU clusters during model queries. The company claims its chips run inference 15 times faster than competing GPUs on open-source models. The trade-off is rigidity: a dinner plate-sized processor is harder to redeploy than a fleet of interchangeable GPUs.

The market isn’t pricing what Cerebras has done. It’s pricing what OpenAI and Amazon might help it do next.

From UAE outcast to Wall Street darling

Cerebras’ road to the public markets was anything but smooth. The company first filed to go public in September 2024, then withdrew after the Committee on Foreign Investment in the United States opened a national security review of its partnership with G42, a UAE-based AI firm that accounted for more than 80% of revenue. CFIUS eventually cleared the arrangement. Cerebras refiled in April 2026 with a reshaped customer base.

But concentration hasn’t disappeared — it’s relocated. A UAE university accounted for 62% of 2025 revenue, with G42 contributing another 24%. In January, Cerebras signed a multiyear cloud deal with OpenAI worth more than $20 billion, covering 750 megawatts of compute through 2028. In March, AWS announced it would deploy Cerebras systems in its data centers. Both companies hold warrants to purchase Cerebras stock.

CEO Andrew Feldman, who co-founded Cerebras in 2016 and holds a stake worth close to $2 billion at the IPO price, told CNBC that large single customers are “one of the characteristics of this market.”

The drought breaks

Cerebras arrives after one of the longest tech IPO dry spells in memory. Just 31 technology companies went public in 2025, down from 121 four years earlier, according to University of Florida IPO expert Jay Ritter. The broader semiconductor sector has roared back — Intel, AMD, and Micron have all posted triple-digit gains in 2026, and the VanEck Semiconductor ETF is up 58% year to date.

The offering was more than twenty times oversubscribed, Reuters reported. Underwriters raised the price range from $115–$125 to $150–$160 before ultimately pricing at $185 — well above the original band.

The $95 billion question

Cerebras swung to net income of $88 million in 2025 from a $481.6 million loss the prior year, but the profit flatters. GAAP operating losses continued, according to the company’s filings, and net income was boosted by non-cash accounting items.

The structural risks are plain. TSMC manufactures every Cerebras wafer with no guaranteed long-term allocation — and Cerebras is a small customer next to Nvidia, AMD, and Apple. Nvidia spent more than $18 billion on R&D last fiscal year and acquired assets from inference startup Groq for $20 billion in December. The CUDA ecosystem, with a decade of developer tooling, is not a moat hardware alone can dismantle. Lock-up provisions restricting insider sales expire in the coming months.

Whether this is justified confidence or late-cycle exuberance depends on one belief: that the AI infrastructure build-out is still in its early chapters. The pipeline of upcoming debuts — SpaceX, potentially OpenAI and Anthropic — suggests Wall Street is all in on that thesis. History says IPOs of this scale, at these multiples, tend to be followed by a reckoning.

Both things can be true.

Sources