When Cerebras Systems priced 30 million shares at $185 each on Wednesday evening, the AI chip startup had already secured its place in Wall Street history: $5.55 billion in gross proceeds, the largest technology initial public offering since Uber went public in May 2019. By the time the Nasdaq closing bell rang on Thursday, the stock had blown past even the most optimistic whisper numbers, surging 89% to close near $350 under the ticker CBRS and pushing the company’s fully diluted market capitalization to roughly $95 billion.
For a company that carried a roughly $4 billion valuation in its last private funding round, the leap is extraordinary. It also sends an unmistakable signal: public-market investors are willing to write enormous checks for any credible challenger to Nvidia in the accelerating race to build AI infrastructure.
Inside the biggest tech debut in seven years
The deal terms, laid out in a final prospectus filed with the SEC, show Cerebras priced at the top of its marketed range. By gross proceeds, the $5.55 billion raise tops Arm Holdings’ roughly $4.87 billion offering in September 2023 and trails only Uber’s $8.1 billion raise.
The gap between those marquee listings tells its own story. Rising interest rates, bruising aftermarket losses for the IPO class of 2021, and a prolonged investor aversion to unprofitable growth companies kept the mega-cap IPO window mostly shut for years. Cerebras pried it back open by riding a spending cycle that shows no sign of slowing: Microsoft, Alphabet, Amazon, and Meta collectively committed more than $200 billion to data-center capital expenditures in 2025, according to their public earnings reports, and their 2026 budgets appear even larger.
The 89% first-day pop, reported by Bloomberg, ranks among the strongest opening sessions for a large-cap tech listing in recent memory. Uber, by contrast, slipped below its offer price on day one. A relatively tight initial float and heavy institutional demand likely amplified Thursday’s move, but the sheer breadth of buying interest suggests more than scarcity at work.
The company behind the chip
Cerebras was founded in 2016 by Andrew Feldman, a serial entrepreneur who previously sold his low-power server startup SeaMicro to AMD for $334 million in 2012. The company is built around a single audacious idea: instead of slicing a silicon wafer into hundreds of individual chips, use the entire wafer as one massive processor.
The result is the Wafer-Scale Engine, now in its third generation. The WSE-3 spans roughly 46,000 square millimeters of silicon and contains trillions of transistors. For perspective, Nvidia’s flagship B200 GPU measures about 800 square millimeters. That size advantage translates into raw throughput for training and running large language models, the workloads fueling the current AI boom.
Cerebras pairs the chip with its own cluster-scale systems and a software stack designed to let customers deploy models without re-engineering their code. Rather than selling standalone processors, the company delivers integrated solutions. Its pitch centers on speed and efficiency: Cerebras claims its hardware can train certain models in a fraction of the time required on comparable GPU clusters, while consuming less energy per operation. The company has also pushed into inference, the process of running trained models in production, with dedicated appliance configurations aimed at cloud and enterprise buyers.
The S-1 filing and earlier public disclosures name several customers and deployments that illustrate the breadth of adoption. The Mayo Clinic has used Cerebras systems for drug-discovery and medical-imaging workloads. AstraZeneca has deployed the hardware for molecular-simulation tasks in pharmaceutical research. The U.S. Department of Energy’s Argonne National Laboratory and Lawrence Livermore National Laboratory have both installed Cerebras clusters for scientific computing, and the United Arab Emirates’ technology investment arm, G42, has been disclosed as a significant buyer of Cerebras hardware for large-scale AI training.
Revenue is growing quickly but remains modest by semiconductor-industry standards. The S-1 registration statement filed in April 2026 disclosed that the company is still operating at a loss, a common profile for hardware startups investing heavily in R&D and manufacturing partnerships. Cerebras fabricates its wafers through TSMC, the same foundry that produces chips for Nvidia and Apple, which lends manufacturing credibility but also ties the company to a concentrated supply chain.
Where the risks live
The prospectus is blunt about customer concentration. A small number of large buyers, including hyperscale cloud operators and government-linked AI programs such as G42, account for a disproportionate share of revenue. If even one major account pulls back or decides to design its own silicon (as Google, Amazon, and Microsoft have all done to varying degrees), the financial hit could be severe. The filings flag this risk without disclosing the exact revenue split among top customers in the most recent period.
Geopolitics loom over the business as well. U.S. export controls on advanced semiconductors have tightened repeatedly since late 2022, and Cerebras’ risk disclosures acknowledge exposure to evolving restrictions. Any expansion of those rules could shrink the company’s addressable market overnight, particularly if sales to certain Middle Eastern or Asian buyers come under additional scrutiny.
Then there is the competitive reality. Nvidia controls an estimated 80%-plus share of the AI accelerator market, according to industry analysts, and is iterating on a roughly annual cadence. Its next-generation Rubin architecture, unveiled at Computex in 2024, is expected to begin shipping in 2026. AMD, Intel, and a growing roster of custom-chip efforts inside the hyperscalers are all chasing the same workloads. Cerebras’ wafer-scale approach is genuinely differentiated, but the IPO documents concede that rapid innovation cycles could erode any technical lead. Sustaining an edge requires continuous R&D spending at a level the company has never attempted as a public entity.
Capital deployment is another open question. The prospectus earmarks proceeds for research, manufacturing expansion, and general corporate purposes but does not commit specific dollar amounts to any single initiative. Investors are effectively trusting Feldman and his team to allocate $5.5 billion wisely in a market where the cost of being wrong on product timing can be existential.
What the $95 billion price tag assumes
At roughly $95 billion, Cerebras is valued at a level that prices in years of aggressive revenue growth and eventual profitability. The company provided no forward guidance in its filings, which means investors are building their own models from limited public data. Conventional valuation multiples are difficult to apply: the revenue base is small, margins are negative, and the AI hardware market is evolving so quickly that five-year forecasts are closer to speculation than analysis.
First-day pops in newly listed stocks can also be misleading. Lock-up expirations, typically 90 to 180 days after an IPO, will eventually release insider shares into the market and increase supply. Momentum-driven buying tends to fade as initial excitement cools and analysts publish their first independent estimates. Thursday’s 89% surge is a vote of confidence, but it was cast in a single session with constrained float, not over the quarters that will ultimately determine whether Cerebras can justify its valuation.
Why the IPO window swung open for wafer-scale silicon
Cerebras has reopened the market for large-scale tech IPOs at a moment when AI spending is reshaping how capital flows through Silicon Valley and Wall Street alike. The company raised more money in one day than most semiconductor startups raise in a lifetime, and it did so at a valuation that would have seemed fantastical even 18 months ago. Whether it can grow into that price tag will depend on execution against Nvidia, the durability of the current AI infrastructure cycle, and policy decisions in Washington and Beijing that are still being written. The IPO gave Cerebras the capital. Now the clock is running.
