In 2018, Elon Musk tried to buy Cerebras Systems, but the founders, Andrew Feldman, Gary Lauterbach, Michael James, Sean Lie, and Jean-Philippe Fricker, said no. Back then, some saw this as a bold move, while others thought it was risky. The company had already spent three years and a lot of money on a chip design that most in the industry thought couldn’t be built.
Last week, Cerebras filed to go public on Nasdaq under the ticker CBRS. Now, according to CNBC, the founders’ decision seems much more justified.
According to its IPO filing, Cerebras reported $510 million in revenue for 2025, a 76% increase from the previous year. The company also went from a $481 million net loss to $87.9 million in net income. Morgan Stanley, Citigroup, Barclays, and UBS are leading the underwriting, while Mizuho and TD Cowen are bookrunners. The share price and offering size have not yet been announced.
The company has also shared that it has a $24.6 billion backlog in remaining performance obligations as of December 31.
Most AI computing today uses clusters of thousands of small GPU chips, each about the size of a fingernail, connected in large, power-hungry data centres. Now that AI inference is becoming more important than training, these limitations are getting harder and more costly to overlook. Cerebras chose a completely different approach.
Its Wafer-Scale Engine is the first and only commercial processor to use the entire wafer. It measures 46,225 square millimetres, contains 4 trillion transistors, and 900,000 cores. Putting everything on a single surface eliminates the need for complex interconnections between chips.
The result is a memory bandwidth 7,000 times that of a standard GPU and inference speeds that have drawn customers ranging from Argonne National Laboratory to GlaxoSmithKline. The current CS-3 system integrates the WSE with power, cooling, and networking into a single deployable unit.
NVIDIA leads the competition, with its GPUs powering most of the world’s AI infrastructure and a market value of over $4.5 trillion in early 2026. AMD and Broadcom are strong competitors, and more startups focused on inference are entering the field.
Cerebras’ situation is unusual because AMD, a direct competitor, joined its most recent funding round. When a rival invests in your Series H, it can mean they want to protect themselves from disruption, believe in the technology, or both.
That Series H, closed in February 2026 and led by Tiger Global, valued Cerebras at $23 billion, nearly triple the $8.1 billion valuation it carried just five months earlier after its Series G. Benchmark Capital, which has backed the company since its $27 million Series A in 2016, raised a dedicated $225 million SPV to increase its position. Total private capital raised across all rounds now stands at approximately $2.8 billion.
This IPO comes as many AI companies are testing whether the public market is interested in the infrastructure behind the AI boom, with Cerebras being the first to file. The response to its debut will likely influence how future AI hardware IPOs are received.
