A wave of expected initial public offerings of artificial intelligence companies started with a bang on Thursday.
Cerebras, a Silicon Valley maker of A.I. chips, opened trading of its shares on the stock market at $350, up 89 percent from an I.P.O. price of $185, valuing the company at $75 billion. In the week leading up to its market debut, Cerebras had lifted its offering price twice from a preliminary $115 and increased the number of shares it made available to investors, raising at least $5.6 billion for itself.
That made Cerebras the largest public offering so far this year and the biggest tech debut globally since 2019, said Matt Kennedy, a senior strategist at Renaissance Capital, which follows I.P.O.s.
“The A.I. I.P.O. boom is really starting to happen now,” he said.
Cerebras presages a series of potential “mega I.P.O.s” from A.I.-related firms including SpaceX, OpenAI and Anthropic. SpaceX, Elon Musk’s rocket maker, which owns his A.I. initiatives, has valued itself at more than $1 trillion and could go public as soon as next month. OpenAI and Anthropic, which have developed foundational A.I. models and tools like chatbots, are also eagerly anticipated by investors. All of them could be among the biggest I.P.O.s to date.
These companies would reach the stock market amid a frenzy over A.I., which is transforming everything from software coding to geopolitics. Tech giants including Google, Meta and Microsoft are pouring billions into building data centers to power A.I. development and some are working with OpenAI and Anthropic to win the technology contest.
The mania has catapulted Nvidia, which is the biggest maker of A.I. chips, into position as the world’s most valuable public company. Cerebras is among the companies trying to challenge Nvidia’s dominance.
Based in Sunnyvale, Calif., Cerebras makes advanced computer chips that can be used to train A.I. models. It is particularly known for what it claims is the largest computer chip ever built — as big as a dinner plate, about 100 times the size of a typical chip — which can be used in data centers to help accelerate the progress of A.I.
“We are building the fastest A.I. infrastructure in the world,” the company said in its offering prospectus.
Cerebras, which is derived from the Latin word for “brain,” was founded in 2015 by Andrew Feldman, a semiconductor executive, and the chip industry veterans Jean-Philippe Fricker, Michael James, Gary Lauterbach and Sean Lie. The start-up began selling chips in 2019 and has raised more than $2.55 billion in venture capital from investors including Benchmark and Foundation Capital. The company was last valued in the private markets at $23 billion.
Cerebras had previously filed to go public in 2024. At the time, it relied heavily on a single customer, which was also an investor — G42, an A.I. company backed by the United Arab Emirates. G42 accounted for 87 percent of Cerebras’s revenue in the first half of 2024, according to company filings.
Cerebras said at the time that it had notified the Committee on Foreign Investment in the United States about selling shares to G42 so the committee could review the partnership for national security risks. The company later announced that CFIUS had cleared the sale, but it pulled its I.P.O. plans with no explanation.
Now Cerebras is riding the demand from tech firms that want computing power to develop A.I. It has struck deals with Amazon and OpenAI, though it still relies mostly on a handful of major customers. G42 accounted for 24 percent of Cerebras’s revenue last year and the Mohamed bin Zayed University of Artificial Intelligence, a research lab in the U.A.E., accounted for 62 percent, according to company filings.
In total, Cerebras generated revenue of $510 million last year, up from $290 million in 2024, according to its disclosures. Profit was $238 million, compared with a loss of $482 million in 2024.
(The New York Times has sued OpenAI and Microsoft, claiming copyright infringement of news content related to A.I. systems. The two companies have denied the claims.)
Cade Metz contributed reporting.
