Most people watching the markets Thursday had never heard of Cerebras Systems. By the end of the day, the company was worth $95 billion.
Cerebras priced its IPO at $185 a share, opened at $350, and closed up 68% at $311. It raised $5.55 billion. That’s a remarkable debut by any measure — and it happened four days ago. The stock has already pulled back to around $280, which is actually when these stories get interesting.
Before anyone decides what to do with that information, here’s what Cerebras actually is.
Cerebras Systems (CBRS)
Nvidia built its empire by putting thousands of smaller processors on a single chip and connecting them. Cerebras took a completely different approach. Their chip, called the Wafer Scale Engine, is literally the size of a dinner plate — about 46,000 square millimeters. Instead of cutting a silicon wafer into hundreds of small chips and then trying to wire them together, Cerebras uses the entire wafer as one chip. The result is a processor with 4 trillion transistors and an enormous amount of memory sitting right next to the compute. That matters for certain types of AI work, particularly inference — running an AI model in real time to answer a question or generate content — where the bottleneck is often moving data, not processing it.
The company did $510 million in revenue last year, up 76% from 2024. Morgan Stanley projects $8 billion by 2028. Those are serious numbers. OpenAI signed a $20 billion compute agreement with them. That’s not a small customer.
But here’s the thing every investor needs to know before buying: 86% of Cerebras’s 2025 revenue came from just two customers, both linked to the UAE. That’s not a red flag that kills the story. It’s a risk that shapes it. A company at $67 billion in market cap with $510 million in revenue and two customers accounting for nearly all of it is a different kind of bet than most people realize when they see a stock up 68% on day one.
The pullback from $311 to around $280 is the market processing that reality. Could be the start of more selling. Could be the first dip before a real run. Too early to say.
What I can say: the technology is real. The customers are real. The growth trajectory is real. But this is early-stage, concentrated-revenue, high-multiple territory. If you’re interested, position size accordingly and understand what you own.
The Nvidia comparison is going to come up constantly. Cerebras is not Nvidia. Nvidia dominates training — building AI models from scratch. Cerebras is staking its ground in inference — running those models in the real world. Whether that turns into a durable business depends entirely on whether they can broaden their customer base over the next two years.
That’s the story to watch.
