Cerebras priced its IPO at $185 and closed at $311. Andrew Feldman and Sean Lie became billionaires.

Cerebras Systems opened on Nasdaq on May 14 at $385, more than double the $185 IPO price set the night before. CBRS finished at $311.07, up 68%, with a market cap just under $95 billion. It’s now the year’s biggest U.S. tech IPO.

The window for AI hardware IPOs has been closed since 2022. Cerebras filed back in 2024, pulled the offering over CFIUS scrutiny of an Abu Dhabi investor, and refiled this spring on the back of a $510 million revenue year and a $10 billion OpenAI inference deal. The pop on day one is what re-opens that window for everyone else, and the two co-founders who stayed on the cap table are now the only listed pureplay-AI-silicon billionaires that aren’t Jensen Huang.

What we know

The pricing, the open, and the close are all on the public record from CNBC, TechCrunch, and Bloomberg.

• The raise. Cerebras sold 30 million shares at $185 for $5.55 billion in proceeds before underwriting fees. The initial range had been $165 to $175; the company priced above it.
• The pop. Indications crossed +89% pre-open. The first print came in at $385 (a 108% pop), it dropped during the morning, and the close was $311.07. Both numbers got cited in the headlines: TechCrunch led with the intraday peak, CNBC and Bloomberg led with the close. The right framing is that $185 is what insiders sold at, and $311 is what the public paid by 4 p.m.
• The billionaire question. CEO Andrew Feldman’s pre-IPO stake at the $185 price was worth roughly $1.9 billion. CTO Sean Lie’s was about $1 billion. Both held those shares through the open, so the close-of-day numbers are higher again. Lead investor Benchmark, whose partner Eric Vishria almost didn’t take the original meeting, now sits on a position multiple times the size of its $25M Series A check from 2016.
• The financials. 2025 revenue was $510 million, up 76% year over year, and net income was $237.8 million. That compares with a roughly $500 million net loss in 2024. The flip from a half-billion loss to a quarter-billion of net income in twelve months is the line analysts spent the morning re-reading.
• The customer mix. Cerebras supplies inference compute to OpenAI, Amazon Web Services, G42, and the Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi. The G42 stake is the one that stalled the 2024 IPO attempt over CFIUS review; this filing closed without the same blocker.

What we don’t know yet

Three numbers Wall Street is going to want by the next earnings cycle aren’t in the S-1.

• Lock-up bleed. Standard insider lock-ups run 180 days. The next pricing event is the November window when employees and early investors can sell. Tech IPOs from 2020 to 2022 frequently lost 30 to 60% of their open-day price by then; Cerebras enters that window in early-to-mid Q4.
• Concentration risk. The S-1 disclosed that a single customer (G42) accounted for the majority of 2024 revenue. The 2025 mix isn’t fully broken out in the day-one filings. Wall Street will want a number, and the 10-Q in August is when they get it.
• Whose IPO is next. Cerebras debuting at this size is the data point that re-opens the window. Groq, SambaNova, and Tenstorrent are all candidates. Bloomberg cited at least one banker who said the books on the next AI chip IPO will likely open within 60 days.

The setup

Cerebras builds wafer-scale chips: a single piece of silicon roughly the size of a dinner plate that holds 900,000 cores and skips the inter-chip networking step that drives Nvidia’s NVL72 rack costs. The pitch to customers is “no chip-to-chip traffic, no cabling, no rack-scale cluster math.” The pitch to public-market investors is “we are the only listed AI chipmaker that isn’t Nvidia or AMD.”

That positioning explains the day-one demand more than the financials do. There are exactly two pureplay AI silicon stocks available on U.S. exchanges as of yesterday. One has a $4 trillion market cap. The other is Cerebras. Buyside index funds that need any kind of AI-hardware exposure beyond Nvidia have a very short shopping list, and CBRS just landed at the top of it.

Andrew Feldman has been here before. He sold his prior company, SeaMicro, to AMD for $334 million in 2012. AMD discontinued the product line three years later. Whether wafer-scale chips have a longer commercial life than microservers did is the bet inside the bet.

What this means for you

If you operate inference infrastructure, the immediate consequence is supply. A freshly capitalized Cerebras with $5.5 billion on the balance sheet can build out manufacturing capacity at a pace it couldn’t last year, and that affects pricing on every workload that competes with Nvidia H200/Blackwell instances. Watch what AWS does with its Cerebras-backed inference SKU pricing in the next two quarters. That’s the leading indicator.

If you’re an early-stage AI chip startup, your seed deck just got easier and your valuation conversation just got harder at the same time. Easier because there is now a public comparable trading at 13 times revenue. Harder because every late-stage investor you talk to will assume your IPO needs to look something like this one to clear. The next AI chip IPO will be priced relative to today’s CBRS chart, not relative to your DCF.