AI hardware has been a private-market story for years. The firms buying chips, building inference stacks, and negotiating compute contracts have done so largely in the dark, without the transparency that public markets force. That changes when a company rings the opening bell. And on May 14, 2026, Cerebras Systems rang it very loudly.
The Nvidia competitor, which designed its chip from scratch and purpose-built it for AI workloads, has been trying to go public since 2024. Regulatory scrutiny, customer concentration concerns, and shaky financials stalled those plans. What finally cleared the path was a dramatic turnaround in the numbers — and the market responded accordingly.
A Price That Blew Past Every Estimate
Cerebras priced its IPO shares at $185 on Wednesday evening, well above its original range of $115 to $125, which had already been raised to $150 to $160. The company increased the size of the offering to 30 million shares. That alone signaled strong institutional appetite. But the retail market went further: shares opened Thursday at $385, up 108% from the IPO price, before cooling to above $330 by mid-day.
The Valuation Story Behind the Pop
Even at the $185 IPO price, Cerebras entered its first trading day with a fully-diluted valuation of $56.4 billion. At that figure, co-founder and CEO Andrew Feldman's stake was worth nearly $1.9 billion. Co-founder and CTO Sean Lie's stake came in at roughly $1 billion. If the above-$300 price holds through the close, both founders end the day worth considerably more than those figures.
What Killed the First IPO Attempt
Cerebras first filed to go public in 2024 but ran into sustained scrutiny from the Committee on Foreign Investment in the United States, known as CFIUS, over a large investment from Abu Dhabi-based Group 42. Investor hesitation compounded the regulatory delay: Group 42 accounted for almost all of Cerebras's revenues at the time, a customer concentration risk serious enough to shelve the offering entirely. The company spent the following year diversifying its customer base and turning its finances around.
The Numbers That Unlocked Investor Appetite
By April 2026, Cerebras was able to report revenues of $510 million for 2025, a 76% increase year-over-year, across a handful of customers. More striking was the swing in net income: $237.8 million in profit, compared to a loss of nearly half a billion dollars the year prior. That shift from deep loss to substantial profit, at that scale and speed, is what put institutional investors back at the table. The revised IPO range, the upsized share count, and the eventual pricing all followed from those filings.
The Customer List Now Carries Weight
Cerebras has positioned itself as a supplier of chips for inference — the ongoing compute required for AI models to respond to prompts. Its current customer roster includes OpenAI, which TechCrunch noted holds a complicated circular-deal relationship with the company, as well as G42, Saudi Arabia's Mohamed bin Zayed University of Artificial Intelligence, and Amazon Web Services. That list represents a meaningful spread across geographies and use cases, addressing the concentration risk that spooked earlier investors.
The $5.5 billion raised in a single day, on a stock that more than doubled at open, confirms that public market appetite for AI infrastructure is intact and intense. Retail investors in particular drove the opening surge, bidding shares to $385 before the price settled above $330 mid-day.
Whether Cerebras can hold that valuation will depend on whether its 2025 revenue trajectory continues and whether its inference positioning survives intensifying competition from Nvidia and others. But for every agency and brand investing in AI-powered production workflows, a publicly traded Cerebras means more pricing transparency, more quarterly data, and a clearer read on where the compute market is heading. If the inference segment keeps growing at this pace, the hardware decisions your clients make today could look very different — and very costly to reverse — within 18 months.