The AI Chip Bet That Made Wall Street Ring the Bell: Cerebras Goes Public
There was a moment, roughly twelve months ago, when Cerebras Systems looked like a cautionary tale rather than a market debut. Its IPO filing had stalled in regulatory limbo, its revenue was dangerously concentrated in a single foreign customer, and the window for AI hardware plays seemed perpetually about to close. On Thursday, none of that mattered. The company raised $5.5 billion, watched its stock more than double on the first day of trading, and gave every AI infrastructure investor who stayed patient a very good morning.
Cerebras priced its shares at $185 on Wednesday night — well above the initial range of $115 to $125 that the company had floated, and then raised again to $150–$160 as demand climbed. The offering itself expanded to 30 million shares. By the time trading opened on Thursday, buyers pushed the price to $385, a first-day jump of 108%. Even by the standards of a frothy AI market, the reaction was striking.
By the Numbers: $5.5B raised in IPO · Priced at $185/share · Opened at $385 (108% first-day gain) · Fully diluted valuation at IPO price: $56.4B · 2025 revenue: $510M (+76% YoY) · 2025 net income: $237.8M (vs. ~$500M loss in 2024) · CEO Andrew Feldman stake at IPO price: ~$1.9B · CTO Sean Lie stake: ~$1B
The underlying business justifies some of the excitement. Cerebras reported $510 million in revenue for 2025, a 76% increase year-over-year, spread across a more diversified customer base than the company had a year prior. More striking was the profit swing: from nearly half a billion dollars in net losses in 2024, Cerebras posted $237.8 million in net income for 2025. That kind of trajectory, in an industry where most hardware startups are still burning cash as fast as they raise it, gave institutional investors reason to bid aggressively.
The path to this moment required clearing a significant regulatory obstacle. When Cerebras first filed to go public in 2024, a large strategic investment from Abu Dhabi-based technology group G42 triggered an extended review by the Committee on Foreign Investment in the United States. CFIUS reviews are opaque by design, but the practical effect was simple: the IPO couldn't move forward while the review was open. The company shelved its plans and waited. By the time it re-emerged this spring with updated financials, those concerns had been addressed — and the new numbers gave underwriters something to sell.
"A company that once looked like a warning sign has become the opening act of 2026's IPO season."
What Cerebras actually sells is worth understanding in the context of where AI infrastructure spending is heading. The company's chips are purpose-built for AI inference — the ongoing compute work required every time a model responds to a prompt. This is a different market than training, which is dominated by Nvidia. Inference is where the volume is, and where cost pressures are sharpest: for every enterprise deploying a large language model at scale, inference costs are a recurring operational expense, not a one-time investment. Cerebras's architecture, centered on a wafer-scale chip that fits an entire neural network on a single die, trades conventional flexibility for dramatically lower latency.
The customer list has grown accordingly. Alongside G42, Cerebras now counts OpenAI, Saudi Arabia's Mohamed bin Zayed University of Artificial Intelligence, and Amazon Web Services among its buyers. The OpenAI relationship is particularly notable — and somewhat unusual. OpenAI is both a customer of Cerebras and a company with deep ties to Microsoft, which competes with Cerebras in cloud infrastructure. The circularity is a feature of the current AI landscape, not a bug: the market is large enough, and the technology differentiated enough, that traditional competitive lines have become unreliable guides to actual commercial relationships.
For investors, Thursday's debut signals something beyond Cerebras itself. The IPO market for AI companies had been largely dormant through late 2025, constrained by valuation expectations that public market buyers weren't willing to meet. Cerebras cleared that bar convincingly. Whether the stock holds above its opening price or retraces toward the offering level is secondary to the message it sends: there is genuine public market appetite for differentiated AI infrastructure plays, provided the financials can support the story. Several AI hardware and infrastructure companies that have been private-market holdouts will be watching the trading closely over the coming weeks.
Co-founder and CEO Andrew Feldman's stake, valued at roughly $1.9 billion at the IPO price, reflects how much the company's fortunes have shifted in twelve months. CTO Sean Lie's position is worth approximately $1 billion. Neither number accounts for Thursday's price action, which pushed the stock well above those baseline figures before cooling somewhat through the afternoon session. The founding team's paper gains are substantial — but more relevant to the broader market is what Cerebras's debut says about the economics of building specialized AI hardware when the largest customers in the world are still trying to figure out which chips they need most.