Cerebras closed its first public trading day up 68 percent at $311.07, after selling 30 million shares at $185 and raising $5.55 billion. CNBC said the debut valued the AI-chip company at about $95 billion. [1]
Sunday's paper argued that Cerebras was a customer-concentration swap, not just an IPO pop. Monday did not retire that frame. It gave it a market price.
The company has an enviable problem. Investors want pure-play AI hardware. Cerebras has a public ticker, a dramatic first-day print and enough growth language to make Nvidia look like a category rather than a single stock. But the public-market question is not whether artificial intelligence needs chips. It is whether a company can earn a $95 billion valuation while its revenue base still depends on a small cast of very large counterparties.
The prospectus history is the tell. CNBC reported that Cerebras filed in September 2024, withdrew a little over a year later after scrutiny focused on its reliance on G42, and then refiled in April. In the refreshed prospectus, G42 fell from 85 percent of revenue in 2024 to 24 percent last year, while Mohamed bin Zayed University of Artificial Intelligence accounted for 62 percent of revenue. [2]
That is diversification in one sense and concentration in another. It is better than a single customer with almost the whole business. It is not yet the kind of dispersed enterprise revenue that lets public investors stop reading the customer table before they read the income statement.
Andrew Feldman, the co-founder and chief executive, described the shape of the market plainly. "There's some whales out there, there's some really big customers," he told CNBC. [2] That sentence is more useful than most debut-day adjectives. Whales can make a young company rich fast. They can also turn a revenue chart into a dependency map.
CNBC's headline number is the debut pop. X's instinct is the bubble argument: AI chips, sovereign money, OpenAI adjacency, and a market eager for a Nvidia alternative. The paper's job is narrower. It asks what has become public enough to test. The answer is a table: G42, MBZUAI, OpenAI, Amazon Web Services, warrants, cloud deals and a stock price that now moves every trading day.
Cerebras has tried to change the mix. CNBC reported that the company announced a cloud deal with OpenAI worth more than $20 billion that expires in 2028, and that AWS plans to put Cerebras chips in its data centers so developers can run AI models quickly. Amazon and OpenAI both have warrants to purchase Cerebras stock. [2]
Those deals matter because they are the counterargument to the concentration critique. They move the company away from selling boxes into a cloud-service model. They also put Cerebras against cloud providers with scale, procurement muscle and their own accelerator strategies. Google, Microsoft, Oracle and CoreWeave are not background characters in that paragraph. They are the next public-market comparison set.
The cleanest bullish case is that Cerebras has arrived just as AI infrastructure demand broadens beyond Nvidia GPUs. Revenue jumped 76 percent last year to $510 million, and the company swung to net income of $88 million from a loss of $481.6 million a year earlier, according to CNBC. [1]
The cleanest skeptical case is that the market has paid a premium before it has seen a normal customer base. A business can be real, technologically interesting and still vulnerable to the purchasing schedule of a handful of institutions. That distinction is the difference between a story stock and an operating company.
Monday's consequence is therefore not the first-day gain. It is the discipline the public market now imposes. Every quarter will ask whether revenue is becoming less concentrated, whether OpenAI and AWS are usage channels rather than names on a slide, and whether the UAE university line falls because other customers arrive or because one whale slows.
The secondary discipline is language. "AI infrastructure" can make every buyer sound permanent and every warrant sound like proof of demand. Public investors should ask the older questions: who pays, when do they renew, what happens if they pause, and which revenue line survives a customer leaving.
The next receipt is not another celebratory interview. It is the first public quarter that shows whether the debut broadened the company or merely made its concentration easier to price.
-- THEO KAPLAN, San Francisco