Cerebras Beat Estimates and Tanked Anyway—Welcome to the IPO Penalty Box

Cerebras delivered the kind of earnings report that should have moved stock prices higher. Revenue came in at $193.4 million, roughly $10 million above what Wall Street had pencilled in. The adjusted loss of $2.48 million was smaller than analysts expected. By any conventional measure, this was a beat. The stock fell 10% anyway.

Welcome to the IPO penalty box, where positive news arrives with a timer attached.

The chipmaker's real offense was not what it reported—it was what comes next. Cerebras guided investors toward a gross margin of 36 to 38 percent in the second quarter. That margin sat at 46.5 percent in the first quarter. The company is forecasting roughly 1,000 basis points of compression in a single quarter. For context, that is not a gentle deceleration. That is a cliff.

This is where the stock market's relationship with numbers becomes less a matter of arithmetic and more a matter of sentiment. Revenue grew 92 percent year-over-year, climbing from $99.5 million to $193.4 million. The company is shipping chips at accelerating velocity. Yet investors responded to margin guidance the way a restaurant critic responds to a menu written in a font that suggests the kitchen stopped caring sometime in 2019.

CEO Andrew Feldman attempted damage control on Wednesday, telling investors they had misunderstood the guidance. "We shared that plan as we went public a few months ago, and we're beating that plan," he said. This is technically accurate. Feldman is also correct that nothing in the numbers was, strictly speaking, disappointing. The problem is that Wall Street did not come to Cerebras' earnings call to be technically correct about previous guidance. It came to hear whether margins were improving or deteriorating. The answer was unambiguous.

Context matters here. Cerebras went public at $185 per share and opened trading at $350, nearly doubling on day one. The stock touched $386.34 before lunch. Six weeks later, after one earnings report, it closed Tuesday at $226.72. The damage amounts to a 28 percent decline from the opening print. This is the arithmetic of IPO excess meeting the reality of execution.

The revenue story is genuinely remarkable. Cerebras announced during the first quarter a deal worth over $20 billion to supply OpenAI with computing power. Yet 86 percent of 2025 revenue came from two UAE-linked entities: G42 (24 percent) and Mohamed bin Zayed University of Artificial Intelligence, MBZUAI (62 percent). In practice, the company replaced one large customer with two entities considered related parties to each other. This concentration is not necessarily disqualifying, but it is worth noting when a chipmaker's growth story depends almost entirely on a handful of sources.

Morgan Stanley analysts remained bullish and raised their price target to $273 from $250, writing that nothing in the numbers was disappointing. They are not wrong about the fundamentals. They are, however, operating under the assumption that stock prices are rational functions of underlying business performance. The market's response to Cerebras suggests otherwise.

The lesson here is not new, though it appears to require annual refreshing. Beating estimates is a necessary condition for stock appreciation during an IPO correction. It is not sufficient. Investors want to know whether margins are expanding or contracting, whether management is executing on previously stated plans, and whether the company's reliance on a narrow set of customers is sustainable or merely convenient. Cerebras beat estimates on the earnings line. It failed the margin test and the concentration test, and it is paying the price in real time.

The stock opened at $350 on momentum and closed Tuesday at $226.72 on math. That is not a misunderstanding. That is the market working.