Cerebras Beat Wall Street's Revenue Estimate by 8%. Its Stock Still Fell 16% Over the Next Two Days.

Cerebras Systems lost $450.5 million last quarter. By the accounting measure Wall Street actually models, it lost four cents a share instead and beat estimates by roughly 81%. Both numbers are true, and the gap between them is the whole story.

[[IMG1]]

The wafer-scale computing company, which builds chips the size of dinner plates as an alternative to Nvidia's GPUs, reported its second quarterly update as a public company on August 12. Core revenue, the non-GAAP figure that strips out stock-based compensation and other one-time items, came in at $209.9 million, up 103% year over year and roughly 8% ahead of consensus. Cloud and services revenue, the fastest-growing slice of the business, jumped 287% to $127.7 million.

What Changed

The headline GAAP numbers told a rougher story. Total GAAP revenue came in at $180.1 million, up 74% year over year but well below the $194 million figure some outlets compared it against, a number that actually reflected the core consensus rather than a GAAP one. Net loss under GAAP hit $450.5 million, or $2.98 a share, inflated by the wave of stock-based compensation that typically vests around a company's first year as a public entity. Cerebras only listed on the Nasdaq in May, in what several outlets called the biggest tech IPO of 2026, raising $5.5 billion and opening as much as 108% above its $185 IPO price before settling to close its first day up 68%.

Shares closed at $262.06 on August 12, the day of the report, fell to $231.01 the next session, and closed at $218.98 by August 14. That's a 16% slide across two trading days for a stock that beat both revenue and adjusted earnings expectations and raised its own guidance in the same release.

Why It Matters

Newly public companies with heavy stock compensation almost always show a wide gap between GAAP and adjusted results, and investors who only skim the top-line loss figure can end up selling a stock for the wrong reason. Cerebras beat on core revenue, beat on core earnings by a wide margin, raised full-year guidance, and still watched its market value shrink by roughly a sixth in two days.

The reaction fits a pattern showing up across AI infrastructure names all year. Nvidia lined up $500 billion in financing to fund AI data center construction earlier this month, and its own stock fell the same day the deal was announced. Bond investors have grown skeptical too: nearly 80% of AI data center bonds sold since early 2025 now trade below their issue price. Strong demand and heavy accounting losses keep showing up in the same earnings reports across the sector, and the market hasn't settled on how to price that combination.

Cerebras isn't short on evidence that demand is real. The company's backlog, what it calls remaining performance obligations, stood at $25.4 billion at quarter's end, roughly 29 times its raised full-year revenue guidance. Cathie Wood's ARK Invest funds treated the selloff as a buying opportunity, purchasing 106,941 shares worth $28 million on August 13, the day after the report.

What to Watch Next

Cerebras guided to $214 million to $216 million in core revenue for the third quarter, alongside an operating margin loss of 25% to 23%. Whether the stock stabilizes will likely depend less on hitting that number and more on whether investors start reading the core figures instead of the GAAP ones. The bigger test is competitive: Cerebras is trying to prove that wafer-scale chips can win a meaningful share of AI inference workloads away from Nvidia, and every quarter of demand growth without a path to GAAP profitability will keep inviting this same argument between the bulls and the skeptics.

The Pulse24 Take

The loss headline was real, and so was the growth underneath it. Cerebras' second quarter is a reminder that a single number rarely tells the whole story with a freshly public company, especially one still absorbing the accounting effects of its own IPO. Investors chasing the GAAP loss sold first and asked questions later. The ones who read the backlog and the guidance, ARK Invest among them, bought instead. Neither reaction proves the wafer-scale bet will work. What it does show is that AI infrastructure stocks are being priced on a knife's edge right now, where a single earnings report can swing 16% in either direction depending on which numbers the market decides to trust that day.