This post was originally published on this site.
These results support the view that demand for AI inference is spreading beyond Nvidia to specialist chip designers. They give the AI infrastructure trade another sign that spending is shifting from training models to running them. The gap between GAAP and core profitability will be scrutinised by investors still pricing a recent IPO, and a GAAP loss alongside core profit tends to make for volatile trading early on. Peers in AI cloud capacity and data centre power may also draw support from the scale of the backlog and capacity build-out.
–
Earlier:
- Applied Materials beats on Q4 and guides above forecasts as chip equipment demand holds
- Cisco beats and lifts FY27 view as AI revenue from hyperscalers set to more than double
–
Cerebras is turning from a wafer-scale chip seller into an AI cloud provider, with a $26bn backlog and a promise to triple revenue next year. It now has to deliver on both.
Summary:
- Q3 revenue rose 55% to around $211m, and core revenue rose 70% to around $231m
- Cloud and other services revenue more than tripled to around $142m, while hardware revenue fell to around $69m from about $97m
- GAAP net loss widened to around $161m, or 67 cents a share, while core net income was around $30m
- Q4 core revenue guidance is $310m to $330m, with core gross margin of 47% to 49%; full-year core revenue is seen at $940m to $960m
- Backlog of around $26bn and about $7.8bn in cash and short-term investments
- New deals include an Oracle cloud integration due in 2027 and two hyperscaler wins; the CEO expects revenue to triple in 2027
Cerebras Systems posted record third-quarter revenue and raised its full-year outlook in its first earnings season as a listed company. Fast growth in its AI cloud business made up for a fall in hardware sales, and chief executive Andrew Feldman said he expects revenue to triple in 2027. Revenue rose 55% from a year earlier to around $211 million for the quarter ended 30 September. The company’s preferred “core” measure, which strips out certain items, rose 70% to around $231 million.
Cerebras is based in Sunnyvale, California. It builds specialised computer systems for artificial intelligence and presents itself as a challenger to Nvidia. Its signature product is a wafer-scale chip: a single processor about the size of a dinner plate, made from an entire silicon wafer, where rivals cut each wafer into many smaller chips. The company says this design makes its systems especially fast at inference, the work of running trained AI models to answer queries. It sells the systems outright and, more and more, rents out their computing power through its own cloud service. Cerebras listed on Nasdaq under the ticker CBRS this year.
That shift to cloud is now driving the business. Cerebras said in its third-quarter release that cloud and other services revenue more than tripled to around $142 million. Hardware revenue fell to around $69 million, from about $97 million a year earlier. The company reported a GAAP net loss of around $161 million, or 67 cents a share, but posted core net income of around $30 million.
For the fourth quarter, Cerebras expects core revenue of $310 million to $330 million, core gross margin of 47% to 49% and an operating margin of around breakeven. For the full year, it expects core revenue of $940 million to $960 million. It reported a backlog of around $26 billion and about $7.8 billion of cash and short-term investments. New deals include an Oracle cloud integration due in 2027 and two hyperscaler wins.
The main risk is how much Cerebras depends on a few customers. It relies on a small number of large customers, including OpenAI. To turn that backlog into revenue, it also has to deliver about 700 megawatts of data centre capacity by the end of 2027.
This article was written by Eamonn Sheridan at investinglive.com.
