Cerebras: fast inference, slow delivery as OpenAI ties draw scrutiny
OpenAI CEO Sam Altman recently posted on social media describing Cerebras as a partner, with the two working on the frontier of inference speed, and Cerebras shares rebounded in premarket trading; the move followed a sharp drop reportedly triggered by word that OpenAI used Nvidia chips for the Ultrafast mode of GPT-6.1 Sol. Citing disclosed figures, the article says Cerebras reported roughly $510 million in fiscal 2025 revenue, up about 75.7% year over year, with $237.8 million in net profit; second-quarter cloud revenue reached $126 million, up 52% quarter over quarter, including $56.8 million recognized from OpenAI, about 32% of quarterly revenue. It also notes about $392.8 million in negative free cash flow and a roughly $25.4 billion backlog as of end-June, of which only about 22% is expected to be recognized within 24 months and 43% in months 25–48, while the top three customers accounted for 76% of second-quarter revenue.
The disclosures give a rare, concrete look at whether specialized inference silicon can turn orders into revenue, while slow backlog conversion and customer concentration show delivery remains the deciding factor.