Cerebras is shifting from selling wafer-scale AI supercomputers toward building cloud capacity for inference customers. Its largest commitment is OpenAI’s deal, initially valued at $10 billion for 750 megawatts of systems and later described as potentially worth $20 billion to $30 billion if capacity expands through 2030 and services continue through 2032. AWS, Oracle Cloud, and Microsoft Azure have also signed deals, though they have not yet generated revenue or entered the reported backlog. In the third quarter, hardware sales fell 28.8 percent to $68.9 million, and gross margin was 11 percent, hurt by factory expansion and legacy customer pricing. Cerebras ended the quarter with $7.36 billion in cash and a $26.1 billion backlog, mostly tied to OpenAI.
The next generation is central to the company’s growth plans: its WSE-5 processor has returned from fabrication, is running models, and is undergoing system tests. Built on TSMC’s 5-nanometer process, it will power the CS-5, slated for production in early summer 2027, with more than twice the speed of the CS-4. Cerebras expects the CS-4 to enter full production soon, also bringing major performance gains. The company’s chief executive expects sales to triple in 2027; the projection follows estimated 2026 GAAP revenue of about $885 million, up 73.4 percent year over year. The broader case is that falling inference costs will embed generative AI widely, sustaining demand even as AI infrastructure spending may approach a peak.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.