Cerebras Systems reported second-quarter 2026 results on 12 August 2026, furnished to the SEC on Form 8-K. GAAP cloud and other services revenue was 126.0 million dollars, up 281 percent year over year, and GAAP total revenue was 180.1 million dollars, up 74 percent. On the company’s own core non-GAAP basis, which strips out customer warrant amortisation, stock compensation and data center pass-through revenue and costs, core revenue was 209.9 million dollars, up 103 percent, with core gross margin of 41 percent and a core operating margin of negative 16 percent. GAAP gross margin was 14 percent and GAAP operating margin was negative 265 percent, so the gap between the two presentations is unusually wide and worth reading carefully.
The forward-looking numbers are larger than the quarter. Cerebras disclosed 25.4 billion dollars of remaining performance obligations as of 30 June 2026, data center capacity live or contracted for delivery by the end of 2027 of more than 600 MW, and manufacturing capacity set to increase more than tenfold during 2026 through new lines at contract manufacturers Flex, Sanmina and Rocket EMS. CFO Bob Komin said the company plans to “more than triple revenue in 2027.” Liquidity stood at 8.6 billion dollars of cash, equivalents, restricted cash and short-term investments after a 6.4 billion dollar IPO, plus an 850 million dollar revolving facility.
Two supply-chain claims stand out because they cut against the prevailing bottleneck story. Cerebras says its wafer-scale architecture avoids HBM memory, CoWoS packaging and 3nm fabrication, all of which are currently supply constrained, and that it has secured the TSMC wafer supply needed for growth. On the demand side it lists new cloud capacity agreements with coding companies Cognition and Lovable, a disaggregated inference partnership with AMD claimed to raise throughput up to fivefold and due in production in Q4 2026, an AWS partnership targeting Amazon Bedrock in Q1 2027, and support for OpenAI’s GPT-5.6 Sol at 750 tokens per second. These are company statements, not independently verified benchmarks.
For a buyer of inference capacity, the useful signal is that speed is now being sold as a distinct product category with contracted backlog behind it, and that at least one supplier claims a path around the memory and packaging shortages that constrain GPU supply. The caution is equally clear: 25.4 billion dollars of remaining performance obligations against roughly 200 million dollars of quarterly revenue means almost all of the value is in delivery risk, the customer base is concentrated in a handful of names including OpenAI, and the company is still loss-making on both GAAP and core measures. The 2027 tripling is guidance, not a result.