Cerebras Systems Charts Path to Triple Revenue in 2027 on Fast Inference Demand
The AI chip maker secured 600 megawatts of data center capacity and announced disaggregated inference partnerships as it works through a $25 billion revenue backlog.
Record Quarter Sets Stage for Acceleration
Cerebras Systems reported a stronger-than-expected second quarter and laid out an aggressive expansion plan built on surging demand for fast AI inference, with CEO Andrew Feldman telling investors the company expects to "more than triple our core revenues in '27 and continue to grow at multiples in the years following" (earnings call, 2026-08-12).
The company, which trades at a market capitalization near $50 billion, beat guidance across all key metrics including core revenue, core gross margins, and core operating margin. Feldman framed 2026 as "a foundation-building year" as Cerebras works through $25 billion in remaining performance obligations.
Data Center Expansion Addresses Key Bottleneck
Cerebras has secured more than 600 megawatts of data center capacity across 13 locations globally, either live or scheduled for delivery by end of 2027. The facilities span U.S. cities including Dallas, Denver, Minneapolis, and Santa Clara, plus international sites in France, Finland, Norway, and multiple Canadian provinces (earnings call, 2026-08-12).
Feldman noted that "data center space continues to be the bottleneck for the entire industry," but emphasized Cerebras benefits from serving inference workloads that don't require the gigawatt-scale footprints needed for training clusters. The company has built what will become "among the largest non-hyperscale AI cloud" and expects its data center pipeline to reach gigawatt scale (earnings call, 2026-08-12).
AMD Partnership Targets Economics and Throughput
The company announced a disaggregated inference partnership with AMD combining Cerebras systems with AMD's Helios racks. The solution separates inference into prefill and decode stages, using GPUs for the parallelizable prefill work and Cerebras' wafer-scale engine for the memory-intensive decode generation.
According to Feldman, the combined solution "maintains Cerebras speed while increasing throughput by 5x," meaning "up to 5x as many high-speed, high-value tokens are made by each Cerebras system" with corresponding improvements in per-token economics and data center profitability (earnings call, 2026-08-12). The company also disclosed a similar disaggregation partnership with AWS Trainium.
Product Roadmap and Strategic Positioning
Cerebras now supports OpenAI's GPT-4.5 Sonnet, delivering the frontier model at "10x faster" speeds, which Feldman said "lays to rest any of the remaining concerns regarding our ability to support large frontier models" (earnings call, 2026-08-12).
The company expects to maintain its performance advantage while increasing throughput by more than 20x over the next 18 months. Feldman said Cerebras plans to "deliver new systems that double our speed each year for the next several years," starting from what he called "a 15x performance advantage over everyone else in the industry" (earnings call, 2026-08-12).
Manufacturing capacity is expanding more than 10x in 2026 through partnerships with Flex and other providers, with continued expansion planned for 2027. The company emphasized it avoids several industry supply constraints by operating on TSMC's 5-nanometer node rather than cutting-edge processes, and by not requiring HBM memory or CoWoS packaging.
Cerebras' AWS partnership is expected to deliver generally available solutions through Bedrock in Q1 2027, while discussions with other hyperscalers are progressing toward mid-2027 first revenue.
MarginX data shows recent insider activity including conversions and sales by Dhiraj Mallick.
This article was generated by MarginX from the earnings call on 2026-08-12. It is not investment advice.