Cerebras Systems Inc. (CBRS) Earnings

Management highlights

- Overall Quarter Performance * Q2 2026 was a strong quarter with a successful completed IPO, and Cerebras beat all prior guidance metrics for core revenue, core gross margin, and core operating margin * 2026 is a foundation-building year to support massive revenue growth in 2027–2029, with $25.4 billion in remaining performance obligations (RPO) as of quarter-end - Capacity Expansion * Secured over 600 megawatts of data center capacity (live or under contract for delivery by end of 2027) across 13 locations in the U.S., Canada, and Europe, with a gigawatt-scale pipeline of additional expansion opportunities * Expanded manufacturing capacity to 4x 2025 H1 levels, with a target of 10x full-year 2026 capacity growth, and contracted 3–4x additional capacity growth for 2027 * Secured sufficient wafer supply from TSMC; Cerebras uses 5nm nodes (lower cost, less supply constrained) and does not require HBM memory, advanced packaging, or 3nm capacity, avoiding most industry-wide supply chain constraints - Technology and Product Capabilities * Delivered production support for OpenAI's GPT-5.6 Sol frontier model, serving the model at 10x faster speeds than alternative solutions, demonstrating the maturity of Cerebras' software stack * Launched disaggregated inference solutions in partnership with AMD (Helios) and AWS (Tranium), which splits inference into prefill (handled by GPUs) and decode (handled by Cerebras' wafer-scale engine). This combined solution maintains Cerebras' industry-leading speed while increasing total throughput by 5x, improving data center economics by enabling up to 5x more high-value tokens per Cerebras system * Product roadmap targets doubling inference speed annually for the next several years (starting from a 15x existing performance advantage) and increasing total throughput by more than 20x by end of 2027; the 4th generation CS4 system will be unveiled the following week, with CS5 launching in H2 2027 * Has ongoing strategic research partnerships with the U.S. government for stacked memory and integrated wafer-scale optical solutions - Customer and Market Expansion * The AWS Cerebras inference solution is on track for general availability on the AWS Bedrock platform in Q1 2027, with first revenues expected from AWS and other hyperscalers starting mid-2027; no hyperscaler business is included in the current $25.4 billion RPO * Signed six new deals over $30 million in Q2, including new customers in fast-growing AI coding (Figma, Cognition, Lovable), agentic AI (Block, AlphaSense, GSK), and AI-powered enterprise security (CrowdStrike), a new market enabled exclusively by fast inference that is expected to grow substantially

Guidance

- Q3 2026 guidance: Core revenue is expected to be in the range of $214–$216 million, core gross margin 38%–40%, and core operating margin -25% to -23%. Q3 is expected to be the low point for core gross margin, as temporary rented cloud capacity is still in use - Full year 2026 guidance: Management upwardly revised core revenue to a range of $880–$890 million, upwardly revised core gross margin to 41%–43%, and upwardly revised core operating margin to -19% to -17%
- Long-term guidance: Management expects to more than triple core revenue in 2027, with continued multiple-year growth thereafter. Core gross margin is expected to improve significantly starting in Q4 2026, and trend toward a long-term target of 60%+ as lower-cost owned capacity replaces temporary rented capacity, throughput increases lower per-token costs, and scale drives supply chain cost improvements

Segment performance

Cerebras reports two core product segments: 1) Core cloud and other services: Q2 2026 revenue was $127.7 million, growing 287% year over year. This segment contributed 60.8% of total Q2 core revenue, with a core gross margin of 41.8% (1,600 basis points higher than Q2 2025). 2) Core hardware: Q2 2026 revenue was $82.1 million, growing 17% year over year. This segment contributed 39.2% of total Q2 core revenue, with a core gross margin of 38.8% (510 basis points higher than Q2 2025). Total Q2 2026 core revenue was $209.9 million, up 103% year over year, with an overall core gross margin of 40.6% (940 basis points higher year over year) and a core operating loss of $33.6 million (core operating margin of -16%, a 2,600 basis point improvement year over year).

Risks & headwinds

- Data center capacity remains the primary industry-wide bottleneck for growth; even after securing 600 megawatts of contracted capacity, the current secured capacity is not enough to meet all existing customer demand - Short-term gross margin is pressured by the temporary need to rent back system capacity from customers to meet unmet demand for cloud inference services, reducing sequential core gross margin from Q1 2026 levels - All forward-looking growth and execution plans are subject to inherent risks and uncertainties that could cause actual results to differ materially from expectations, as detailed in Cerebras' SEC filings

Analyst Q&A

• Q: OpenAI makes up a large share of current incremental revenue, with AWS expected to contribute significant new revenue in 2027. What will customer concentration look like next year, and how are discussions with other hyperscalers progressing?

  A: OpenAI will remain a meaningful portion of 2027 revenue, but its percentage share will decline over time as AWS and other new high-growth segments (coding, security) grow in contribution. Cerebras has historically used early large customer relationships to expand into broader markets, and discussions with other hyperscalers are progressing well.

• Q: How will the AMD disaggregated solution be commercialized, what is the revenue model, and how does AMD's recent inference hardware acquisition impact the partnership?

  A: The joint disaggregated solution (AMD Helios for prefill, Cerebras for decode) is already commercially compelling and has active buyers. Cerebras purchases the Helios racks to deploy in its cloud, and retains all customer revenue from the solution. AMD's acquired startup will likely target applications outside of data center inference, so it does not conflict with the Cerebras partnership. The solution will be generally available for deployment in Q4 2026.

• Q: Will 20x throughput growth for Cerebras' native systems by end of 2027 eliminate the need for disaggregated heterogeneous inference solutions?

  A: Cerebras is pursuing all available avenues to increase throughput. Native system throughput improvements and disaggregated solution gains are complementary, not mutually exclusive. Increasing throughput via both strategies delivers better overall performance and economics, so both paths will continue to be pursued as part of the product roadmap.

• Q: Will the contracted 600 megawatts of data center capacity accelerate Cerebras' 2027 revenue ramp?

  A: Faster deployment of secured data center capacity will directly enable faster revenue growth for both Cerebras' cloud business and on-prem hardware shipments. Data center expansion is a top corporate priority, with a dedicated team accelerating development timelines to bring capacity online as quickly as possible to meet unmet demand.