Cerebras’ wafer-scale inference advantage and $25.4 billion backlog offer exceptional growth potential, but at $147 the valuation already demands flawless execution and leaves asymmetric downside.
Overview
Cerebras Systems develops wafer-scale AI compute designed to reduce the interconnect and memory bottlenecks of conventional GPU clusters. It sells systems such as CS-3 and CS-4 and monetizes dedicated cloud capacity and per-token inference APIs. Its technical differentiation is compelling for latency-sensitive inference: WSE-3 combines 900,000 cores and 44 GB of on-chip SRAM, while CS-4 is specified at up to 750 PFLOPS. Commercial validation includes a more than $20 billion OpenAI agreement for 750 MW and an AWS integration partnership. FY2025 revenue rose 76% to $510 million; FY2026 core revenue guidance is $880–$890 million, and Q2 core revenue grew 103.1% year over year. Cloud and Services reached 70% of Q2 GAAP revenue, while RPO reached $25.4 billion. **The growth opportunity is exceptional, but concentrated:** OpenAI represents more than 75% of backlog, and FY2025 revenue was 86% attributable to MBZUAI and G42. **Profitability remains immature:** Q2 core gross margin was 40.6%, but GAAP net loss was $450.53 million, including heavy SBC and warrant amortization. At about $147, EV/FY2026 core revenue is 31.7x; the probability-weighted five-year value is $109.44. Near-term catalysts are OpenAI deployment, AWS distribution, and CS-4 execution; infrastructure and valuation leave asymmetric downside.