Cerebras reported Q2 2026 cloud and services revenue of $126 million, a 281% year-over-year increase, while hardware sales fell 23% to $54.1 million. Total GAAP revenue reached $180.1 million (up 74% YoY), though the company missed analyst expectations of $194.2 million, causing shares to plunge 18% in after-hours trading. Core revenue, excluding non-cash items, totaled $209.9 million, up 103% YoY. The earnings shift reveals a critical pivot: customers now prefer to rent Cerebras compute through the cloud rather than own hardware outright, trading capex for opex and shifting risk to Cerebras to fund infrastructure in advance.
Cerebras's rapid transition from a hardware vendor to a cloud-compute provider carries substantial financial headwinds. The company reports $25.4 billion in remaining performance obligations—indicating extraordinary AI demand—but faces a capital-intensive business model: it must build and operate data centers, secure 600+ MW of capacity through 2027, and expand manufacturing 10x in 2026, all before collecting full revenue. CEO Andrew Feldman acknowledged the challenge: gross margin is still thin (14% GAAP, 41% core), and the company posted a $450.5 million net loss (partly $377M stock-based comp from its May IPO). OpenAI's $20B master agreement anchors demand—Cerebras committed 750 MW with an option for another 1.25 GW—but OpenAI's working capital loan of ~$1B is merely a partial hedge against long-term build costs.
Architects considering Cerebras for inference workloads should recognize this as a vendor making a decades-long bet on speed-first inference. With $8.6B in liquidity post-IPO and $850M credit facility, Cerebras has runway, but profitability depends on sustained pricing power and utilization. The 281% cloud growth and $25.4B backlog validate demand; the 23% hardware decline and margin compression signal that pure edge hardware plays no longer compete. Watch for Q3 2026 results and any updates to capacity commitments or customer wins to gauge whether Cerebras can execute the cloud-compute buildout faster than rivals.