Cerebras Is About as Big as Nvidia's Data Center Business Was Nearly a Decade Ago. The Similarities Mostly End There.

Cerebras expects $880 million–$890 million in 2026 core revenue, about 74% growth at the midpoint, but its third-quarter forecast of $214 million–$216 million implies year-over-year growth slowing to roughly 58%. The company remains unprofitable, with a second-quarter core operating margin of negative 16%, and three customers accounted for about three-quarters of revenue. At roughly $165 per share, its market value is about $39 billion, or 44 times expected 2026 core revenue; the article argues investors should wait for a broader customer base and a path to consistent profits.

Analysis

The key issue is not whether Cerebras can grow quickly, but whether its speed advantage creates a durable, profitable inference niche before customer concentration and Nvidia’s ecosystem limit its bargaining power. A concentrated buyer base can make quarterly growth look stronger—or weaker—than underlying adoption: monitor repeat usage and revenue share by customer, not just headline growth. If customers use Cerebras for latency-sensitive workloads while retaining Nvidia for broader training and inference, the market may expand without direct one-for-one displacement; that supports Cerebras but weakens the case for treating it as a smaller Nvidia.

Near term, the Q3 delivery versus guidance and any update on customer concentration are likely to matter more than the long-run market narrative. Over 1–3 months, evidence of broader customer additions and improving core operating margin would support the premium; slowing growth without margin progress would expose valuation sensitivity. Over 6–18 months, the structural test is whether utilization and customer diversification can turn specialized performance into repeatable economics. Nvidia benefits if the workload remains general-purpose or its software and system breadth keep customers from switching; a credible specialized alternative is a competitive risk, but not yet evidence of broad share loss.

Contrarian point: a falling post-IPO share price does not itself establish cheapness, while exceptional growth does not validate the current premium. Potential IPO-related supply is an additional overhang to verify against lockup and insider-sale terms, not assume. The thesis weakens if Cerebras broadens its customer base and improves margins; it strengthens if growth decelerates, concentration persists, or operating losses fail to narrow.

Key Decisions for Investors

• No outright long CBRS on growth alone. Wait for Q3 results and verify revenue by customer, repeat usage, and core operating margin; broadening customers alongside continued margin improvement would be a constructive entry signal.
• For relative-value books, consider a small, risk-defined short CBRS / long NVDA pair only after checking CBRS borrow availability, IPO-related supply, and hedge sizing. This expresses valuation and execution-risk divergence, not a forecast that Nvidia will lose no inference share; cap risk against a sharp CBRS squeeze or materially better profitability evidence.
• Set a thesis review trigger around the next filing: persistent concentration combined with growth below guidance or no further operating-margin improvement argues against owning CBRS at a premium. Conversely, diversification and a clear path toward positive operating economics would falsify the cautious view.
• Track inference workload adoption and customer alternatives over the next 6–18 months. Evidence that Cerebras wins incremental latency-sensitive workloads without displacing Nvidia broadly is positive for the niche but insufficient on its own to justify a broad NVDA short.