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

The key risk is not whether Cerebras can grow, but whether growth converts into durable gross profit before customer concentration and the cost of scaling its own compute footprint become binding. Concentrated demand can support rapid initial deployment, yet losing or delaying one large account could leave capacity underutilized and amplify operating losses. Conversely, utilization gains could create meaningful operating leverage; watch customer additions, repeat usage, and operating cash flow rather than headline revenue alone.

Near term, strong reported growth may support sentiment, but the slower implied quarterly trajectory raises the bar for another upside surprise. Over 1–3 months, customer concentration, cash consumption, and any evidence of improving unit economics are the catalysts to track. Over 6–18 months, the thesis depends on whether fast inference becomes a sufficiently large workload category and Cerebras can compete without relying on a few buyers. NVIDIA’s software ecosystem and incumbent customer relationships are a structural hurdle; faster inference alone does not establish a lower-cost or more profitable alternative.

Contrarian point: the comparison with NVIDIA’s early data-center revenue is weak evidence of a comparable outcome. The relevant difference is the absence of an established profit engine, while the current valuation leaves little room for execution slippage. But an outright short is also vulnerable to rapid growth, scarce AI exposure, and high volatility. Verify GAAP-to-core revenue reconciliation, cash runway, capacity investment, customer retention, and actual inference economics before sizing a directional position.