Cerebras Drops 19% in 3 Months: Buy, Sell or Hold the Stock?

Cerebras shares have fallen 19% over three months as Q3 2026 core revenue guidance of $214-$216 million implies only 2-3% sequential growth, while core gross margin is expected to bottom at 38-40% and operating margin deteriorate to negative 25-23%. The company reported a $450.5 million Q2 GAAP net loss, faces significant customer concentration—including OpenAI exposure—and is investing heavily in infrastructure amid competition from Nvidia, AMD and Intel. Offsetting these risks, Cerebras has $25.4 billion in RPO, over $8.6 billion of liquidity, plans for core revenue to more than triple in 2027, and Wall Street's target indicates 56.93% upside; Zacks rates the stock Hold.

Analysis

The key underwriting question is not booked demand but its economic quality: a capacity-heavy inference model can convert nominal backlog into low-margin revenue if utilization, power costs, and customer-specific service obligations rise faster than system productivity. The next 1-3 months are therefore likely governed by proof that owned capacity displaces rented capacity without a revenue interruption; a gross-margin rebound alone is insufficient if operating losses widen through depreciation, power, and network costs.

Customer concentration creates asymmetric downside because a single large buyer can use its scale to renegotiate pricing, deployment timing, or exclusivity as competing inference alternatives improve. Conversely, diversified hyperscaler deployments would matter disproportionately: they would validate that demand is not merely a bespoke workload and could justify a shift from a project-finance valuation framework toward a recurring infrastructure multiple. NVDA remains the relative winner if customers prioritize ecosystem portability and rapid availability, while AMD is the more direct beneficiary if buyers seek a lower-cost, standards-based alternative rather than a proprietary architecture.

The consensus may be too focused on the eventual margin target and too little on financing-adjusted returns on the data-center buildout. Large liquidity buffers reduce near-term solvency risk but do not eliminate multiple compression if incremental capital earns below the cost of capital. The stock is not compelling as a directional long until management demonstrates sequential gross-margin recovery, a meaningful decline in top-customer concentration, and evidence that backlog converts to cash collections rather than simply reported revenue.

Key Decisions for Investors

• Remain neutral/avoid initiating a standalone CBRS long into the next earnings print; reassess only if gross margin exits the guided trough and management quantifies owned-capacity utilization, power economics, and cash conversion. Falsifier for caution: two consecutive quarters of margin expansion with stable operating-loss intensity and a material non-OpenAI revenue contribution.
• For a 1-3 month relative-value expression, consider long NVDA / short CBRS in beta-adjusted sizing. The trade captures ecosystem and execution-premium divergence if inference deployments are delayed or priced aggressively; exit if CBRS discloses a major diversified hyperscaler contract with favorable unit economics or NVDA’s AI demand guidance weakens.
• Use TSM as the cleaner 6-18 month capacity beneficiary rather than CBRS: wafer demand is monetized before downstream data-center utilization risk emerges. Keep sizing modest because CBRS-specific volume is unlikely to be material to TSM; the thesis is invalidated by broad AI accelerator order cuts rather than CBRS execution alone.
• Set an alert for customer prepayments, receivables days, and capitalized infrastructure commitments at the next filing. Rising receivables or commitments without proportional operating cash flow would signal that reported backlog is lower quality and support increasing the CBRS short leg.