Cerebras Stock Has Been Cut in Half. It Still Costs About 145 Times Next Year's Estimated Earnings.

Cerebras shares trade near $184, down 52% from their $386.34 post-IPO peak, despite Q2 core revenue more than doubling 103% year over year to $209.9 million and management raising full-year core-revenue guidance to $880 million-$890 million. The sell-off is characterized as a valuation reset rather than a deterioration in operations: the stock still trades at roughly 145x expected 2027 EPS of $1.25, while core operating margin remained negative 16%. Its $25.4 billion backlog, including an OpenAI capacity agreement, provides growth visibility, but only $5.6 billion is expected to convert into revenue over the 24 months ending June 2028 and profitability remains unproven.

Analysis

CBRS is transitioning from a hardware-led valuation framework to a capacity-delivery and cloud-utilization story. That raises the relevant diligence threshold: reported growth matters less than contracted capacity becoming billable, sustained utilization, cash collection, and incremental gross margin after power, hosting, depreciation, and customer-warrant economics. A concentrated hyperscale-scale commitment can create apparent multi-year visibility while also making the equity highly sensitive to one customer's deployment schedule, financing capacity, and inference-demand assumptions.

The likely near-term overhang is not another fundamental miss but post-IPO supply, estimate dispersion, and a valuation reset toward execution milestones. Over the next 1-3 months, any evidence that revenue growth is being purchased through low-margin pass-through infrastructure or that operating leverage is delayed should compress the multiple further; conversely, cash-flow and gross-margin proof could rapidly reverse sentiment given the size of the drawdown. Six to eighteen months out, CBRS must demonstrate that its systems produce attractive unit economics versus NVDA's ecosystem, not merely that customers are willing to reserve capacity.

The contrarian bull case is that a depressed share price can make the risk/reward asymmetric if contracted workloads convert on schedule and fixed infrastructure costs are absorbed faster than expected. However, the market may still be underpricing delivery risk: AI capacity contracts are economically closer to project-finance commitments than conventional software backlog, with power availability, data-center commissioning, customer credit, and cancellation provisions potentially more important than headline RPO. The thesis is falsified positively by sustained GAAP gross-margin expansion, positive operating cash flow, and backlog conversion without incremental dilution; it is falsified negatively by a material schedule revision, receivables build, or margin stagnation despite higher utilization.

Key Decisions for Investors

• Remain uncommitted in CBRS until the next two reporting periods establish GAAP gross-margin trajectory, operating cash flow, receivables aging, and customer concentration. A raised revenue outlook alone is not a buy trigger; require evidence that incremental revenue converts to durable contribution profit.
• For a tactical bearish expression over 1-3 months, consider a small CBRS short versus long NVDA, sized beta-neutral. The pair isolates execution and valuation risk from broad AI-capex upside; cover if CBRS reports positive operating cash flow or materially better-than-expected GAAP margin, and reassess if NVDA demand indicators weaken.
• Do not underwrite the backlog at face value without reviewing cancellation rights, minimum-payment terms, power/hosting obligations, and customer prepayments. Treat disclosure of these items as the primary catalyst watch list rather than initiating on price weakness.
• If CBRS falls materially below IPO pricing while delivery milestones remain intact, revisit through defined-risk calls rather than common equity. Entry should wait for independently verifiable utilization and cash-conversion data; the upside case requires multiple stabilization, while the maximum loss must be limited because timing slippage can produce another sharp de-rating.