Cerebras 10-Qquarter ended 2026-09-30

Executive Summary

Cerebras sells wafer-scale AI computing systems and cloud inference services to AI developers, enterprises and research institutions. It is shifting toward dedicated inference capacity, anchored by a 750 MW OpenAI agreement and $26.1B of remaining performance obligations. Cloud services now dominate revenue, but growth slowed relative to the year-to-date pace, margins weakened and capacity delivery fell behind schedule. The IPO funded substantial expansion; investors must judge whether Cerebras can turn contracted demand into profitable services before infrastructure commitments and delivery penalties consume that advantage.

What's Happening in the Business

Cloud and other services replaced hardware as Cerebras’s largest revenue source. Growth came from OpenAI, new inference agreements and support for deployed clusters. Management expects dedicated capacity coming online to improve near-term gross profit and margin, although expedited deployment will bring additional startup costs.

Execution is already lagging. OpenAI capacity delivered by 9/30/2026 was below contemplated levels before excused-delay adjustments, and management forecasts delays into part of 2027. Service credits began accruing on 10/1/2026, subject to those adjustments. Current forecasts put less than 5% of total committed capacity at risk of termination in 2026; unrecovered infrastructure costs could compound lost revenue.

Cerebras is also investing in next-generation products, engineering and software while transitioning hardware inventory. The IPO triggered substantial equity-compensation recognition, so the expense surge reflects both expansion and compensation accounting—not cash spending alone.

Key Financial Highlights

• Growth drivers: OpenAI contributed $72.4M in Q3 and $147M over nine months, net of warrant amortization. Lower hardware deliveries reduced quarterly revenue by $10.8M; services growth more than offset that decline.
• Revenue deductions and inventory: Customer-warrant amortization reduced Q3 revenue by $36.4M and nine-month revenue by $82.7M without representing current cash payments. Inventory charges were $23.5M over nine months, mainly from the next-generation transition.
• Compensation: Stock-based compensation reached $155M in Q3 vs $11.3M, and $542M over nine months vs $33.8M. IPO-triggered vesting recognition explains much of the increase; further service-period and executive-award expense is expected.
• Earnings comparability: Nine-month 2025 results included a $363M gain from extinguishing a forward-contract liability, rather than operating earnings.
• Cash and funding: Nine-month OCF was $(389)M vs $(134)M, less negative than the same-period net loss principally because substantial compensation expense was noncash; working-capital investment nevertheless increased cash use. Capital expenditures reached $1.05B vs $301M. Net IPO proceeds of $6.23B helped fund expansion.
• Commitments: Uncommenced data-center leases require ≈$8.5B of undiscounted payments, far exceeding current operating cash generation. Contracted future revenue is not cash: 33% of remaining performance obligations is expected within 24 months and 44% in months 25–48, including estimated pass-through costs.

Risk Assessment

• Financial controls: Unresolved material weaknesses cover revenue, inventory, data-center assets, equity and technology controls; no material misstatement was identified, but remediation remains unproven.
• OpenAI execution and financing: Delivery failures can trigger credits or termination, while certain termination events can accelerate the related loan after infrastructure funds have been spent.
• Customer and export concentration: MBZUAI and OpenAI supplied 72% of Q3 revenue; dependence on these customers magnifies purchasing risk, while UAE export-license conditions can disrupt MBZUAI and G42 sales.
• Fixed-cost exposure: Long-term data-center commitments can outlast shorter or consumption-based customer contracts, leaving idle capacity and potentially requiring additional financing.
• Supply and competition: TSMC is the sole wafer source without a formal long-term allocation commitment; supply disruption or stronger competing platforms could impair deliveries and pricing.

What to Watch

• OpenAI deployment: Delays may extend into 2027. Capacity acceptance and reduced credits would validate execution; tranche cancellations would undermine contracted revenue.
• Broader cloud adoption: AWS agreements and Microsoft inference milestones provide additional routes to demand. Expansions would broaden utilization; stalled deployments could leave excess capacity.
• Margin recovery: Management expects dedicated-capacity ramp-up to improve margins. Improvement alongside service growth would support the model; persistent pricing and startup-cost pressure would weaken it.
• Funding runway: OpenAI loan principal is $829M, including $707M current, with repayment expected through services and permitted credits. Watch whether those offsets and infrastructure spending preserve liquidity or require further debt or equity.