Cerebras 10-Qquarter ended 2026-06-30

Executive Summary

Cerebras sells AI computing systems and cloud inference capacity to enterprises, research institutions and AI developers, earning hardware sales and service revenue. Its expansion centers on cloud inference, particularly contracted OpenAI capacity and a new joint-compute arrangement with AWS. Q2 revenue grew 74%, but growth slowed against H1 and gross margin fell to 14.2% as customer-warrant deductions and capacity-ramp costs weighed on results. The IPO provided ≈$6.2B to support expansion; investors now face the question of whether contracted demand can become profitable, reliably delivered services before infrastructure commitments absorb that funding.

What's Happening in the Business

Cloud and other services became 70% of Q2 revenue, versus 32% a year earlier, driven by consumption, dedicated capacity and ramping customer arrangements. OpenAI contributed $56.8M in Q2 and $74.4M in H1. Its agreement commits to 750 megawatts of inference capacity deploying during 2026–2028; another 1.25 gigawatts is optional, not committed.

In June, Cerebras signed a global hardware-leasing agreement with AWS for a joint compute solution in AWS data centers. Lease commencement conditions remained unmet at quarter-end, so it had generated no revenue. Pricing, exclusivity and manufacturing-capacity commitments could constrain resources available to other customers.

The build-out is substantial: construction in progress reached $604M, from $163M at year-end. New data-center leases signed in Q2 and Q3 carry ≈$2.3B of minimum payments outside quarter-end lease liabilities, exceeding H1 capital spending several times over. Remaining performance obligations total $25.4B, substantially including OpenAI; ≈22% is expected through 6/30/2028 and 43% over the following two years. These are future revenue obligations, not available cash.

The product transition also carries costs: H1 inventory charges reached $19.2M, primarily from moving to the next generation. Management expects near-term margins to be significantly lower than recent periods and volatile as cloud capacity ramps.

Key Financial Highlights

• Customer warrants: Amortization reduced Q2 revenue by $44.3M, including $28.0M in hardware; the remaining asset will reduce revenue through 10/2031.
• Stock-based compensation: Q2 $377M vs $13.3M; H1 $387M vs $22.4M. H1 included $274M triggered by IPO-related vesting conditions; that catch-up is distinct from ongoing compensation.
• Cash use: H1 OCF was $(47.5)M vs $(124)M. It differed from the net loss because $560M of noncash charges, chiefly compensation, outweighed $142M of working-capital cash use. Capital expenditures rose 197% to $549M, producing FCF of $(596)M.
• Funding: Quarter-end cash and equivalents were $6.74B, plus $1.18B of short-term investments. OpenAI loan principal was $918M, repayable partly through service credits; liquidity is substantial, but infrastructure spending sits outside OCF.

Risk Assessment

• Reporting controls: Material weaknesses span technology controls, staffing, revenue, inventory, data-center assets and equity administration; remediation remains incomplete, creating reporting-error risk.
• OpenAI execution: Missed capacity or service commitments can trigger credits, damages or termination, potentially accelerating the secured loan while leaving infrastructure costs behind.
• Customer and export concentration: MBZUAI supplied 49% of H1 revenue; UAE system sales depend on export licenses and safeguards, making relationship or regulatory disruption consequential.
• Supply and competition: TSMC is the sole wafer foundry without formal long-term allocation commitments; shortages or competing AI architectures could undermine delivery and pricing.
• Shareholder influence and dilution: Class B shares carry 20 votes each, while substantial awards and warrants create potential dilution with limited public-shareholder influence.

What to Watch

• OpenAI deployment: Revenue recognition has begun; subsequent filings should show capacity tranches and service revenue expanding without delivery remedies or termination.
• AWS launch: The lease has not commenced; commencement and revenue would validate the partnership, while delays could strand reserved capacity.
• Cloud economics: Gross profit is growing slower than service revenue; improving reported margins would support scalable economics, while persistent ramp costs would weaken that case.
• Control remediation: Weaknesses remain unresolved; tested remediation would strengthen reporting confidence, while persistent deficiencies or corrections would undermine it.