CBRS Cerebras 10-Q 2026-09-30 section summaries

Business and operating performance

Cerebras designs and manufactures wafer-scale AI compute systems for inference and other AI workloads, delivered in racks that can be deployed from individual data centers to supercomputer scale. It is rapidly expanding its cloud capacity while continuing heavy investment in R&D and infrastructure. Cloud and other services drove revenue growth, but rising costs, lower hardware revenue, and substantially higher operating expenses weighed on margins and losses.

Consolidated results

($ millions)

Revenue increased 55.3% in Q3 and 72.6% in 9M. The services category includes revenue recognized over time of $141.691 million in Q3 and $349.508 million in 9M, compared with $38.522 million and $100.276 million, respectively; the remainder was recognized at a point in time. Gross margin fell to about 20.0% in Q3 from 40.6%, and to about 26.3% for 9M from 38.1%. Cloud-service costs rose sharply, including to $107.425 million in Q3 from $32.610 million. Inventory charges for excess, obsolete, and scrapped products were $4.3 million in Q3 and $23.5 million in 9M, mainly reflecting transition to the next-generation product.

The 9M 2025 net income included $380.431 million of other income, including a $363.336 million gain from extinguishment of a forward-contract liability; that prior-period result is not directly comparable with current operating performance. In 2026, higher interest and dividend income was partly offset by non-cash interest expense on the OpenAI financing component of $18.568 million in Q3 and $57.170 million in 9M.

Cash flow, investment, and balance sheet

($ millions)

Investing cash outflow was $4.892 billion, including $5.120 billion of investment purchases, partly offset by $1.274 billion of maturities and sales. Financing provided $7.869 billion, including $6.233 billion of net IPO proceeds, $1.014 billion from Series H preferred shares, and $1.005 billion from a working-capital loan. Cash, cash equivalents, and restricted cash ended 9M at $3.519 billion, versus $930.378 million at the start of the year.

The May 2026 IPO sold 34.5 million Class A shares at $185 per share and generated approximately $6.2 billion in net proceeds. Preferred stock converted into 124.653 million Class B shares. Common shares outstanding increased from 57.907 million at year-end 2025 to 243.290 million at September 30, 2026 (150.982 million Class A, 78.593 million Class B, and 13.716 million Class N). Class B shares carry 20 votes each; Class A shares carry one vote, and Class N shares are non-voting.

Customers, contracts, and capacity expansion

The company has one reportable segment. Geographic revenue was:

The December 2025 OpenAI agreement covers 750 MW of AI inference capacity and related services over an initial three- or four-year term, extendable by OpenAI to five years. OpenAI may contract for an additional 1.25 GW by the end of 2030. OpenAI generated $72.4 million of Q3 and $146.8 million of 9M revenue, net of customer-warrant amortization. Remaining performance obligations were $26.1 billion at September 30, 2026; approximately 33% is expected over the next 24 months and 44% in months 25–48, with the balance thereafter. A significant amount relates to OpenAI. The RPO includes certain estimated pass-through costs, and its fulfillment depends on executing a large infrastructure deployment.

Revenue was concentrated: Customer A and B contributed 38% and 34% of Q3 revenue, respectively; for 9M they contributed 45% and 25%. Customer A and C are related parties to each other. At period-end, Customer B represented 64% and Customer C 16% of accounts receivable, creating customer and collection concentration risk.

The company received approximately $1.0 billion under a secured OpenAI working-capital loan in January 2026, restricted to permitted infrastructure uses. It bears 6% annual interest unless waived or deemed paid, and matures no later than December 31, 2032. After $175.489 million of non-cash billings credited against it, $829.082 million of principal remained, including $706.518 million current. Repayment may be in cash or through services and permitted credits; specified termination events can accelerate repayment. A separate revolving facility expanded to $850 million, matures April 14, 2031, and had no borrowings at period-end; it carries a $150 million minimum-liquidity covenant.

Net property and equipment grew to $1.508 billion from $437.396 million, including $1.099 billion of construction in progress. In addition, signed, non-cancelable data-center leases not yet commenced entail approximately $8.5 billion of undiscounted minimum payments; further Q4 2026 leases added approximately $217.8 million. These commitments will be recorded when leases commence and represent substantial future fixed costs.

Compensation, risks, and other developments

Stock-based compensation rose to $155.105 million in Q3 from $11.346 million, and to $541.706 million in 9M from $33.781 million. RSUs accounted for $449.890 million and PRSUs $64.791 million of 9M expense. The IPO satisfied the PRSU market-capitalization condition; remaining expense is recognized over tranche-specific service periods and may accelerate if thresholds are reached earlier. The CEO and CTO also received 1.239 million RSUs with $101.6 million aggregate grant-date fair value and 9.0 million PRSUs tied to service, performance, and $75 billion, $150 billion, and $250 billion market-capitalization thresholds.

Customer-warrant assets reached $1.1 billion, and amortization reduced revenue by $36.4 million in Q3 and $82.7 million in 9M; remaining amounts are expected to reduce revenue through October 2031. Two suppliers accounted for 12% and 11% of 9M purchases. The company also faces supply constraints, execution and margin risks as cloud services scale, and significant infrastructure and compensation costs. A putative copyright-infringement class action filed in October 2025 remains unresolved; the company intends to defend itself and cannot estimate a reasonably possible loss. U.S. deferred tax assets remain subject to a full valuation allowance.

CBRS — Q3 and nine months ended September 30, 2026

Consolidated results

Revenue grew 55% in Q3 and 73% in 9M, led by cloud and other services. That business expanded to 67.3% of Q3 revenue from 28.7%, and 60.0% of 9M revenue from 30.1%. Near-term mix and growth depend in part on OpenAI’s deployment choices; some cloud revenue includes pass-through data-center costs.

Hardware revenue declined 29% in Q3 but was down only 1% in 9M. Customer-warrant-asset amortization reduced hardware revenue by $17.1M in Q3 and $46.1M in 9M; lower deliveries reduced Q3 revenue by a further $10.8M, while higher 9M deliveries added $42.9M. Cloud and other services revenue rose 265% in Q3 and 245% in 9M, reflecting the OpenAI arrangement, new inference-services agreements, and services supporting deployed clusters. OpenAI-related and other cloud/service growth contributed $122.3M in Q3 and $212.2M in 9M; warrant amortization reduced this category’s revenue by $19.3M and $36.7M, respectively.

Gross margin pressure reflected warrant amortization, stock compensation and related tax expense, higher material costs, preferential pricing for a large strategic customer, and inventory charges. Hardware margins were 11% in Q3 and 23% in 9M; even excluding warrant amortization and specified stock-compensation costs, hardware margin declined 19 percentage points in Q3 and 2 points in 9M. Cloud/services margins were 24% in Q3 and 28% in 9M. Excluding specified non-core items, total gross margin was flat year over year in Q3 and improved 4 points in 9M; adjusted cloud/services margin improved 30 points and 11 points, respectively. Higher cloud costs include capacity, equipment, personnel, and capacity-rental costs. Management expects additional startup costs as it expedites capacity, while anticipating higher near-term gross profit and margin as dedicated cloud capacity ramps.

Operating results and cash flow

Stock-based compensation was $155.105M in Q3 versus $11.346M and $541.706M in 9M versus $33.781M. IPO completion on May 13, 2026 satisfied liquidity-based RSU vesting conditions; 9M expense included $273.6M of RSU compensation, with further accelerated-attribution expense expected over remaining service periods. Management also expects significantly higher expense from executive awards and increased public-company, compliance, and growth costs. Operating-expense increases additionally reflected hiring and investment in engineering, product and software development, next-generation products, prototypes, and legal and professional services.

Operating cash use increased due mainly to working capital for cloud growth, supplier prepayments, inventory purchases, and billing and collection timing, partly offset by customer advances and accounts-payable timing. Investing cash use reflected marketable-security purchases and increased infrastructure investment. Financing inflows were primarily from the IPO and OpenAI working-capital loan, partly offset by tax-withholding payments. The company plans substantial further data-center and systems spending.

Liquidity, financing, and risks

The total includes $850.0M available under a revolver maturing April 14, 2031. The secured facility requires at least $150.0M of quarterly covenant liquidity. The OpenAI working-capital loan is restricted to permitted infrastructure spending; $175.5M of 9M billings were applied as non-cash offsets, leaving $829.1M of outstanding principal at September 30. Repayment is expected through services and other permitted non-cash credits, which reduce cash otherwise collected; contractual cash repayment and acceleration provisions also apply. The company reported a $1.5B accumulated deficit and expects available cash, securities, and revolver capacity to fund anticipated needs for at least 12 months, while noting it may require additional debt or equity financing.

Dedicated-capacity contracts are generally take-or-pay, and customers may shift from on-demand usage to dedicated capacity as workloads grow. Capacity-delivery shortfalls under the OpenAI MRA could result in service credits or termination and reduce expected growth. Contractual commitments also include recognized data-center and office lease liabilities, non-cancelable data-center leases not yet commenced, and ordinary-course purchase commitments. Management reported no material changes to critical accounting policies or estimates since the Prospectus.

Market Risk

As of September 30, 2026, cash, cash equivalents and restricted cash totaled $3.5B, primarily in money market instruments. A hypothetical 10% change in interest rates would change other income, net by approximately $4.8M. Investments totaled $4.3B, primarily U.S. Treasury securities with remaining maturities of one year or less; a hypothetical 10% relative change in Treasury market yields would change their aggregate fair value by approximately $6.0M.

The company had $829.1M outstanding under its Working Capital Loan with OpenAI, carrying a fixed 6.0% interest rate. Repayment may be made in cash or through services under the related commercial arrangement. Because the rate does not reset with market rates, interest-rate changes do not affect the loan’s contractual rate or related interest expense.

Foreign-currency transactions outside subsidiaries’ functional currencies were few and immaterial, and the company does not hedge this exposure. A hypothetical 10% change in the U.S. dollar’s relative value would not materially affect the unaudited condensed consolidated financial statements.

Item 4 — Controls and Procedures

For the quarter ended September 30, 2026, management, including the CEO and CFO, concluded that the company’s disclosure controls and procedures were not effective at the reasonable-assurance level because of previously identified material weaknesses in internal control over financial reporting. The weaknesses relate to inadequate IT general controls over financial-reporting applications, insufficient segregation of duties, and a shortage of appropriately qualified accounting, finance, and operations personnel. Process and control gaps affect revenue recognition, inventory management and costing, data center asset accounting, and equity administration. The company stated that these deficiencies did not result in a material misstatement of its financial statements.

Remediation is ongoing. Measures include hiring technically qualified personnel; developing controls over significant accounts and disclosures, reconciliations, segregation of duties, and journal-entry preparation and review; implementing IT access and program-change controls; and engaging external SOX, technical-accounting, internal-control, and financial-statement consultants. Management reports progress under audit committee oversight, but cannot conclude that the weaknesses are fully remediated until remediation is complete and the controls have been tested and demonstrated to operate effectively over a sufficient period. Additional measures may be necessary.

Other than these remediation measures, no changes during the three months ended September 30, 2026 materially affected, or were reasonably likely to materially affect, internal control over financial reporting.

Risk factors

Growth, profitability and customer concentration

Cerebras has grown rapidly, but may not sustain that growth or scale operations, manufacturing, supply chains, data-center capacity, cloud infrastructure, customer support and public-company systems effectively. Revenue has increased substantially, while the company warns that infrastructure and product investment may outpace revenue growth and strain liquidity and operating performance.

“Core” results exclude stock-based compensation, pass-through revenue and costs, customer-warrant-asset amortization, and other adjustments; they are not a substitute for GAAP results. Accumulated deficit was $1.5B at September 30, 2026. IPO-related liquidity conditions for RSUs and PRSUs began triggering significant stock-based compensation expense. The company expects growing spending on R&D, cloud infrastructure, inference services, data centers and public-company operations, and cannot assure that these investments will produce durable revenue or profitability.

Revenue and receivables are concentrated among a small number of customers. MBZUAI represented 38% of Q3 2026 revenue (70% in Q3 2025) and 45% of 9M 2026 revenue (56% in 9M 2025); G42 represented 7% (14%) and 9% (30%), respectively. OpenAI, under a December 2025 master services agreement (MRA), represented 34% of Q3 and 25% of 9M 2026 revenue; the company expects it to remain a substantial customer for several years. Two customers accounted for 80% of receivables at September 30, 2026, versus one customer at 78% at December 31, 2025. Reduced purchases, nonpayment, disputes, or customers choosing competitors or internal solutions could materially affect results.

OpenAI commitments, infrastructure and financing

The MRA depends on delivery of data-center capacity tranches to milestones. Delivered capacity was below the contemplated amount as of September 30, 2026, before adjustments for excused delays, and forecasts indicate delays could continue into part of 2027. Service credits for uncured shortfalls began accruing October 1, 2026, subject to excused-delay adjustments. A shortfall uncured 90 days after a milestone can permit termination of the affected tranche; current forecasts put less than 5% of total committed MRA capacity at risk of termination in 2026. Termination could reduce contract value and remaining performance obligations, while committed costs may not be recoverable if capacity cannot be resold. MRA requirements for capacity, sites, security and service levels may also constrain deployment options or expose the company to credits, refunds, damages and termination.

OpenAI provided a $1.0B working-capital loan bearing 6% annual interest, subject to possible waiver. It matures no later than December 31, 2032, with amortized repayment over three years beginning after delivery of the final tranche of the initial 250 MW. Certain MRA termination or trigger events could give OpenAI control over disposition of the funds and require immediate repayment of principal and accrued interest—potentially after funds have been committed or spent and expected revenue has fallen.

Cerebras is making long-term data-center and supplier commitments while cloud customer contracts may be shorter-term, cancellable or consumption-based. Underestimating demand could constrain growth; overestimating it, or facing customer cancellations, low utilization or delayed deployments, could leave costly idle capacity and unrecoverable deposits. Facility construction, power, cooling, permitting, equipment, financing and third-party provider performance can delay delivery, increase costs and impair service. Power and data-center costs may rise, and regulatory or land, energy and water constraints can hinder expansion. The company may need substantial additional capital; equity issuance could dilute shareholders, while debt could add restrictive covenants, collateral requirements, repayment and refinancing risks. The April 2026 Revolving Credit Facility carries covenants that, after the Phase Two Effective Date, restrict certain debt, investments, liens, distributions and asset transfers, among other matters.

Customer agreements with AWS, Oracle and Microsoft create additional execution and concentration risks. AWS terms include a multi-year initial system lease, purchase options, pricing commitments, exclusivity, manufacturing-capacity guarantees and other protections that could restrict capacity for others. If expected expansions do not occur, the company could have excess inventory or capacity; Microsoft cloud inference also depends on milestones.

Technology, supply chain and competition

The business historically relied on hardware sales; its inference cloud offering, launched in August 2024, remains early-stage and requires substantial infrastructure, hardware, staffing and sales investment. Long infrastructure commitments may not match customer demand, and near-term investment may pressure cash flow and margins. Hardware revenue timing can depend on customer financing, shipment, site readiness, installation and acceptance; the company may incur costs before collection and face defaults, inventory write-downs or difficulty recovering or reselling equipment.

Cerebras relies on third parties for manufacturing and components, generally without long-term capacity or pricing commitments. TSMC is the sole wafer source for its proprietary wafer-scale engine, with no formal long-term supply or allocation commitment; switching foundries would require substantial development. Long lead times, limited supplier priority, low yields, advanced packaging, component shortages and geopolitical or natural-disaster disruption—particularly around Taiwan and South Korea—could delay products, increase costs, create excess inventory or impair margins. GPU supply is also constrained for its newer disaggregated inference approach. Supplier delays may not excuse the company from customer obligations.

Competition includes NVIDIA, AMD, Intel, hyperscalers, frontier-model developers and specialized cloud providers, many with greater capital, capacity, supplier leverage and software ecosystems. Hyperscalers and model developers are building their own solutions. Cerebras sells a premium product; lower-cost alternatives, industry overcapacity, commoditization, rapidly changing architectures, customer-specific engineering needs or model developers’ preference for incumbent platforms could erode pricing and demand. Product defects, reliability or security issues, delayed launches, or failure to support emerging models and frameworks could lead to credits, damages, lost customers and reputational harm.

Regulation, legal exposure and controls

Export controls and licensing—particularly affecting the UAE, where G42 and MBZUAI are headquartered—could delay or prevent sales, restrict customer access or disrupt supply. Cerebras has licenses for CS-2, CS-3 and CS-4 systems exported to G42 and MBZUAI, subject to security and compliance obligations. Licenses may be denied, delayed or conditioned; noncompliance could lead to penalties or loss of export privileges. Evolving AI, privacy, data-transfer, cybersecurity and government-contracting rules may add costs, constrain products or markets, or delay commercialization. The EU AI Act’s phased requirements include penalties of up to 7% of global revenue.

Cyber incidents, including those involving third-party providers, could disrupt operations, expose sensitive or customer data, trigger litigation, regulatory action and response costs, and harm reputation; insurance may not cover losses. AI-training, copyright, patent, open-source and other IP claims could require damages, licenses, redesigns or product withdrawal, and customer indemnification may be uncapped.

Material weaknesses in internal controls remained unresolved at September 30, 2026, including deficiencies in skilled review of revenue recognition, inventory and data-center accounting, equity administration, and IT general controls and segregation of duties. Remediation is underway but unproven. Management’s annual ICFR assessment begins with the FY2027 Form 10-K; reporting failures could prompt restatements, investigations or litigation.

Governance and shareholder risks

Class B shares carry 20 votes each versus one for Class A. Based on March 31, 2026 beneficial ownership, Class B holders held approximately 99.2% of post-IPO voting power; directors, executives and holders above 5%, with affiliates, held approximately 38.1% of common shares but approximately 50.9% of voting power. This structure and other governance provisions may limit Class A influence and deter takeovers.

At September 30, 2026, outstanding options could issue 15.619M Class A and 3.885M Class B shares; RSUs could issue 15.211M Class A and 10.927M Class B shares. Future awards and registered resale shares could pressure the stock price and complicate future fundraising. The company does not expect to pay dividends.

Q3 2026 Trading Arrangements

• On August 24, director Steve Vassallo adopted a Rule 10b5-1 trading arrangement to sell up to 230,000 common shares. It is scheduled to expire on May 24, 2027.
• On August 28, director Lior Susan adopted a Rule 10b5-1 trading arrangement to sell up to 376,714 common shares. It is scheduled to expire on February 26, 2027.
• Sales under either plan may begin only after the applicable mandatory cooling-off period. Each plan ends when its sales are completed or on its stated expiration date, whichever occurs first.
• No directors or officers adopted or terminated non-Rule 10b5-1 trading arrangements during Q3 2026. The disclosed plans describe potential future sales, not completed transactions.