CBRS Cerebras 10-Q 2026-06-30 section summaries

CBRS — Q2 and H1 2026

Cerebras’ revenue grew rapidly as cloud and AI-inference services expanded, but operating costs and IPO-triggered stock compensation drove substantial losses. Its May 2026 IPO transformed the balance sheet and funded a major infrastructure build-out. The company’s long-term growth opportunity is closely tied to a large OpenAI agreement, alongside significant customer, supplier and capacity commitments.

Consolidated results

Revenue increased 74.3% in Q2 and 84.2% in H1. Services drove growth, while hardware gross profit was only $0.978 million in Q2, when hardware cost of revenue exceeded hardware revenue. Customer-warrant-asset amortization reduced revenue by $44.3 million in Q2 and $46.3 million in H1; the remaining asset is expected to reduce revenue through October 2031.

The sharp expense increase included $377.008 million of Q2 stock-based compensation, versus $13.281 million, and $386.601 million in H1, versus $22.435 million. Q2 expense included $332.920 million for RSUs and $34.223 million for PRSUs; H1 expense included $273.6 million following IPO-related satisfaction of RSU liquidity vesting conditions. The 2025 comparison for other income includes a $363.336 million gain from extinguishment of a forward-contract liability. In 2026, other income included $48.839 million of Q2 and $71.571 million of H1 interest and dividend income, partly offset by OpenAI financing-component interest expense.

Revenue by geography

The company operates as one reportable segment. EMEA led H1 revenue and growth; U.S. revenue growth was strongest in Q2.

Cash flow and balance sheet

Cash, cash equivalents and restricted cash totaled $7.427 billion at H1 end, up from $930.378 million at the start of the period. Net property and equipment nearly doubled, with construction in progress rising to $604.030 million from $163.451 million. Inventory declined to $32.666 million from $63.626 million, while reserves increased to $20.1 million from $0.9 million. Charges for excess, obsolete and scrapped inventory rose to $11.8 million in Q2 and $19.2 million in H1, primarily reflecting transition to the next-generation product.

OpenAI agreement, financing and commitments

The OpenAI MRA covers 750 MW of committed inference capacity and related services over an initial three- or four-year term, extendable by OpenAI to five years. OpenAI may purchase up to an additional 1.25 GW by the end of 2030, potentially bringing total capacity to 2.0 GW. MRA revenue was $56.8 million in Q2 and $74.4 million in H1. Remaining performance obligations were $25.4 billion at June 30, substantially including OpenAI obligations: about 22% is expected to be recognized through June 2028, 43% in months 25–48, and the balance thereafter.

OpenAI funded a secured working-capital loan of $1.004571 billion in January, restricted to MRA-related infrastructure and capabilities. After $86.327 million of non-cash billings offset the balance, $918.244 million of principal remained ($736.036 million current and $182.208 million long-term). The loan bears 6% stated interest, matures no later than December 31, 2032, and qualifying MRA termination could accelerate repayment and permit collateral remedies. Its financing component generated non-cash interest expense of $19.653 million in Q2 and $38.602 million in H1. The revolving facility expanded from $250 million to $850 million after IPO conditions were met; it was undrawn at June 30, matures in 2031, and has a $150 million minimum-liquidity covenant.

Data-center capacity commitments are also expanding. Q2 agreements not yet commenced carried approximately $1.5 billion of undiscounted minimum lease payments and were not on the June 30 balance sheet; additional non-cancelable leases signed in Q3 carried approximately $753.4 million. Existing operating lease liabilities had a $568.827 million present value, against $690.304 million of future payments. H1 operating lease right-of-use assets obtained were $318.429 million, versus none in H1 2025.

IPO, dilution and key risks

The May 15 IPO issued 34.5 million Class A shares at $185 per share, including the fully exercised 4.5 million-share underwriters’ option, for approximately $6.2 billion of net proceeds. Preferred shares converted one-for-one into 124.653 million Class B shares. Common shares outstanding rose from 57.907 million at December 31, 2025 to 228.204 million at June 30, 2026. Class B shares carry 20 votes each, compared with one vote for Class A; Class N shares have no votes. At June 30, 76.011 million potential shares were excluded as anti-dilutive, including 26.108 million warrant shares, 23.797 million options, 16.532 million RSUs and 9.000 million performance RSUs. CEO and CTO awards included 9.0 million PRSUs with $531.2 million aggregate grant-date fair value, tied to market-capitalization thresholds.

Customer concentration is material: Customers A and B generated 34% and 32% of Q2 revenue and 49% and 20% of H1 revenue; at period-end, Customer B represented 57% of receivables and Customer D 19%. Three suppliers represented 13%, 12% and 10% of H1 purchases, leaving the company exposed to supply interruptions and industry-wide shortages. A putative copyright class action filed in October 2025 is being contested; the company cannot currently estimate a reasonably possible loss. In July 2026, OpenAI partially exercised its warrant, purchasing 10,033,508 Class N shares at $0.00001 per share.

CBRS — Q2 and H1 2026 MD&A

Revenue mix and consolidated results

Q2 revenue grew 74% year over year, while H1 revenue grew 84%. Growth came from increased customer consumption, new and ramping customer arrangements, dedicated cloud capacity, and related services. Cloud and other services reached 70.0% of Q2 revenue (32.0% in Q2 2025) and 55.9% of H1 revenue (31.0% in H1 2025), reflecting a substantial shift toward services. Customer-warrant-asset amortization reduced reported revenue.

Hardware revenue fell 23% in Q2, including a $28.0M reduction from warrant-asset amortization, partly offset by higher sales under existing arrangements. H1 hardware revenue rose 18%, as $53.8M of higher revenue under existing arrangements was partly offset by $29.0M of warrant amortization. Cloud and other services revenue rose 281% in Q2 and 232% in H1; warrant amortization reduced that category by $16.2M and $17.3M, respectively.

Reported gross margin contracted sharply in Q2. Hardware gross profit was $0.978M versus $23.646M, with margin of 2% versus 34%; cloud gross profit was $24.581M versus $8.453M, with margin of 20% versus 26%. For H1, hardware gross profit was $46.640M versus $44.910M (28% versus 32% margin), and cloud gross profit was $65.095M versus $28.793M (31% versus 46% margin). Management attributes reported pressure to warrant amortization, pass-through data-center costs recorded in both revenue and cost, and start-up costs to accelerate cloud capacity; it expects near-term margins to be significantly lower than recent periods and volatile. Excluding specified non-core items, management reported improved Q2 margins, attributing performance to pricing, utilization, mix, lower materials costs, and manufacturing efficiency.

Expenses, earnings, and liquidity

Stock-based compensation reached $377.008M in Q2 versus $13.281M and $386.601M in H1 versus $22.435M. H1 included $273.6M for RSUs whose liquidity condition was satisfied in connection with the IPO; remaining expense will be recognized over service periods. IPO-related compensation and payroll taxes drove much of the operating-expense increase, alongside investment in engineering and product development and higher public-company costs.

Non-GAAP core operating loss was $33.613M in Q2 versus $43.901M and $37.125M in H1 versus $63.217M; core net loss was $6.908M versus $40.543M in Q2 and $9.389M versus $55.256M in H1. These supplemental measures exclude specified items, including stock compensation and warrant amortization, and do not replace GAAP results.

Other income fell substantially because prior-year periods included a $363.3M nonrecurring gain from cancellation of a committed purchase obligation and release of a forward-contract liability. 2026 other income also reflected noncash interest expense tied to the OpenAI working-capital loan, partly offset by higher interest and dividend income.

The May 15 IPO issued 34.5M Class A shares at $185 per share and generated approximately $6.2B of net proceeds. At June 30, cash, cash equivalents, and restricted cash totaled $7.4B, with $1.2B of marketable U.S. government securities; the company also reported a $1.4B accumulated deficit and $918.2M outstanding under the OpenAI working-capital loan.

Operating cash use included a $464.5M net loss and $142.4M of working-capital cash use, offset by $559.5M of noncash charges, chiefly stock compensation. Investing cash use reflected $1.3B of investment purchases, partly offset by $514.5M of maturities and sales, as well as data-center systems and other property and equipment. Financing inflows included the IPO, $1.0B of net Series H preferred-stock proceeds, and $1.0B from the OpenAI loan.

Strategic commitments and execution risks

Under a December 2025 master agreement, OpenAI committed to purchase 750 MW of inference capacity and related services, expected to deploy in tranches during 2026–2028, and holds an option for another 1.25 GW by the end of 2030. CBRS began recognizing revenue under the arrangement in Q1 2026. The loan was funded in January 2026; CBRS has repaid part using noncash service credits and expects to continue doing so. The OpenAI ramp could increase cloud-services mix but also brings pass-through costs, start-up spending, and execution risk; revenue mix depends partly on deployment choices.

In June, CBRS signed a global hardware-leasing agreement with AWS to develop and deploy a joint compute solution in AWS data centers. Lease commencement conditions had not been met by June 30, so no revenue had been recognized. CBRS also issued an AWS warrant. The company expects R&D, sales and marketing, G&A, and public-company costs to remain elevated.

At June 30, fixed operating-lease payment obligations for data centers and offices were $690.3M. This excludes non-cancelable data-center leases executed in Q2 and Q3, whose aggregate undiscounted minimum payments are approximately $2.3B over their lease terms. The expanded $850M revolving facility matures April 14, 2031; it requires at least $150M of qualifying unrestricted cash plus undrawn commitments at each quarter-end and includes other financial and operating covenants.

Market Risk

As of June 30, 2026, cash, cash equivalents, and restricted cash totaled $7.4B, primarily in short-term money market instruments. The company expects interest-rate changes to have no material effect on these holdings; a hypothetical 10% rate change would increase or decrease other income, net, by $22.0M.

The company had $918.2M outstanding under its Working Capital Loan with OpenAI. The loan carries a fixed annual interest rate of 6.0%, insulating its contractual rate and related interest expense from market-rate changes. It may be repaid in cash or through services under the related commercial arrangement; accrued interest associated with service repayments is deemed paid under the note’s terms. A hypothetical 10% interest-rate change would not materially affect the unaudited condensed consolidated financial statements through this loan.

Foreign-currency exposure is negligible because transactions outside the reporting unit’s functional currency are few and immaterial. The Canada and India subsidiaries use local currencies as their functional currencies, while the France subsidiary uses the U.S. dollar. The company has not hedged foreign-currency exposure but may do so if that exposure increases. A hypothetical 10% change in the U.S. dollar’s relative value would not materially affect the unaudited condensed consolidated financial statements.

Controls and Remediation

As of June 30, 2026, the CEO and CFO concluded that 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 include inadequate IT general controls over financial-reporting applications, insufficient segregation of duties, and a shortage of appropriately qualified accounting, finance, and operations personnel. Related process and control gaps affect revenue recognition, inventory management and costing, data-center asset accounting, and equity administration. The company said these deficiencies had not caused a material misstatement of its financial statements.

Remediation is ongoing and includes hiring technically qualified staff; designing and implementing controls for significant accounts and disclosures, including reconciliations and journal-entry preparation and review; strengthening risk management and segregation of duties; implementing IT access and program-change controls; and engaging external SOX and technical-accounting consultants. Management reports progress, with management review and audit committee oversight, but cannot conclude remediation is complete until the measures have been implemented and tested as effective over a sufficient period. Additional actions may be necessary.

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

Part II, Item 1A — Risk Factors

Financial performance and customer concentration

Cerebras cautions that its recent growth may not continue: forecasting is difficult given its short operating history at current scale, lumpy hardware deliveries, concentrated customers and a fast-changing AI infrastructure market. Its accumulated deficit was $1.4B at June 30, 2026. The company expects substantial further investment in R&D, cloud infrastructure, data centers and public-company operations. IPO-related liquidity vesting conditions for RSUs and PRSUs significantly increased stock-based compensation.

Two customers represented 76% of accounts receivable at June 30, 2026, compared with one customer representing 78% at December 31, 2025. OpenAI’s December 2025 master agreement is expected to represent a substantial portion of revenue over several years. Lower demand, defaults, relationship or contract problems, or customers shifting purchases to competitors could materially harm results.

Cloud expansion, major agreements and funding

Cerebras historically generated most revenue from hardware systems; its inference cloud, introduced in August 2024, is still early-stage. Building data-center capacity and scaling cloud services require significant capital, while demand, utilization, monetization and eventual profitability remain uncertain. Infrastructure commitments are often long-term and costly to exit, whereas cloud customers may use services on shorter-term or consumption-based arrangements. Cancellations or weaker renewals could leave costly capacity idle; underestimating demand could constrain service and lose business. Startup customers may also fail.

Under the OpenAI agreement, Cerebras must deliver capacity tranches to specified timelines and service levels. Failures can result in credits, damages or termination of some or all of the agreement. OpenAI provided a $1.0B working-capital loan at 6% annual interest, maturing no later than December 31, 2032. Amortized repayment over three years begins after delivery of the final tranche of the initial 250 MW. Certain termination or trigger events could give OpenAI control over loan funds and require immediate repayment of outstanding principal and accrued interest.

A June 2026 AWS agreement provides for joint compute deployment, an initial multi-year system lease, options for future purchases, and pricing, exclusivity and manufacturing-capacity commitments. These commitments may constrain capacity for other customers; reduced AWS purchases could leave excess inventory or capacity, while delays or performance failures could impair the opportunity.

The company expects significant additional capital needs. Equity or convertible financing could dilute holders; borrowing could add fixed obligations, senior claims and restrictive covenants. A revolving credit facility entered into in April 2026 could divert cash to debt service and, following its Phase Two Effective Date, restrict certain debt, investments, distributions and asset transactions.

Execution, supply and competitive risks

Growth depends on scaling manufacturing, supply chains, data-center capacity, power, cooling, cloud operations, cybersecurity, staffing and customer support while meeting delivery commitments. Facility-provider delays, outages, construction or retrofit problems, power constraints and regulatory changes could delay service, raise costs or trigger customer remedies. Energy is a significant operating cost, and Cerebras may not be able to secure reliable power at reasonable prices or pass cost increases through to customers.

Cerebras uses a fabless model and relies on TSMC as its sole wafer foundry, including for specialized wafer-scale-engine manufacturing. It has no formal long-term wafer allocation commitment; many other components and manufacturing services are also sourced through purchase orders without capacity guarantees. Shortages, long lead times, yield variability, geopolitical disruption, supplier prioritization or quality issues could delay deliveries, raise costs or leave excess inventory. Switching foundries would require substantial development work. Taiwan and South Korea supply-chain exposure is heightened by geopolitical and natural-disaster risks.

The company competes with large hardware vendors, hyperscale cloud providers and specialized AI clouds, many with greater resources, capacity, supplier leverage and customer reach. Premium pricing may be pressured by cheaper alternatives or commoditization. Rapid changes in models, architectures and workloads could require costly redesign or make current products less relevant; customers may defer purchases ahead of product refreshes, use internal solutions or shift to competitors. Long sales cycles, evaluations and security reviews make revenue timing difficult to predict.

Regulation, technology and operational controls

International expansion brings export-control, sanctions, privacy, data-residency and other compliance risks. Cerebras has export licenses for CS-2, CS-3 and future CS-4 systems for G42 and MBZUAI in the UAE, subject to extensive safeguards; license delays, restrictions or revocation could disrupt sales. The EU AI Act introduces requirements with penalties of up to 7% of global revenue, while U.S. and other jurisdictions’ AI rules are evolving. Government sales may require costly certifications, security clearances and nonstandard terms.

Cerebras processes sensitive data and relies on third-party IT, cloud and data-center providers. Cyberattacks, outages, privacy failures or vendor incidents could disrupt services, expose data, cause liability and damage customer trust; insurance may not cover all losses. The company also faces IP and AI-training litigation risk, including uncertainty over copyright rules, open-source obligations and third-party licenses. Product defects or failures may require repairs, replacements, credits or damages.

As of June 30, 2026, Cerebras had material weaknesses in internal control over financial reporting, involving qualified accounting resources, GAAP application in areas including revenue, inventory, data-center assets and equity administration, and IT general controls. Remediation is underway but incomplete; failure could lead to reporting errors, restatements or reduced investor confidence. Management’s first annual internal-control report is due with the 2027 Form 10-K.

Capital structure and shareholder risks

Class B shares carry 20 votes per share. Based on beneficial ownership as of March 31, 2026, Class B holders held approximately 99.2% of total voting power; directors, executives, 5%-plus holders and affiliates owned about 38.1% of common shares but controlled 50.9% of voting power. Governance provisions may limit takeover opportunities and shareholder influence.

Up to approximately 171.1M shares could be released during the lock-up period, including up to 15.0M held by directors and officers. At June 30, 2026, outstanding options could issue 21,906,818 Class A and 5,315,618 Class B shares; outstanding RSUs could issue 15,145,010 Class A and 12,483,860 Class B shares. Sales, anticipated sales or equity awards could pressure the share price and make future financing more difficult. The company has never paid dividends and does not anticipate doing so in the foreseeable future.

Rule 10b5-1 Trading Arrangements

During Q2 2026, directors and officers adopted Rule 10b5-1 trading arrangements covering up to 5,035,667 shares, excluding a separate plan that was terminated before any sales.

Director Lior Susan adopted a separate plan on May 19, 2026, covering 417,363 shares, and terminated it on June 24, 2026; no shares were sold under the plan. Transactions under the plans may begin only after the applicable mandatory cooling-off period. Plans terminate when sales are completed or at their stated expiration. No directors or officers adopted or terminated non-Rule 10b5-1 trading arrangements during Q2 2026.