CoreWeave Q3 2026 Revenue Hits $3.66 Billion, Up 168% YoY, With $132.5 Billion Revenue Backlog (November 10, 2026)

Event overview ​

On November 10, 2026, CoreWeave (Nasdaq: CRWV) published financial results for the quarter ended September 30, 2026. The company described the period as a breakout quarter, citing a sequential revenue increase of more than $1 billion and adjusted operating margins reaching double digits a quarter ahead of plan.

The income statement tells two stories at once. Revenue rose from $1.365 billion to $3.661 billion, a 168% increase and the first quarter in which net new revenue exceeded $1 billion. Net loss widened from $110 million to $709 million, with the net loss margin moving from -8% to -19%.

Co-founder, chairman and chief executive officer Michael Intrator pointed to three funding sources: financing backed by customer contracts, operating cash flow, and approximately $9 billion in customer prepayments and deposits expected within the next three months.

Key facts ​

Nine-month revenue was $8.314 billion (versus $3.559 billion), and nine-month net loss was $2.075 billion (versus $715 million).

Raised full-year 2026 guidance:

1. Revenue and losses expanding together ​

Read the line items and a pattern appears: revenue grew by $2.296 billion while operating expenses grew by $2.141 billion. The incremental revenue was almost entirely consumed by incremental cost.

The two largest expense lines were technology and infrastructure at $2.025 billion (from $747 million) and cost of revenue at $1.183 billion (from $369 million). Together they account for 93% of operating expenses. What that money bought is visible on the balance sheet: net property and equipment rose from $30.557 billion at end-2025 to $56.778 billion, an increase of $26.2 billion in nine months, while operating lease right-of-use assets rose from $8.231 billion to $18.922 billion.

The loss, however, does not originate in operations. Operating income was positive at $207 million, but net interest expense of $846 million was four times that figure; adding $83 million of income tax expense produced the $709 million net loss.

Depreciation is the second pressure line: $1.889 billion in the quarter versus $630 million a year earlier. This matters for how the non-GAAP figures should be read. Adjusted EBITDA reached $2.256 billion at a 62% margin, while adjusted operating income was only $379 million at 10%. The gap between them is largely depreciation and amortization. GPU clusters are heavy assets: revenue is recognized roughly linearly, depreciation runs linearly, and the refresh cycle for equipment tends to be shorter than the accounting depreciation period. That is the structural tension in this business.

2. How to read a $132.5 billion backlog ​

Revenue backlog stood at approximately $132.5 billion as of September 30, 2026. The footnote defines it as remaining performance obligations plus other amounts the company estimates will be recognized as revenue in future periods under committed customer contracts, in each case subject to delivery and service availability requirements.

Two layers of that definition deserve separation. Remaining performance obligations are an accounting concept covering contracted but undelivered work; the additional estimated amounts are the company's own projection and therefore carry more flexibility. The figure is a multi-year cumulative contract value, not current revenue and not cash already received.

Comparing it against guidance gives only a rough scale reference: $132.5 billion is roughly ten times the midpoint of 2026 revenue guidance. That is a ratio between two quantities, not a judgment about how quickly the backlog converts — recognition depends on delivery progress, service availability and customers paying on schedule, and any of the three can shift the timing.

Media reports put the figure at $104 billion at the end of June, rising to $132.5 billion by the end of September. The disclosure does not break out customer concentration or revenue per customer, so it is not possible to determine how much of the backlog depends on a small number of large buyers.

3. A capital structure built on prepayments and contract-backed debt ​

The most consequential part of this report is not the growth rate but how the expansion is financed.

At quarter end, total liabilities were $87.456 billion against total assets of $91.481 billion, leaving stockholders' equity of $4.025 billion — a very high leverage ratio. Within that, recourse debt was $9.183 billion current and $32.417 billion non-current, while non-recourse debt was $1.389 billion current and $3.884 billion non-current.

Financing activity was dense in the quarter. The company raised approximately $4.2 billion through an upsized convertible senior notes offering, established an inaugural at-the-market equity program for up to 35 million Class A shares, and closed a $2.6 billion DDTL 5.5 facility described as its first financing facility designed to support shorter-duration customer contracts.

The three instruments differ in kind: converts push maturity out while diluting future equity, the ATM sells shares when the price suits, and the DDTL facility addresses maturity mismatch between short contracts and long debt. Together with the roughly $9 billion of customer prepayments and deposits the CEO cited, they form four channels of capital: equity, convertible debt, contract-backed lending and customer prepayment.

The cost of that structure shows up in interest. Net interest was $846 million in the quarter, and the company guides to $1.05-1.15 billion in Q4 2026 — a quarterly interest bill approaching the full-year adjusted operating income guidance of $1.275-1.375 billion.

4. Power becomes the capacity metric ​

For an AI cloud provider, how much power has been secured says more about expansion speed than how many GPUs can be bought. This quarter's release places power metrics prominently:

• More than 200 MW of active power added in the quarter, bringing total active power above 1.7 GW, double the level at the start of the year;
• Total contracted power increased by more than 900 MW to over 4.6 GW;
• International contracted power exceeded 1.3 GW, including India, Indonesia and Spain.

The year-end target is more than 1.85 GW of active power. The gap between active power (energized and deliverable) and contracted power (signed but not yet energized) is precisely the delivery burden carried into coming quarters.

Technology milestones moved in parallel: multi-rack NVIDIA Vera Rubin NVL72 deployments came online, with the first production workloads for Cognition achieving 4.8x greater token throughput than GB200 NVL72; support was announced for NVIDIA Vera, described as the first CPU designed for AI agents; and CoreWeave reported the highest server throughput among cloud providers using an NVIDIA GB300 NVL72 rack in MLPerf Inference v6.1.

The 4.8x figure applies to specific workloads against a specific comparison platform and is a vendor-reported measurement, not a general cross-platform performance multiple.

Impact and limits ​

First, these results offer a concrete sample of the AI infrastructure business model: rapid revenue growth, operating profitability achieved, and net income suppressed by interest and depreciation. Judging such a company on revenue growth alone or on net loss alone misrepresents it; operating income, adjusted EBITDA, interest and depreciation need to be read together.

Second, the combination of backlog and prepayments shows AI compute increasingly being reserved through multi-year contracts with upfront payments. For customers this locks in capacity; for the provider it trades future revenue for present financing capacity, effectively redistributing the risk of capital expenditure across time.

Third, power metrics matter more than they used to. When active power becomes the binding constraint on deliverable compute, site selection, grid interconnection and green power procurement become as important as GPU procurement. This site previously compared how Oracle, CoreWeave and Nvidia differ in AI infrastructure financing (Oracle, CoreWeave and Nvidia: AI infrastructure financing compared); these results are that structure being exercised in practice.

Limits to state explicitly:

• Adjusted EBITDA, adjusted operating income and adjusted net loss are non-GAAP measures and must not be mixed with GAAP figures; the adjusted operating margin falling from 16% to 10% does not contradict the claim of entering double digits, but the trend matters more than the single point;
• The backlog definition includes company estimates; it is neither contracted cash nor a figure that can be spread linearly across future years;
• Two capital expenditure measures exist for the quarter: $9.441 billion of purchases of property and equipment in the cash flow statement, and $14.511 billion cited by some media including finance leases and other components. The two cannot be merged or substituted;
• Customer concentration, revenue per customer, GPU depreciation life and margin per unit of compute are not disclosed. This site does not extrapolate them.

Official sources ​

• CoreWeave press release: CoreWeave Reports Strong Third Quarter 2026 Results (November 10, 2026)
• CoreWeave Q3 2026 earnings presentation (PDF)
• Sina Finance: CoreWeave Q3 revenue of $3.661 billion, up 168%, shares up 7.9% after hours (November 11, 2026)