CoreWeave has $104bn of AI backlog, but 79% sits beyond the next two years

CoreWeave, Inc. (Nasdaq: CRWV) ended the second quarter of 2026 with US$104.2 billion of revenue backlog, but the headline number obscures a crucial timing detail. The company expects only 21% of that backlog to be recognised as revenue within the 24 months following June 30, 2026.

That equates to approximately US$21.9 billion of expected revenue within two years. A further 39%, or about US$40.6 billion, is positioned for recognition between 25 and 48 months, while the remaining 40%, or approximately US$41.7 billion, sits more than four years into the future.

The distinction matters because CoreWeave’s backlog is equivalent to 8.1 times the US$12.8 billion midpoint of its upgraded 2026 revenue guidance. That ratio demonstrates extraordinary demand visibility, but it should not be interpreted as eight years of guaranteed revenue at the current run rate. CoreWeave is growing quickly, contracts extend across different periods, and recognition remains subject to the company delivering and keeping the required computing capacity available.

Second-quarter revenue increased 112% year over year to US$2.575 billion, while adjusted operating income rose sequentially from US$21 million to US$128 million. Management raised full-year revenue guidance to US$12.4 billion to US$13.2 billion and increased the capital expenditure forecast to US$35 billion to US$39 billion.

The central investment question is therefore no longer whether customers want CoreWeave’s computing capacity. Near-term capacity is effectively sold out, according to management. The more consequential questions concern how quickly contracted demand can be converted into active infrastructure, how much financing that conversion requires and whether operating margins can expand faster than depreciation and interest costs.

How much of CoreWeave’s US$104.2 billion backlog can become revenue within two years?

CoreWeave’s second-quarter backlog consisted of US$103.7 billion of remaining performance obligations and US$500 million of other estimated future revenue under committed customer contracts. Approximately 99.5% of the total was therefore represented by remaining performance obligations rather than the company’s broader estimate of additional contract revenue.

The backlog increased 246% from US$30.1 billion a year earlier. Sequential growth was less dramatic, rising US$4.8 billion, or approximately 4.8%, from US$99.4 billion at the end of the first quarter. Booking activity then accelerated after June, with CoreWeave disclosing more than US$25 billion of net new customer commitments during the early weeks of the third quarter.

Those later commitments were explicitly excluded from the US$104.2 billion quarter-end backlog. Adding the two figures produces more than US$129.2 billion of disclosed backlog and subsequent commitments, but it would be inaccurate to describe that entire amount as CoreWeave’s official backlog as of June 30.

The revenue-recognition schedule provides a more useful view. Approximately 60% of the backlog is expected within four years, while 40% extends beyond that period. This is unusually strong long-term visibility, but it also means the majority of the economic benefit depends on execution across several generations of GPUs, data centres and financing arrangements.

Management said more than 50% of the second-quarter backlog was already attached to contracts where customer delivery had commenced. CoreWeave expects that proportion to exceed two-thirds by the end of 2026. Delivery commencement does not mean the associated backlog is immediately recognised, but it indicates that more than half is connected to operating or ramping infrastructure rather than entirely undeveloped projects.

Why do CoreWeave’s quarterly capital expenditure figures appear to conflict?

Some reports placed CoreWeave’s second-quarter capital spending at approximately US$6.4 billion, while the company reported capital expenditure of US$9.4 billion. Both figures can be traced to official financial information, but they measure different things.

The cash-flow statement recorded US$6.422 billion of cash purchases of property and equipment during the quarter. CoreWeave’s capital expenditure measure includes additions to property and equipment, including assets acquired under finance leases, and then subtracts the change in construction in progress.

Using that definition, gross property and equipment increased by US$11.689 billion during the second quarter. Subtracting the US$2.337 billion increase in construction in progress produces capital expenditure of US$9.352 billion, rounded by the company to US$9.4 billion.

The US$9.4 billion figure is therefore the appropriate number when comparing the quarter with management’s capital expenditure guidance. The US$6.4 billion figure is appropriate only when discussing cash paid for property and equipment. Mixing the two produces a misleading picture of either cash consumption or infrastructure deployment.

CoreWeave recorded US$16.139 billion of capital expenditure during the first half. Reaching the full-year guidance range would require another US$18.9 billion to US$22.9 billion during the second half. The company expects US$11.5 billion to US$13.5 billion in the third quarter alone.

At the midpoints, the US$37 billion capital expenditure plan is 2.9 times CoreWeave’s US$12.8 billion revenue guidance. This does not mean the entire investment is charged against 2026 earnings. The assets are capitalised and depreciated over time. It does, however, illustrate how much infrastructure must be financed before the corresponding contracted revenue is fully recognised.

Can CoreWeave activate enough power to convert customer commitments on schedule?

CoreWeave expanded active power by nearly 500 megawatts during the second quarter, reaching approximately 1.5 gigawatts. Contracted power stood at approximately 3.7 gigawatts at quarter-end, meaning active capacity represented about 40.5% of the contracted total.

The company subsequently added another 500 megawatts of contracted power, taking the figure to approximately 4.2 gigawatts by August 11. That later increase expands the long-term capacity pipeline, but it does not immediately increase the amount of power generating revenue.

Management expects active power to exceed 1.85 gigawatts by the end of 2026, up from previous guidance of more than 1.7 gigawatts. More than 300 megawatts became active during June alone, which helps explain the expected acceleration in third-quarter revenue to between US$3.45 billion and US$3.60 billion.

The difference between contracted and active power is not automatically a shortfall. Contracted capacity is intentionally secured before customer delivery begins. Nevertheless, the gap identifies the main operational dependency behind the backlog: CoreWeave must align power, data-centre shells, networking equipment, cooling systems and GPU deliveries before it can begin recognising the related revenue.

The backlog may therefore be demand visibility, but active power is the conversion mechanism. Delays in any of those inputs could move revenue between quarters while depreciation, leases or financing costs begin earlier.

Why does a 59% adjusted EBITDA margin coexist with a US$626 million net loss?

CoreWeave reported US$1.510 billion of adjusted EBITDA in the second quarter, producing a 59% adjusted EBITDA margin. That measure excludes two of the largest costs created by the company’s expansion model: depreciation and interest.

Depreciation and amortisation reached US$1.393 billion, equivalent to approximately 54% of quarterly revenue. Net interest expense was US$640 million, or nearly 25% of revenue. Together, those two items represented approximately 79% of second-quarter revenue.

After operating expenses, CoreWeave recorded a US$49 million GAAP operating loss. The company then reported a US$626 million net loss, despite US$125 million of other income partially offsetting interest and tax expenses. Adjusted operating income of US$128 million provides evidence of sequential improvement, but it represented only 5% of revenue and remained below the US$200 million reported a year earlier.

CoreWeave reported US$35.068 billion of recourse and non-recourse debt at June 30, compared with US$5.524 billion of cash and cash equivalents. That implies approximately US$29.5 billion of debt net of unrestricted cash, before separately considering operating lease liabilities. Including restricted cash lifted reported liquidity to more than US$6.9 billion, but restricted balances are not equivalent to freely deployable corporate cash.

The financing burden is not expected to disappear during the next quarter. CoreWeave guided for third-quarter interest expense of US$860 million to US$940 million. At the respective guidance midpoints, interest would consume approximately 25.5% of quarterly revenue, broadly consistent with the second-quarter ratio.

Has CoreWeave reduced customer concentration enough to materially lower contract risk?

CoreWeave generated approximately 36%, 26% and 10% of second-quarter revenue from its three largest customers. Together, those customers accounted for 72% of quarterly revenue.

That represents meaningful diversification from the second quarter of 2025, when the largest customer alone produced approximately 71% of revenue. However, moving from one dominant customer to three large customers reduces concentration risk without eliminating it.

The regulatory filing does not identify which customers correspond to the 36%, 26% and 10% figures. It would therefore be unsafe to assign those percentages directly to Microsoft, OpenAI, Meta Platforms or any other disclosed customer.

CoreWeave has separately disclosed a Meta commitment initially valued at approximately US$21 billion through 2032 and an OpenAI order form worth up to approximately US$6.5 billion through May 2031. Jane Street also entered a US$6 billion commercial relationship and made a US$1 billion equity investment. These agreements demonstrate customer breadth, but CoreWeave itself expects substantial concentration among a limited number of customers to continue because of the size and duration of the contracts.

The risk is not limited to customer default. A reduction in spending, contract modification, delivery dispute or slower rollout by one large customer could materially change capacity utilisation and the timing of revenue recognition.

What does CoreWeave’s post-earnings share-price rally say about investor expectations?

CoreWeave shares closed at US$105.26 on August 14, the latest completed US trading session used for this analysis. The stock gained 19.3% on August 12, the first session after the results, before surrendering a small portion of that increase over the following two sessions.

The August 14 close represented a gain of approximately 16.1% over five trading sessions and 31.7% over one month. The 52-week change was a more modest 5.8%, illustrating how much volatility investors have absorbed during the period.

CoreWeave remained approximately 31% below its 52-week high of US$153.20 but stood about 74% above its 52-week low of US$60.55. The market reaction indicates that investors rewarded the revenue guidance increase, faster power activation, improving adjusted operating margins and more than US$25 billion of subsequent commitments.

The rally does not show that concerns about leverage have disappeared. Instead, it suggests investors became more willing to tolerate the capital intensity after receiving stronger evidence that infrastructure deployments were translating into revenue and sequential margin improvement.

What must CoreWeave prove before its backlog can outweigh leverage and execution risk?

CoreWeave’s demand position is difficult to dispute. Its official backlog exceeds US$104 billion, subsequent commitments exceed US$25 billion, near-term capacity is effectively sold out and management expects revenue to more than double in 2026.

The harder test is converting that demand without allowing financing costs to absorb the operating gains. CoreWeave must deliver more than 1.85 gigawatts of active power by year-end, increase the share of backlog attached to commenced delivery beyond two-thirds and achieve the expected low-teens adjusted operating margin during the fourth quarter.

Progress on those measures would support management’s claim that the current margin pressure is primarily caused by the timing of infrastructure deployment. Missed capacity targets, another capital expenditure increase or interest expense growing faster than revenue would challenge that argument.

The US$104.2 billion backlog is real under CoreWeave’s disclosed definition, and nearly all of it consists of remaining performance obligations. Yet only 21% is expected to become revenue within two years, while the infrastructure needed to serve the longer-dated portion must be funded much earlier.

That timing mismatch is the defining feature of CoreWeave’s model. The company possesses exceptional contracted demand, but the value ultimately created for shareholders will depend on the cost and reliability of converting that demand into powered, operating and profitable computing capacity.