CoreWeave (NASDAQ: CRWV) backlog hits $104bn as $39bn AI buildout tests leverage

CoreWeave Inc. (NASDAQ: CRWV) reported second-quarter 2026 revenue of $2.575 billion, up 112% from a year earlier, while its revenue backlog reached approximately $104.2 billion before more than $25 billion of additional customer commitments secured early in the third quarter. Near-term computing capacity is effectively sold out, allowing CoreWeave to secure new artificial intelligence infrastructure contracts on improving commercial terms while raising its full-year revenue and capital spending expectations. The company now expects 2026 revenue between approximately $12.4 billion and $13.2 billion while planning between $35 billion and $39 billion of capital expenditure to expand its power, data centre and accelerator footprint. The tension is that second-quarter interest expense reached $640 million, slightly exceeding CoreWeave’s $626 million net loss, demonstrating how heavily the growth strategy still depends on external capital. CoreWeave shares closed at $107.73 on August 12, gaining 19.3% during the session, approximately 19.8% over five trading sessions and 29.3% over one month.

Why does CoreWeave’s $104 billion backlog matter more than its 112% quarterly revenue growth?

The 112% revenue increase confirms that demand for artificial intelligence computing capacity remains exceptionally strong, but CoreWeave’s backlog provides the more important signal about what the business could look like several years from now. At approximately $104.2 billion, the backlog is more than eight times the midpoint of the company’s newly raised 2026 revenue outlook.

That comparison does not mean CoreWeave already has eight years of guaranteed revenue. The backlog includes remaining performance obligations and additional expected revenue under committed customer contracts, subject to CoreWeave delivering and maintaining the promised services. Construction delays, hardware availability or changes to contractual arrangements can affect when that revenue is recognised.

Even with those qualifications, the number provides unusual visibility for a company operating in a rapidly changing technology market. More than half of the backlog is already associated with contracts where service delivery has begun, reducing the proportion dependent entirely on future construction.

The additional $25 billion-plus of customer commitments secured in the opening weeks of the third quarter makes the demand signal harder to dismiss as an accounting artefact. Those commitments alone are almost ten times CoreWeave’s entire second-quarter revenue.

The backlog also changes the debate around potential artificial intelligence overcapacity. CoreWeave is not currently constructing speculative capacity and hoping demand eventually appears. It is confronting the opposite constraint, customers want more computing capacity than the company can immediately provide.

That gives CoreWeave pricing leverage while supply remains scarce. The risk is that this advantage can weaken rapidly if hyperscalers, specialist neocloud companies and major technology customers simultaneously bring large amounts of competing infrastructure online.

The value of the backlog therefore depends on timing. Capacity delivered during the current shortage can earn attractive returns. Capacity that arrives after the market becomes oversupplied may face weaker renewal economics even when the original contract remains profitable.

Can CoreWeave spend up to $39 billion in 2026 without losing control of its capital structure?

The most important number in CoreWeave’s earnings may be neither revenue nor backlog. It may be the company’s new capital expenditure forecast of between $35 billion and $39 billion.

At the midpoint, approximately $37 billion, expected 2026 capital expenditure would equal roughly 2.9 times the midpoint of CoreWeave’s annual revenue guidance. Very few publicly traded technology companies operate at that level of capital intensity.

Second-quarter capital expenditure alone reached approximately $9.4 billion, or about 3.6 times quarterly revenue. This is the economics of building an artificial intelligence utility rather than selling conventional software.

CoreWeave must acquire high-value accelerators, servers, networking equipment and storage while funding electrical infrastructure and data centre construction before customers can consume the resulting capacity. Revenue follows deployment, but much of the cash requirement arrives first.

The balance sheet reflects that sequence. CoreWeave ended June with approximately $5.5 billion of cash and roughly $35 billion of current and non-current recourse and non-recourse debt liabilities. Gross debt therefore exceeded unrestricted cash by about $29.5 billion before considering additional financing completed after the quarter.

CoreWeave has repeatedly demonstrated access to the debt and equity markets. It has raised more than $30 billion of debt and equity capital during 2026 and completed another $2.6 billion delayed draw term loan facility immediately before reporting second-quarter results.

That facility is particularly revealing because its approximate five-year maturity extends beyond the roughly three-year average duration of the customer contracts supporting it. Lenders are therefore accepting some renewal or re-leasing risk rather than requiring the initial customer contract to cover the entire debt maturity.

The flexibility comes at a price. The new facility carries a spread of Term SOFR plus 5.50%, illustrating that capital remains expensive despite strong customer demand.

CoreWeave can sustain the model when new contracts generate returns comfortably above financing and infrastructure costs. The balance becomes less attractive if equipment prices rise, customers gain bargaining power or refinancing becomes more expensive.

What does 1.5 gigawatts of active power against 3.7 gigawatts contracted reveal about CoreWeave’s execution burden?

CoreWeave increased active power by almost 500 megawatts during the second quarter to approximately 1.5 gigawatts. Contracted power reached approximately 3.7 gigawatts.

That means only about 40.5% of currently contracted power is active. Another roughly 2.2 gigawatts sits between commercial commitment and operating infrastructure.

This gap represents both CoreWeave’s growth opportunity and its largest execution challenge.

Securing power has become one of the principal bottlenecks facing artificial intelligence infrastructure. Developers need utility interconnections, substations, transmission capacity, backup generation, cooling systems and increasingly large volumes of equipment before accelerator clusters can be switched on.

CoreWeave’s contracted power position gives the company an advantage because customers increasingly value guaranteed delivery dates as much as raw computing specifications. A graphics processor that exists on a procurement spreadsheet but lacks electricity is not particularly productive.

The company has demonstrated improving construction velocity. Adding almost 500 megawatts during one quarter is a significant operating achievement, particularly when artificial intelligence racks require much higher power density than conventional cloud infrastructure.

However, scaling from 1.5 gigawatts toward a 3.7-gigawatt contracted portfolio is not merely a repetition of earlier construction. Larger deployment introduces more utilities, landlords, equipment suppliers, permitting jurisdictions and financing structures.

The risk becomes correlated execution. A transformer shortage, delayed utility connection or data centre construction problem can postpone several customer deployments simultaneously.

Contract terms can protect CoreWeave financially to some extent, but customers ultimately purchase computing availability rather than excuses. Repeated delays could damage its ability to win the next generation of contracts even when current backlog remains intact.

Why is CoreWeave expanding into software and cross-cloud tools instead of remaining a specialised GPU rental provider?

CoreWeave’s long-term strategy increasingly depends on becoming more difficult to replace than a supplier of rented accelerators. Hardware capacity can command exceptional pricing during a shortage, but computing eventually becomes more competitive as new suppliers enter the market.

The company is therefore expanding its active software and infrastructure portfolio around the underlying processors. CoreWeave Interconnect provides private fibre connectivity with other hyperscale clouds, beginning with Google Cloud, while SUNK Anywhere allows customers to use CoreWeave’s orchestration technology across a broader cloud portfolio.

LOTA Cross-Cloud is designed to improve access to data stored across different cloud environments, addressing one of the practical costs of moving artificial intelligence workloads between infrastructure providers.

CoreWeave has also added agentic artificial intelligence capabilities including CoreWeave ARIA and CoreWeave Sandboxes. These products address experimentation, reinforcement learning, model evaluation and secure execution environments rather than simply selling accelerator hours.

The strategic logic is straightforward. Customers become less likely to switch providers when their data access, model operations, agent workflows, observability and development processes are integrated with the same platform.

CoreWeave also completed the bring-up and validation of NVIDIA Corporation’s Vera Rubin NVL72 architecture during the quarter. Early access to new hardware remains an important competitive advantage because artificial intelligence laboratories often want new accelerator generations before general cloud capacity becomes widely available.

The challenge is keeping the software credible across a multi-cloud environment. CoreWeave cannot market flexibility while creating its own form of lock-in that customers consider more restrictive than existing hyperscalers.

A successful outcome would give CoreWeave higher-value software relationships surrounding its physical infrastructure. That could improve margins and customer retention while reducing the extent to which every future contract depends on having the cheapest or newest accelerator.

Does CoreWeave’s $640 million quarterly interest bill weaken the case for its 59% adjusted EBITDA margin?

CoreWeave generated adjusted EBITDA of $1.51 billion during the second quarter, doubling from $753 million a year earlier. The resulting adjusted EBITDA margin was 59%, an exceptionally high number for a company expanding this quickly.

The figure nevertheless requires careful interpretation because depreciation and financing are not peripheral costs in CoreWeave’s business model. They are structural consequences of owning and financing extremely expensive computing infrastructure.

Depreciation and amortisation reached approximately $1.39 billion during the quarter. Interest expense was another $640 million.

After those and other GAAP items were included, CoreWeave recorded a $49 million operating loss and a $626 million net loss. Adjusted operating income was only $128 million, equivalent to a 5% margin.

The contrast between 59% adjusted EBITDA margin and 5% adjusted operating margin explains why investors can reach dramatically different conclusions while looking at the same results.

Adjusted EBITDA highlights the cash earnings generated by deployed infrastructure before financing and asset consumption. GAAP operating results highlight the enormous cost of placing that infrastructure into service.

Neither measure should be ignored. EBITDA is valuable when assessing project cash generation and debt service capacity, while depreciation matters because artificial intelligence accelerators do not remain economically current forever.

The unusually high interest expense deserves particular attention. At $640 million, quarterly interest represented almost 25% of revenue and slightly exceeded the entire reported net loss.

That means CoreWeave’s path to sustainable net profitability depends heavily on lowering its cost of capital or generating enough operating profit to absorb a rapidly increasing financing burden.

This is why improving contract pricing is strategically important. Higher prices do not merely increase reported revenue. They create the spread required to compensate CoreWeave for the cost of borrowing billions of dollars to build the capacity customers want.

Has CoreWeave diversified its customers enough to reduce dependence on a small number of AI buyers?

CoreWeave highlighted new or expanded relationships during the quarter with customers including Caterpillar Inc., Bentley Systems, Grammarly, Databricks and several artificial intelligence developers. This broadening is important because the company was historically associated with a small number of exceptionally large technology customers.

The latest filing shows that concentration remains significant. CoreWeave’s two largest customers accounted for approximately 40% and 23% of first-half revenue, meaning roughly 63% of revenue still came from two counterparties.

That is an improvement from periods when one customer represented the overwhelming majority of revenue, but it remains high for a company carrying tens of billions of dollars of infrastructure obligations.

Customer concentration works differently under CoreWeave’s contractual structure than in an ordinary software company. Large contracts help finance dedicated infrastructure, while deposits, commitments and project financing reduce some demand uncertainty.

However, concentration still matters when contracts expire. A customer that represents a large percentage of capacity may have considerable negotiating leverage at renewal, particularly when alternative artificial intelligence infrastructure becomes more abundant.

The company’s enterprise expansion can reduce this risk when shorter-duration, higher-priced agreements become a larger part of the portfolio. The new $2.6 billion financing structure is specifically intended to support contracts of this type.

Shorter contracts improve customer diversification and pricing flexibility, but they transfer more renewal risk to CoreWeave and its lenders. That trade becomes acceptable only when the underlying accelerators retain enough commercial value to be leased to another customer.

CoreWeave is therefore moving from one concentration problem toward a more complex portfolio-management challenge. Diversification improves the customer base, but it also requires greater confidence that computing assets can be redeployed efficiently across different buyers.

Why did CoreWeave stock jump 19% after earnings despite concerns over debt and capital expenditure?

CoreWeave shares closed at $107.73 on August 12, up 19.3% from $90.32 in the previous session. The stock gained approximately 19.8% compared with the August 5 close of $89.89 and 29.3% from the July 13 close of $83.31.

The shares remain volatile. CoreWeave’s 52-week range is approximately $60.55 to $153.20, leaving the August 12 close around 29.7% below the high while still roughly 78% above the low.

The earnings reaction suggests investors placed greater weight on demand, pricing and backlog than on the higher capital spending forecast. Near-term capacity being effectively sold out reduces one of the largest fears surrounding artificial intelligence infrastructure, that massive construction programmes could arrive without enough customers.

More than seven brokerages raised price targets following the earnings release, indicating that institutional sentiment improved as revenue growth, contract visibility and pricing strengthened. Some remained cautious because the balance sheet still carries unusually high debt and because artificial intelligence infrastructure returns depend on continued scarcity.

The rally also spread across the sector. Other artificial intelligence infrastructure providers, server manufacturers and semiconductor companies advanced as CoreWeave’s numbers supported the argument that demand has not yet collapsed under the weight of hyperscaler spending.

Investor sentiment should nevertheless be classified as constructive but speculative. CoreWeave is still loss-making, highly leveraged and dependent on continuous access to large quantities of capital.

The stock’s valuation increasingly depends on execution rather than simply demand. Customers have already provided evidence that they want the computing capacity. CoreWeave must now demonstrate that each new dollar of infrastructure can be built at an attractive return without debt costs consuming the economics.

What are the key takeaways from CoreWeave’s $104 billion backlog and $39 billion capex plan?

• CoreWeave’s second-quarter revenue increased 112% to $2.575 billion as artificial intelligence infrastructure demand remained exceptionally strong.
• Revenue backlog reached approximately $104.2 billion before more than $25 billion of additional customer commitments secured early in the third quarter.
• The backlog is roughly 8.1 times the midpoint of CoreWeave’s new 2026 revenue guidance, providing unusual long-term visibility but not guaranteed revenue.
• CoreWeave expects 2026 capital expenditure of between $35 billion and $39 billion, with the midpoint equal to almost 2.9 times expected annual revenue.
• Active power reached 1.5 gigawatts against approximately 3.7 gigawatts of contracted power, leaving roughly 2.2 gigawatts of capacity still requiring activation.
• Quarterly interest expense of $640 million represented almost 25% of revenue and slightly exceeded the company’s $626 million reported net loss.
• CoreWeave’s two largest customers still generated approximately 63% of first-half revenue, leaving meaningful concentration risk despite enterprise diversification.
• The new $2.6 billion financing facility extends beyond the average duration of its supporting customer contracts, increasing flexibility while shifting more renewal risk to CoreWeave and lenders.
• CoreWeave shares gained approximately 19.8% over five trading sessions and 29.3% over one month to close at $107.73 on August 12.
• Stronger contract pricing and sold-out near-term capacity support the bull case, while financing costs, construction execution and eventual AI infrastructure oversupply remain the principal risks.

What ultimately decides whether CoreWeave becomes durable AI infrastructure or a leveraged capacity trade?

CoreWeave has answered the easiest question facing an artificial intelligence infrastructure developer, whether customers actually want its capacity. A $104.2 billion backlog, more than $25 billion of additional commitments and effectively sold-out near-term infrastructure provide compelling evidence that current demand is real.

The harder question concerns the economics of satisfying that demand. CoreWeave expects to spend up to $39 billion during 2026 while carrying about $35 billion of debt and paying hundreds of millions of dollars of interest every quarter. The company must repeatedly convert borrowed capital into computing assets that enter service quickly, command attractive pricing and remain economically useful beyond their initial customer contracts.

The second-quarter results provide early evidence of operating leverage, but they also illustrate why adjusted EBITDA alone cannot settle the investment debate. Artificial intelligence infrastructure depreciates, debt has to be serviced and every new gigawatt introduces another construction and financing cycle.

CoreWeave’s advantage today is scarcity. Its opportunity is to use that period of scarcity to build customer relationships, software and scale that remain valuable after computing capacity becomes more abundant. If it succeeds, the company could emerge as a durable specialised cloud platform with an infrastructure footprint that would be extremely difficult to replicate. If it fails, the same capital intensity that enabled the growth could turn a $104 billion backlog into one of the technology sector’s most expensive lessons in leverage.