CoreWeave Posts Margin Inflection as Backlog Hits $104 Billion and Pricing Power Emerges Across GPU Generations
Q2 2026 earnings call, August 11, 2026
CoreWeave delivered a quarter that management is framing as an inflection point, with revenue of $2.6 billion, up 112% year-over-year, and adjusted operating income of $128 million that beat the high end of guidance. But the more consequential signal for investors is what CEO Michael Intrator described as evidence of a structurally improving pricing and margin environment: new customer contracts signed in the quarter carried contribution margins 5 to 10 percentage points above those added in recent quarters. CFO Nitin Agrawal confirmed the operating margin improvement occurred before the company implemented a roughly 25% price increase across SKUs in July, a move made in direct response to what management characterizes as sustained excess demand relative to supply.
Backlog Swells to $104 Billion, With $25 Billion More Not Yet Counted
Revenue backlog ended the quarter at $104.2 billion, up 246% year-over-year, and critically, that figure excludes more than $25 billion of net new customer commitments signed in the early weeks of the third quarter. More than half of the existing backlog is now attached to contracts where customer delivery has already commenced, a figure management expects to exceed two-thirds by year-end. This detail matters because it converts what could be viewed as speculative long-dated commitments into near-term, cash-generative revenue. Intrator was direct about the demand backdrop: "Our near-term capacity remains effectively sold out. That is translating into signed commitments on increasingly favorable terms from a broadening set of customers."
Old GPUs Are Still Money-Makers
Perhaps the most underappreciated data point from the call is what is happening to CoreWeave's legacy GPU fleet. The company disclosed it recently signed an A100 contract extending into 2029, a chip architecture NVIDIA introduced in 2020. Nitin Agrawal noted that the portion of the fleet coming up for renewal is a "very limited part" of the total, and ASPs on older-generation silicon remain at or above levels seen a year ago. This undercuts a common bear thesis around neo-clouds, namely that depreciating GPU vintages become stranded or discounted assets. Instead, CoreWeave is positioning its aging fleet as a feedstock for its fast-growing managed inference business, where older, already energized compute can be redeployed at attractive economics rather than sold at a discount.
Managed Inference Scales From Nothing to $100 Million ARR in Months
CoreWeave's managed inference platform, launched only months ago, has already generated more than $100 million in booked annual recurring revenue, up from roughly $1 million at launch, and management now guides to at least $250 million in ARR by the end of 2026. Growth in this segment is currently "constrained only by our near-term capacity" per Intrator, an admission that underscores just how supply-constrained the broader platform remains even as CoreWeave races to add power. Customers such as Grammarly and You.com are cited as having moved from experimentation into production inference workloads. Combined with other non-GPU services, storage, CPU, networking and software, these higher-margin adjacent products already exceed $400 million in ARR, giving CoreWeave a second growth engine beyond raw GPU rental.
Power Buildout Ahead of Plan, But Capital Intensity Keeps Rising
CoreWeave ended the quarter with 1.5 gigawatts of active power, having added nearly 500 megawatts in Q2 alone, with 300 megawatts of that added in June, a single month larger than any full prior quarter in company history. Contracted power stands at 4.2 gigawatts as of the call date, up from 3.7 gigawatts at quarter-end, and management reiterated its target of at least 8 gigawatts of active power by 2030. International expansion is also underway, with more than 1 gigawatt of power now contracted outside the United States, including a new 360-megawatt entry into Indonesia expected online in roughly 18 months.
The cost of this buildout is steep and rising. Full-year 2026 capital expenditure guidance was raised to $35 billion to $39 billion, and Q2 capex alone came in at $9.4 billion, slightly above the high end of prior guidance. Construction in progress jumped to $11.9 billion from $9.6 billion sequentially. Interest expense more than doubled year-over-year to $640 million, and net loss widened to $626 million from $290 million a year ago. Investors should note that even with margin expansion at the operating line, the business remains deeply capital-intensive and reliant on debt markets to fund growth ahead of revenue recognition.
Financing Innovation: Shorter-Duration Contracts Now Bankable
CoreWeave raised approximately $18 billion in Q2 across debt, convertibles and equity, including its inaugural Eurobond and its first publicly syndicated delayed-draw term loan backed by shorter-duration customer contracts, DDTL 5.5. Agrawal noted the deal priced "during one of the most dislocated weeks for credit this year" yet was completed at full size, with spreads having since retraced. This financing unlocks CoreWeave's ability to serve enterprise customers who prefer two- to three-year contract terms rather than the traditional five-year take-or-pay structure, a segment Intrator said was previously inaccessible. Total capital secured to date now exceeds $32 billion, and the company says it has cut its weighted average cost of debt by nearly 300 basis points over the past year, translating to roughly $1.1 billion in annualized interest savings.
Customer Diversification Beyond AI Labs
Management devoted significant airtime to demonstrating that demand is broadening beyond frontier AI labs. New customer wins include Caterpillar, which will use NVIDIA's Vera Rubin platform on CoreWeave's infrastructure to train physical AI models for autonomous construction equipment; Isomorphic Labs in life sciences; and systematic trading firms Flow Traders and IMC, which join a growing roster of financial services clients. In the public sector, CoreWeave is partnering with Leidos to deliver AI capabilities for defense and intelligence missions. The company also became the first cloud provider to bring up and validate NVIDIA's Vera Rubin NVL72 platform in the quarter, and third-party benchmarking firm Signal65 estimates CoreWeave's total cost of ownership can run up to 47% below the average hyperscaler.
Guidance Raised Across the Board
CoreWeave raised full-year 2026 revenue guidance to a range of $12.4 billion to $13.2 billion and adjusted operating income guidance to $960 million to $1.15 billion. Expected year-end annualized run-rate revenue was lifted to $18.5 billion to $19.5 billion. For the third quarter, the company guided revenue of $3.45 billion to $3.6 billion and adjusted operating income of $200 million to $260 million, with margins expected to reach the low teens by the fourth quarter. Interest expense guidance for Q3 climbed to $860 million to $940 million, reflecting the growing debt load required to finance accelerating deployments, a reminder that the path to profitability still runs through a heavily levered balance sheet even as the top-line and margin trajectory improve.