U.S. consumer prices rose 3.4% year over year in the latest Consumer Price Index report from the Bureau of Labor Statistics, matching market expectations and giving risk markets little new direction from inflation data. Trading attention instead moved to AI computing infrastructure, where quarterly results from CoreWeave and Nebius pointed to scarce cloud capacity, rising value for rapid deployment, and growing pressure on power supply.
CoreWeave shares closed 19% higher after its earnings release, while Nebius gained 34%. The moves followed results that showed both companies converting demand for AI computing into larger contract backlogs and faster revenue growth, even as available capacity remained limited.
CoreWeave reported quarterly revenue of $2.58 billion, twice its revenue in the comparable period a year earlier. Its backlog reached $104.2 billion, up 2.5 times over 12 months, and the company added $25 billion in newly signed contracts early in the third quarter.
Nebius reported quarterly revenue of $582 million, a 454% increase from a year earlier. The company put its AI cloud annualized revenue run rate at $3 billion and reported positive EBITDA for the first time, indicating that earnings before interest, taxes, depreciation and amortization had moved above zero.
AI cloud providers report limited available capacity
The two earnings reports offered a similar picture of the AI cloud market: customers are reserving computing capacity well before it becomes available, leaving providers with little near-term inventory to sell. That model differs from a conventional cloud-services market where buyers can often provision infrastructure on demand.
For companies building and operating GPU clusters, contracted capacity now carries greater weight than headline hardware ownership. A data center can contain expensive chips, but the revenue opportunity depends on whether it has enough electricity, networking and cooling infrastructure to put those chips to work for customers.
The constraints also give providers more leverage when customers require capacity quickly. Nebius said short-cycle, fast-delivery compute has been priced at $40 million to $50 million per megawatt, compared with $20 million to $25 million per megawatt for longer-term agreements. The difference places a premium on operators that can secure power and deliver functioning clusters without long construction timelines.
A megawatt measures the power available to operate computing equipment. In AI infrastructure, it has become a useful commercial measure because high-performance GPU deployments consume large and steady amounts of electricity. A provider with a data center site, utility access and installed cooling may be able to commercialize new capacity faster than a company that has purchased chips but is still waiting for grid connections or construction permits.
Power availability shapes pricing and expansion plans
The earnings disclosures suggest that power access is becoming a central limit on AI cloud expansion, alongside GPU supply. AI developers can order accelerators, servers and networking equipment, but those assets cannot generate revenue until they are installed in facilities with sufficient electrical capacity and heat-management systems.
That dynamic is pushing AI infrastructure operators toward a business model closer to industrial development. They must line up land, substations, transmission access, backup systems, cooling equipment and construction financing before they can sell computing services. Long-term customer contracts can help support that spending, particularly when the customer commits to paying for a defined amount of capacity over several years.
CoreWeave’s $104.2 billion backlog provides an indication of the scale of those commitments. Backlog does not equal recognized revenue, and its conversion depends on CoreWeave delivering the contracted infrastructure. Yet the size of the figure gives the company a larger base of committed demand as it finances new data-center capacity.
Nebius’s pricing figures show why fast access to powered facilities can be especially valuable. Customers facing immediate compute requirements may pay materially more than those willing to wait for longer-term deployments. That gap could support stronger margins for operators with available capacity, although it also increases the incentive for competitors to build new facilities and eventually reduce shortages.
Older GPU fleets retain commercial value
The disclosures also challenged assumptions that older AI hardware loses most of its value as newer chips arrive. Nebius said its first public auction cleared at a price 15% above its previous record quote, providing one data point that used equipment has retained demand.
CoreWeave said it signed a new contract for capacity based on Nvidia’s A100 chips extending to 2029. Nvidia released the A100 in 2020, yet CoreWeave said older fleets were largely booked at pricing comparable to, or higher than, earlier periods.
The commercial life of a GPU cluster can extend beyond the period assumed in some bearish valuation models if customers continue to use older systems for workloads that do not require the newest generation of chips. Training frontier AI models may favor the latest hardware, while inference, enterprise applications, research workloads and less demanding model development can be served by earlier generations.
Older deployments can also become more profitable after their initial capital costs have been depreciated. CoreWeave and Nebius described renewals on fully depreciated clusters as carrying relatively limited incremental expenses beyond electricity, maintenance and operations. The original financing burden does not disappear from a company’s history, but a renewed contract on an existing cluster can have a different margin profile from a newly built site.
Infrastructure operators face a more demanding test
The market reaction to CoreWeave and Nebius shows that traders are rewarding evidence of contracted demand and improving economics, rather than treating AI infrastructure as a simple proxy for GPU purchases. The stronger test will be whether operators can build capacity on schedule without allowing financing costs, construction delays or power constraints to erode the returns implied by their contract backlogs.
For cryptocurrency-linked infrastructure companies, the comparison is increasingly relevant where they own power-intensive sites, land, cooling systems or electrical interconnections. Facilities originally developed for mining may offer useful starting points for high-performance computing, but conversion requires more than available megawatts. AI customers typically need dense networking, resilient uptime, specialized cooling and long-duration contractual commitments.
The latest earnings results reinforce that physical infrastructure is becoming a larger part of the AI computing value chain. Companies that can translate power access into operating data-center capacity may command higher pricing, while operators without the capital or technical capability to upgrade sites could struggle to capture the same demand.
AI-driven markets move fast—tap into automated strategies with AI Copy Trading to mirror expert crypto traders intelligently.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
