CoreWeave: the landlord powering the AI boom

Some of the companies building the world's most advanced artificial intelligence systems don't own all the computers they need. They rent them.

One company supplying those computers is CoreWeave. It has grown from a cryptocurrency mining business into a specialized cloud-computing provider, building infrastructure for the growing demands of artificial intelligence. Its customers need powerful chips, enormous amounts of computing capacity, and facilities capable of keeping everything running.

The business is expanding quickly, but it requires substantial spending and borrowing. Understanding CoreWeave means understanding both what it rents out and how it pays to build it.

Before AI, there was cryptocurrency

CoreWeave was founded in 2017 as Atlantic Crypto. Its original business was cryptocurrency mining, which uses computer hardware to perform the calculations required to support certain digital currencies.

The company later shifted toward artificial intelligence. The graphics processing units, or GPUs, used in cryptocurrency mining were also well suited to AI workloads.

A GPU is a computer chip designed to perform many calculations simultaneously. That makes it useful for graphics and for the mathematics involved in modern AI. Training an AI model means processing data to adjust the model's internal parameters. Once trained, the model performs inference: generating an answer, creating an image, or completing another task when a user makes a request.

Both activities can require enormous computing resources. As AI models and usage grow, developers need more chips, networking equipment, electricity, cooling, and data-center space.

CoreWeave supplies this infrastructure so customers can rent computing capacity instead of buying and operating all the equipment themselves.

A landlord for computers

Think of CoreWeave as a landlord whose properties are computing systems rather than apartments or offices.

A conventional landlord invests in a building and rents space to tenants. CoreWeave invests in servers equipped with expensive chips, connects them through high-speed networks, and operates the infrastructure that keeps them running. Customers pay to use that capacity.

CoreWeave is often described as a neocloud provider. The term refers to cloud-computing companies that specialize in particular workloads, especially the intensive computing needed for AI. Their services are designed around providing substantial GPU capacity and the systems needed to support it.

This specialization can help customers that need powerful computing resources but do not want to purchase, install, and manage all the equipment themselves.

The landlord comparison has a limitation, however. Buildings can remain useful for decades, while AI hardware can lose its competitive edge as newer chips arrive. CoreWeave must keep its equipment useful while managing the cost of replacing it.

That makes the business capital-intensive. CoreWeave has to invest in infrastructure before all the revenue from that capacity arrives. Customer contracts, borrowing, and continued demand all play important roles in funding its expansion.

The money comes later

CoreWeave has secured several large financing arrangements. In August 2023, it announced a $2.3 billion debt financing facility led by Magnetar Capital and Blackstone. In May 2024, it announced a separate $7.5 billion facility led by Blackstone and Magnetar. In March 2026, it disclosed an $8.5 billion delayed-draw term loan facility.

These are separate arrangements, not three amounts that should simply be added together to calculate current debt. Each has its own terms, and the amount borrowed can differ from the total facility size.

A delayed-draw loan allows a borrower to access funds over time, subject to the agreement's conditions. This can help a company finance construction and equipment as projects progress.

At the end of 2025, CoreWeave reported $9.7 billion outstanding under delayed-draw loan facilities. That was only part of its borrowing, not its total debt.

The timing matters. CoreWeave may need to arrange financing, order equipment, and prepare data-center capacity before customers begin using it and payments arrive. Long-term contracts can give the company greater visibility into future demand, but the company must still cover its costs while building that capacity.

What the numbers reveal

CoreWeave's second-quarter 2026 revenue reached $2.58 billion, compared with $1.21 billion in the same quarter a year earlier.

The company also reported a revenue backlog of approximately $104 billion. Backlog represents contracted revenue expected to be recognized in the future under the relevant agreements. It is not cash already collected, and it will not all arrive immediately.

To serve customers, CoreWeave is planning major investments. It projected capital expenditures of $35 billion to $39 billion for 2026. Capital expenditures, or capex, are spending on long-lived assets such as computing equipment and data-center infrastructure.

As of June 30, 2026, CoreWeave had $35.6 billion in total debt. This broader figure should not be confused with the $9.7 billion outstanding under delayed-draw loan facilities at the end of 2025.

Interest costs are another major expense. CoreWeave reported $558 million in interest expense on debt obligations for the second quarter of 2026, after accounting for interest capitalized into assets under construction. Under accounting rules, some interest incurred while building qualifying assets is added to the cost of those assets rather than recorded immediately as an expense. The distinction affects when the cost appears in the financial statements, not whether the company has to pay the interest.

Together, the figures show a rapidly expanding business with substantial financing costs. Revenue growth and backlog matter, but CoreWeave must also pay interest, fund construction, and eventually replace equipment.

Is the money going in circles?

One question surrounding the AI infrastructure boom is how much of the money flowing through the industry represents independent customer demand and how much reflects commercial and financial relationships among companies that depend on one another.

A chipmaker might agree to buy computing capacity from a cloud provider. That provider might borrow money to purchase the chipmaker's equipment. An AI developer might then sign a long-term contract to rent the resulting capacity.

Each transaction can serve a legitimate business purpose. But the relationships can also make companies financially interdependent.

CoreWeave's agreement with Nvidia illustrates one such relationship. In September 2025, Nvidia entered into a $6.3 billion cloud-computing capacity agreement with CoreWeave. Under the agreement, Nvidia committed to purchase remaining available capacity, subject to the contract's conditions, through April 2032.

The agreement provides a potential source of demand for CoreWeave's infrastructure.

The broader question is whether end customers will ultimately pay enough for the AI services being built on this infrastructure. If demand continues to grow, these contracts can support expansion. If customers reduce spending or AI revenue grows more slowly than expected, companies may have to reassess their commitments.

This is sometimes called circular financing: companies within the same ecosystem provide one another with funding, purchases, or commitments that help sustain growth. Such relationships do not by themselves prove a business is unsound. The terms of the agreements and the underlying customer demand are what matter.

Computers don't age like buildings

New generations of GPUs can deliver better performance, making older chips less competitive even when they still work. Rental prices and resale values can also change as the market evolves.

This creates an accounting question: how long should CoreWeave spread the cost of its equipment across its financial statements?

Depreciation allocates the cost of a long-lived asset over its estimated useful life. A machine costing $120,000, for example, would generate $20,000 in annual straight-line depreciation over six years, assuming no residual value. Over four years, the annual amount would be $30,000.

CoreWeave changed the estimated useful life of certain computing equipment from five years to six years effective January 1, 2023. The company attributed the change to factors including hardware performance, software optimization, and data-center design.

A longer estimated life generally reduces annual depreciation compared with a shorter one, all else being equal. It does not change the original purchase price or establish how much the equipment can earn.

The estimate matters because depreciation affects reported earnings. It also reflects a central challenge for CoreWeave: estimating how long expensive equipment will remain useful while technology and rental prices change.

When the tenants are giants

CoreWeave serves major AI industry customers, including Microsoft and OpenAI. Microsoft is a major customer, while OpenAI is among the AI developers whose computing needs have helped drive demand for specialized infrastructure.

Large customers and long-term agreements can help a provider plan infrastructure investments by giving it greater visibility into demand.

Customer concentration also creates risk. If a major customer delays a project, changes its infrastructure plans, or uses less capacity than expected, the impact can be significant.

Contracts do not remove every uncertainty. Their terms determine when revenue is recognized, what each party must provide, and what happens if plans change.

CoreWeave must turn customer commitments into operating infrastructure and revenue that supports the cost of running the business. It must also account for changes in customer needs and competition over the life of a contract.

That is why backlog needs to be considered alongside spending, debt, and customer relationships.

Why lenders are paying attention

CoreWeave's financing needs have drawn attention from credit markets, where investors assess whether borrowers can meet their obligations and compensate lenders for risk.

CNBC reported on October 14, 2026, that yields on two CoreWeave bonds had approached 15%. A bond yield is the return investors receive at the market price they pay. Rising yields can reflect greater concern about risk or a demand for more compensation.

That does not mean the interest rate on every existing fixed-rate bond automatically rises when its market yield does. Contractual interest payments on fixed-rate bonds generally remain the same. But higher market borrowing costs can make new debt or refinancing more expensive.

CoreWeave needs financing to build capacity. Lenders and bond investors must weigh its contracts and growth prospects against its debt, interest costs, hardware obsolescence, and the possibility of weaker demand.

New tools for AI customers

On October 21, 2026, CoreWeave announced Physical AI Launchables, preconfigured environments for robotics and autonomous-vehicle development, and its Zero Trust AI Framework, a security approach designed to give customers control over access to models, data, and runtime environments. Separately, Blockspace reported on October 20 that its B200 rental index had reached a record level. That index reflects the broader rental market, not CoreWeave-specific prices.

The landlord's dilemma

CoreWeave supplies computing capacity that AI developers need, and its revenue and backlog reflect substantial demand. But the business requires large investments before all the related revenue arrives.

Its $35.6 billion in debt, $558 million in second-quarter interest expense on debt obligations after capitalized interest, and projected 2026 capital expenditures of $35 billion to $39 billion illustrate the scale of the undertaking. Its $2.58 billion quarterly revenue and approximately $104 billion backlog show the other side: rapid growth and substantial contracted revenue expected in the future.

Neither set of figures tells the whole story. Backlog is not cash in the bank, and debt alone does not establish that a company cannot succeed. The central question is whether revenue from its infrastructure can support the cost of building, financing, operating, and replacing it.

A traditional landlord manages occupancy, rent, maintenance, and financing. CoreWeave faces similar questions, but its properties are made of fast-changing technology. It must keep them useful while newer computers are already on the horizon.

That is the challenge at the heart of its business: not just building infrastructure for AI, but making the economics work over the life of the machines.

Educational information only. Not personal investment advice.