Rapidly Rising Data Center Costs May Complicate CoreWeave's Push to 8 Gigawatts of AI Power by 2030

Cloud-computing specialist CoreWeave (NASDAQ: CRWV) has seen its order book grow nearly 300% year over year to a staggering $99.4 billion. At the same time, the capital required to build an artificial intelligence (AI) data center is rising, with costs estimated at $15 to $25 million per megawatt to build AI-ready facilities.

On a recent episode of the All-In Podcast, venture capitalist Chamath Palihapitiya remarked that building a modern 1-gigawatt (GW) AI data center now runs closer to $100 billion fully loaded.

As land fees rise and the scarcity of electrical engineers and components intensifies, specialized AI cloud providers like CoreWeave find themselves in a hyperinflationary infrastructure race to build scale.

Image source: Getty Images.

A purpose-built platform for AI

CoreWeave operates a vertically integrated platform designed for the performance demands of AI workloads. The company buys the latest graphics processing units (GPUs) from its partner Nvidia and combines them with other computing hardware and its own software to deliver performance that general-purpose clouds often struggle to match.

CoreWeave acts as a middleman, renting the compute capacity to AI model companies and hyperscalers. This type of operation keeps capital costs down by leasing the physical data center facilities from third-party developers rather than owning them.

The specialization has made it a key partner to major AI labs, securing multi-year, take-or-pay contracts with clients such as Meta Platforms and OpenAI. CoreWeave's customer concentration is improving, with 10 customers committed to spending at least $1 billion each, and non-investment-grade AI labs now represent less than 30% of its backlog.

The cost of rapid expansion

To fund its expansion, CoreWeave has taken on $25 billion in debt in the form of delayed-draw term loans (DDTLs). It signs a customer contract, uses it as collateral to borrow against future cash flows, and draws money only as capacity is deployed.

Lenders consider the credit-worthiness of the end customer, such as Meta, resulting in better terms. This structure has helped reduce its average borrowing rate from the mid-teens to just below 10% by the end of 2025.

Capital expenditures are expected to exceed $30 billion this year, up from $15 billion last year. The company is on a capital-intensive campaign that is sustainable only as long as it can keep signing new deals and tapping capital markets.

The rising cost of data center construction could weigh on its ability to scale economically as it attempts to grow from around one gigawatt today to 8 GW of capacity by 2030. One factor that helps balance the risk of owning depreciating hardware is the rising demand for inference, which keeps older, less powerful GPUs more valuable for longer than usual.

If AI inference remains highly profitable for providers over the next three to five years, pricing for compute capacity should hold up well. If demand underwhelms or the world moves toward low-cost open-source models, it could put pressure on compute rates for future contracts.

While CoreWeave offers a unique and compelling way for investors to play the AI infrastructure build-out, there's still plenty of uncertainty around who the winners and losers will be over the long run. As the cost of memory chips and other inputs rises, the capital burden on neoclouds and its partners gets heavier. Given the wide range of potential outcomes, investors interested in this stock should keep in mind that taking a reasonable approach to position sizing can help reduce the risk involved with any single stock in a rapidly evolving industry.

Bryan White has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Meta Platforms and Nvidia. The Motley Fool has a disclosure policy.