The economics of CoreWeave’s capital expansion and GPU pricing

Debt scaling and financial pressure

CoreWeave secured more than $30 billion of debt and equity capital this year to support its expansion, including a $2.6 billion delayed draw term loan facility priced at SOFR plus 5.50 percent. This transaction follows a $3.1 billion delayed draw term loan facility completed earlier this year. Blackstone leads a $7.5 billion debt financing facility that includes strategic participation from Magnetar and Coatue. Other participants in the financing include Carlyle, CDPQ, DigitalBridge Credit, funds and accounts managed by BlackRock, Eldridge Industries, and Great Elm Capital Corp. JPMorgan and Mitsubishi UFJ Financial Group served as joint lead arrangers and bookrunners for the $2.6 billion facility. The company faces mounting financial pressure from its rapid expansion. Operating expenses jumped nearly fourfold to $1.19 billion in the second quarter. Analysts from D.A. Davidson claim CoreWeave does not generate enough profit to pay all its debt holders, certainly not equity holders. This financial strain coincides with a stock that fell 11 percent on Wednesday following a larger than expected loss. The $2.6 billion facility carries an approximate five-year maturity, which exceeds the average three-year length of underlying customer contracts. Lenders use this structure to signal confidence in the long-term value of Nvidia GPUs and to underwrite renewal risk. The facility received ratings of Ba2 from Moody’s and BB+ from Fitch. CoreWeave also expanded its data center presence from three to 14 locations in 2023, and its stock price rose nearly three times since its March IPO.

Contract diversification and revenue shifts

The company diversifies its revenue through massive multi-year agreements. Meta signed a $14.2 billion contract to access Nvidia GB300 systems through 2031, with an option to extend to 2032. This deal follows a $22.4 billion contract with OpenAI, which grew from an initial $11.9 billion deal in March and an additional $4 billion in May. Last week, the agreement with OpenAI expanded by another $6.5 billion, bringing the total contract value to approximately $22.4 billion. CoreWeave also provides computing capacity for large US market makers, including Jane Street Group and Hudson River Trading. You should watch the impact of these shifting capital structures on cloud pricing. The company plans to invest $1.25 billion in its new European headquarters in London. While demand for AI infrastructure remains high, the company faces the necessity of managing high-performance compute costs to execute these existing contracts. Demand for AI infrastructure remains high, as Microsoft reported commercial remaining performance obligations of $678 billion, a figure up 84 percent year over year. Meta also plans to spend up to $72 billion on AI infrastructure in 2025 alone. Meta also plans to acquire the chip start-up Rivos to strengthen its own semiconductor abilities and invested $14.3 billion into Scale AI for a 49 percent stake in the startup.

GPU pricing and the capacity squeeze

Pricing differences between hyperscalers and specialized providers remain substantial. CoreWeave is cheaper on all three GPUs both rent, maintaining lower rates than Microsoft Azure for the most sought-after hardware.

On the NVIDIA H100 SXM5 80GB, Microsoft Azure charges $11.06 per hour, whereas CoreWeave charges $6.16 per hour. The NVIDIA H200 SXM 141GB shows a similar disparity, with Azure at $10.60 per hour and CoreWeave at $6.30 per hour. Azure prices the A100 SXM4 80GB at $4.10 per hour, while CoreWeave charges $2.70 per hour. These differences reflect the cost advantages of specialized providers. The movement of Blackstone infrastructure funds this month changes the pricing landscape as specialized providers compete with hyperscalers. The industry faces a capacity squeeze, as US data center capacity will grow from 24 GW to 110 GW by 2030. Will the company maintain its ability to service this debt as competition intensifies?