What does CoreWeave’s IPO say about the market?

Episode Summary

Executive Summary: The episode dissects CoreWeave’s IPO as a highly leveraged, asset-heavy bet on AI infrastructure demand. The hosts question its margins, debt load, customer concentration, and opaque financing, while noting that strong growth and major contracts with Microsoft and OpenAI could support the business if demand and pricing hold up. Overall, they frame it as an unusually risky IPO with many moving parts.

Main Topics: CoreWeave’s business model (Priority: 5/5): CoreWeave is described as a GPU cloud/hyperscaler that buys NVIDIA chips, houses them in data centers, and rents AI compute capacity to customers like Microsoft and OpenAI. Margins, depreciation, and profitability (Priority: 5/5): The discussion focuses on how CoreWeave presents its margins, with criticism that excluding infrastructure depreciation makes gross margins look much stronger than the economically truer figure. Debt burden and financing risk (Priority: 5/5): The hosts emphasize the company’s heavy leverage, high borrowing costs, and large near-term repayment obligations, arguing that financing costs are central to judging its viability. Customer concentration and contract durability (Priority: 4/5): CoreWeave’s dependence on Microsoft and now OpenAI is treated as a mixed blessing: strong counterparties, but also potential long-term demand uncertainty and concentration risk. NVIDIA’s role and strategic incentives (Priority: 4/5): NVIDIA appears as supplier, investor, and likely customer, raising questions about whether it has reasons to support CoreWeave’s IPO and diversify its own customer exposure. IPO risk and market skepticism (Priority: 4/5): The cut in IPO size/valuation and the complexity of the capital structure are presented as signs of weak market demand and reasons for caution around the offering. Long/short segment on tariffs and Blackstone (Priority: 2/5): The show ends with a brief market call on auto tariffs and a short on Blackstone tied to a CoreWeave-related private credit default, highlighting broader financing-market concerns.

Key Arguments: CoreWeave is not a pure software company; it is effectively a leveraged rental business for scarce NVIDIA GPUs, with data centers and software as added value. A commodity-like model can still work if supply is constrained and demand is high, but it depends heavily on pricing power and utilization. Gross margin figures can be misleading if depreciation is excluded from cost of revenue in an asset-intensive business. CoreWeave is reportedly profitable on an operating basis, but interest expense sharply reduces economic earnings. The company faces enormous refinancing and debt-repayment pressure, with $7.5 billion due by the end of next year. Microsoft is a strong counterparty because of its credit quality, but long-term renewal beyond current contracts is uncertain. Diversifying customers may reduce concentration risk, but adding OpenAI increases counterparty risk because it is not comparable to Microsoft. NVIDIA may have strategic reasons to support CoreWeave by helping diversify chip demand and ensuring more outlets for its GPUs. The IPO’s reduced size and valuation suggest limited external demand, making NVIDIA’s investment look like a backstop. CoreWeave’s success ultimately depends on whether demand for GPU compute remains strong enough over the next 5-15 years to justify its leverage and asset purchases.

Data Points: Revenue growth: ~700% in a year - Described as CoreWeave’s rapid revenue expansion in 2024 versus 2023. Gross margin (company-reported): 75% - CoreWeave’s stated margin, criticized for excluding infrastructure depreciation from cost of revenue. Gross margin (New Street estimate): ~45% - New Street Research backed out depreciation and argued the true margin is materially lower. Debt obligations due: $7.5 billion - Estimated debt obligations CoreWeave must repay by the end of next year. Loan facility default: $7.6 billion - Rob Smith said CoreWeave technically defaulted on this facility, which was Blackstone’s largest single loan commitment. Customer contract backlog: $15 billion - Amount already signed up by the end of last year. New OpenAI contracts: $12 billion - Additional contract value added from OpenAI. Equity stake: 6% - NVIDIA’s current equity stake in CoreWeave.

Pivotal Quotes: "It is effectively a giant heap of NVIDIA chips." — John Foley: Explanation of what CoreWeave actually does as an AI infrastructure rental business. "Gargantuan and labyrinthine." — Rob Smith: Description of CoreWeave’s debt structure and financing complexity. "Why would Microsoft, which is an absolutely enormous company, care whether this stuff is on its balance sheet or not?" — Rob Armstrong: Challenge to the argument that Microsoft uses CoreWeave mainly for off-balance-sheet optionality.

Implications: CoreWeave’s IPO is a test of whether investors will fund a highly leveraged AI-infrastructure model with complex financing and concentration risk. It also signals how much faith the market has in sustained AI compute demand and NVIDIA-centered economics.