Trading cash flows for compute: the economics of AI infrastructure

AI infrastructure is becoming a capital-intensive test of demand, financing structures and corporate financial strength.

• The buildout is consuming hundreds of billions, potentially trillions, of dollars, while AI-related financing set records across capital markets, including SpaceX's largest IPO, Alphabet's largest equity raise, OpenAI's largest private funding round and Amazon's largest-ever bond issuance.
• Financing costs are rising for weaker infrastructure providers. CoreWeave's five-year bond yield rose from 10% to 15% in three months, while hyperscalers remain lightly leveraged and their returns on tangible capital stay above the long-term US equity market average.
• Amazon's AWS revenue growth accelerated to nearly 40%, its fastest pace in five years, and operating income rose 56%; rental prices for six-year-old NVIDIA GPUs are still increasing, supporting the case that computing assets can remain productive after depreciation.
• NVIDIA is helping mobilise more than $500 billion of third-party capital with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, while revenue guarantees help neoclouds obtain financing and give NVIDIA access to future computing capacity.
• The financing model carries feedback risk reminiscent of telecom vendor financing in the late 1990s, while China relies more heavily on the state, with state-affiliated sources accounting for over 60% of equity investment in its AI chip and server sector since 2023, according to Rhodium Group.
• LTGG's portfolio had approximately negative 0.3 times net debt to EBITDA at the end of September, versus positive 2.1 times for the MSCI ACWI, with gross margins of 49% versus 37% and free cash flow margins of 12% versus 9%.

Takeaway: Expensive capital should favor companies with strong balance sheets, durable cash generation and pricing power, while investors must distinguish durable AI demand from projects that depend on continued financing and identify downstream adopters such as Shopify and Samsara alongside infrastructure suppliers such as ASML, TSMC and Lam Research.