How customer contract lengths and GPU useful life affect AI infrastructure financing

Disclosure: I wrote the article linked below. I’m sharing the main argument here so the discussion does not require reading it. Of course, you are welcome to read it and comment too! AI infrastructure financing depends partly on how long GPUs can generate enough rental income to repay their purchase and financing costs. A chip can remain technically useful while less competitive on price and, especially, energy efficiency. CoreWeave's August $2.6 billion infrastructure-backed loan facility has an approximately five-year maturity, while the underlying customer contracts average three years. The company explicitly describes lenders accepting renewal risk. Source: CoreWeave’s SEC-filed announcement That difference does not automatically imply a funding shortfall. Loan amortization, collateral protections (beyond the end-user contracts) and the ability to re-lease capacity exist. Older GPUs can also remain economical for workloads that do not require the newest hardware, albeit at a lower rental price. However, lenders must assess demand beyond the initial customer commitments. Rental prices, maintenance costs and competition from newer chips will change before the debt matures. My article examines the proposal to pool and trade AI infrastructure exposures. That could broaden access to capital, but will also expose traditional investors to a risk that is more difficult to price. If you are operating or buying GPU capacity, what most often determines when older hardware stops being economical? Power consumption, memory limits, software support or the rental price of newer alternatives? I am afraid that we have seen a similar movie, almost 20 years ago. It did not end well. Full article: Wall Street Wants To Securitize AI. Have We Forgotten 2008?