For equity investors, AI capital expenditure ratios are nothing special: capital expenditure is capital expenditure, and as long as it generates an attractive return on investment (ROI), everything makes sense. But for credit investors, the situation is entirely different. This is because every type of collateral depreciates at different rates, and these differences can be significant.
Land and the building shell with its power supply system can be extremely durable. Even two or three decades later, after the initial equity investors have exited, a data center can still have value if the long-term power contract is attractive enough.
Grid interconnection infrastructure is one of the most durable assets in the entire project. Some new projects may wait more than five years to become operational, and may never become operational at all. However, chips are completely different. Not only are chips one of the fastest depreciating assets, but the market for financing chips is developing so rapidly that it is forming its own asset class.
In other words, a significant portion of the trillions of dollars used to support the AI construction boom is flowing into assets for which the financing market lacks sufficient long-term historical experience to refer to regarding how much value they will ultimately recover and what their residual value will be.
