Rising anxiety that mounting debt will cause an imminent collapse of the AI industry is misplaced. As my colleague Liz Hoffman noted, and as an unnamed data‑center financier put it bluntly: “We’re not an ATM for your dystopian future.” That quote captures lenders’ impatience, but lack of money alone is unlikely to be the factor that derails the AI train.
The real constraints: supply chains and infrastructure
Growth today is limited by multiple bottlenecks. It’s not only shortages of chips, memory and EUV machines: large numbers of refrigeration units are required to cool GPUs, and raw materials such as glass are needed to make certain components. Were these supply‑chain constraints absent, revenues and user adoption would likely be expanding much faster.
Each time firms improve token efficiency, new AI use cases emerge that consume still more compute — an illustration of Jevons paradox. A recent example is decompiling video games to create mashups: tasks that once took humans years can now be completed by AI over a weekend. That kind of capability may be what persuades younger generations to embrace AI broadly.
Market size and profitability
The total addressable market for AI is effectively the planet’s population; demand is expected to continue rising. Enterprises have only just begun experimenting with AI, and consumer adoption is even less mature. Anthropic told investors this week that it operates with 85% gross margins; even if that figure is overstated, AI services can be highly cash generative, and companies should be able to service their debt.
Why borrow rather than raise equity?
I share the lender’s frustration, but perhaps for a different reason: why lend money for modest interest when equity investment can deliver much larger returns? Hyperscalers and new cloud providers took on debt in part because it was relatively cheap capital.
Current practical bottlenecks
Bloomberg reports that data‑center companies identify talent and power as two of the main constraints to AI buildout today. This underlines that physical and human resource availability — not only financial liquidity — are the more immediate limits on growth.
Conclusion
The current credit nervousness is understandable but does not prove an impending AI market collapse. Infrastructure and supply‑chain limitations, together with shortages of skilled workers and energy, offer a more plausible explanation for growth constraints. At the same time, the sector’s vast potential user base and reports of high gross margins from some players indicate the industry can remain profitable even as investment patterns and financing sources evolve.
