The bull case has merit. Enterprise ROI is measurable. Hyperscaler balance sheets are strong. Investment as a share of GDP remains below historical peaks. Efficiency gains may outpace demand growth.

The bear case has merit. Circular financing amplifies correlation risk. Thermodynamic limits are non-negotiable. Depreciation accounting may be masking massive losses. Valuation metrics are historically extreme.

What cannot be done is to pretend that one case is obviously correct and the other obviously wrong. The uncertainty is genuine. The stakes are enormous. And the resolution will unfold not in earnings calls or analyst reports but in the quiet hum of cooling systems that either work or fail, in grids that either hold or collapse, in a physical world that operates by laws no financial innovation can repeal.

[…]

In the end, this story is not about artificial intelligence. It is about the collision between human ambition and physical reality. We have built machines that think by making machines that heat.

Machines that think. Really.

This otherwise good analysis seems to be predicated on critically digging into everything - the hardware, the financial instruments, the historical parallels. Everything except the software itself. Those claims are taken at face value.