The neocloud boom is a utilization bet, not a cheap AI trade
Retail investors are buying the dip in IREN, NBIS, CRWV, and APLD, but falling share prices have not made these businesses automatically cheap. The winners will be defined by utilization, contract quality, customer concentration, and financing discipline—not AI exposure alone.
The neocloud trade is being framed as a discounted way to own artificial intelligence infrastructure. We think that framing misses the central risk: GPU capacity only becomes valuable when customers use it consistently enough to support durable cash flow. These companies are not interchangeable AI beneficiaries; they are capital-intensive capacity businesses whose outcomes depend on who has signed the contracts, how concentrated the customer base is, and whether the financing can survive a slower ramp.
The strongest counterargument is that demand is not hypothetical. Microsoft’s cost of revenue increased by $4.8 billion, or 47%, in fiscal 2026’s third quarter because of investment in AI infrastructure, and power scarcity may keep GPU-ready capacity in demand. If supply remains constrained, even heavily financed operators could keep their assets busy enough to justify the buildout. Yes, neocloud bulls can point to those signals, along with ’s contracts, Nebius’s revenue surge, and CoreWeave’s commitment-backed revenue. But ecosystem demand does not guarantee that every operator captures attractive returns. The late-1990s fiber buildout offers the better comparison: internet demand was real, yet the winners were determined by contracts, balance sheets, and utilization rather than by exposure to the theme. Neocloud investors face the same discipline test, with GPU economics and financing costs added to the equation.
