AI neoclouds exist because demand for accelerator capacity grew faster than traditional cloud supply. Companies such as CoreWeave and Nscale can specialise around dense GPU infrastructure and move faster than diversified hyperscalers. That specialisation is also the source of their risk: they own expensive hardware whose economics depend on keeping utilisation high through fast technology cycles.
What the evidence establishes
Reuters Breakingviews compared the current neocloud boom with the alternative telecom networks of the late 1990s. The analogy is not that the outcomes must be identical. It is that infrastructure markets can attract enormous capital when demand forecasts are strong, creating capacity before investors know which operators will earn sustainable returns.
The commercial reading
A neocloud has three linked exposures: supplier concentration, customer concentration and financing. Nvidia or another accelerator vendor can capture much of the hardware economics; a small number of AI labs or hyperscalers may dominate revenue; and debt or lease structures can keep cash obligations high even if pricing falls. The strongest operators will need long contracts, low power cost and enough software or service differentiation to avoid becoming commodity GPU landlords.
What to watch next
Focus on utilisation, revenue backlog quality, customer concentration, hardware depreciation, debt maturities and gross margin after power and lease costs. Capacity growth alone is not proof of economic value.
How to use this analysis
Source and verification note
The reporting base for this article is Reuters: data-centre upstarts and the broadband-boom comparison. The link is provided to the source page or release so readers can check the reporting period, definitions and later revisions. Figures are not extended beyond the source's geographic or institutional scope, and forecasts remain labelled as expectations until an official release records the outcome.