The venture capital industry spent the past three years funding every company with "AI" and "cloud" in its pitch deck. Now the bill is coming due, and most of those bets will go to zero.
Gartner's latest forecast is blunt: by 2030, more than four in five of the specialist AI cloud providers—the so-called "neoclouds" that sprang up to offer GPU-as-a-service and model-training infrastructure—will either be acquired, merged, or shuttered. The survivors will be absorbed into the portfolios of hyperscalers or larger infrastructure players. Independence, for most, is a temporary condition.
The economics were always brutal
Running AI infrastructure at scale requires three things in abundance: capital, chips, and customers. The neoclouds have struggled with all three. Nvidia's allocation system has historically favored established buyers; financing GPU clusters requires debt at punishing rates; and enterprise customers increasingly prefer the compliance guarantees and integration depth of AWS, Azure, and Google Cloud. The specialist providers carved out niches by moving faster and offering more flexible terms, but those advantages erode as the hyperscalers invest billions in catching up.
The math is unforgiving. A single H100 cluster capable of training frontier models can cost hundreds of millions of dollars. Utilization rates must stay high to service that debt, but demand is lumpy—AI labs train in bursts, then go quiet. The neoclouds that raised at peak 2023-2024 valuations now face down-rounds or fire sales.
Who survives
Gartner's prediction implies roughly 20% of the current neocloud cohort will remain independent. The likely survivors share common traits: deep relationships with specific AI labs, proprietary software layers that create switching costs, or geographic advantages in markets where hyperscalers face regulatory friction. CoreWeave, Lambda Labs, and a handful of others have built enough scale to plausibly compete. The rest are playing musical chairs with one seat left.
The consolidation has already begun quietly. Several mid-tier providers have sold customer contracts or hardware to larger rivals in the past year. Expect the pace to accelerate as the Federal Reserve holds rates elevated and late-stage venture capital remains scarce.
Our take
The neocloud boom was always a bet that demand for AI compute would outstrip supply indefinitely. It did—until it didn't. The survivors will be those who understood that infrastructure is a scale game, not a growth-hacking exercise. For everyone else, the next four years will be a lesson in the difference between a market and a mania.
