The explosive demand for Nvidia GPUs has exposed a bottleneck at the big cloud providers. While hyperscalers like AWS, Google Cloud, and Azure offer vast infrastructure, their GPU allocations often fall short for companies running large-scale AI workloads. That gap has fueled a wave of "neoclouds" — smaller, specialized providers that focus on renting out high-performance chips with more flexibility and faster turnaround.
Modal, a startup that runs virtual sandboxes for AI agents, initially relied on hyperscalers but quickly hit limits. CEO Erik Bernhardsson said his company needed far more GPUs than the big clouds could provide, so it signed up with major neoclouds and now uses 25 of them. The shift is not just about raw numbers; neoclouds often offer better service and tailored hardware configurations, as video generation startup Reactor found with Nebius, which delivered the specific GPUs it needed at a good price.
Even the hyperscalers themselves are turning to neoclouds for overflow. Google and Microsoft have both started using CoreWeave, a leading neocloud, even as they compete with it. Marc Boroditsky, chief revenue officer of Nebius, said some hyperscalers have approached his company to handle customers they cannot service in time. This creates a strange dynamic where the largest cloud providers are both rivals and customers of the same startups.
Location also matters. Reactor, which generates user-created videos, needs low latency to show results instantly, so data center placement is critical. Neoclouds can often place GPUs closer to end users or in regions where hyperscalers lack capacity. The result is a fragmented but thriving market for GPU access, where neoclouds are no longer a niche but a necessary layer in the AI infrastructure stack.