"Neoclouds" are cloud providers built specifically for AI: they rent out GPUs at scale rather than the full menu of services AWS, Azure or Google Cloud offer. CoreWeave is the biggest. Its revenue backlog reached about $104 billion at the end of June 2026, before more than $25 billion of new commitments early in the following quarter. Q2 revenue was roughly $2.6 billion, roughly doubling year on year, with active power capacity around 1.5 gigawatts.
Why neoclouds exist
- Speed: they built GPU capacity fast when hyperscalers were capacity-constrained.
- Focus: infrastructure tuned for training and inference, like high-speed networking and liquid cooling.
- Big anchor customers: AI labs and tech giants sign multi-year, multi-billion-dollar contracts.
The risks in the model
- Debt: building data centres is capital-intensive, and interest costs are significant.
- Customer concentration: a large share of revenue from a few customers.
- Hardware depreciation: today's top GPU is next year's mid-range.
- Power: expansion depends on grid capacity, which is a growing constraint.
Renting GPUs wisely as a smaller company
- Start with model APIs. Rent GPUs only when you need custom models, fine-tuning or self-hosting.
- Compare total cost: GPU hours plus storage, egress, engineering time and idle capacity.
- Use spot or on-demand for experiments, and commit only to steady workloads.
- Avoid lock-in: containerise workloads so you can move between providers.
- Check data residency if your data must stay in India.
AI infrastructure decisions, made carefully
At NextGen Fusion, our AI development company in Lucknow, we help clients decide between APIs, GPU clouds and self-hosting, and build so they can move as prices change.

