TL;DR
Microsoft's $625B AI backlog (+1,150% growth) still insufficient; neoclouds now capture 10-20% of AI datacenter capex as structural supply constraints shift power and infrastructure bottlenecks.
Key Points
- Microsoft sourcing ~50% of incremental compute from third-party providers (CoreWeave, Crusoe, Nebius, NScale) despite $625B AI commitments
- Four hyperscalers collectively committing ~$700B AI capex in 2026, approaching/exceeding 100% of operating free cash flow
- Neoclouds accumulated $131B in enterprise-scale GPU commitments from zero in <3 years; now represent 10-20% of AI datacenter capex
- Anthropic operating below 98% uptime due to compute scarcity, forcing training resource diversion to maintain inference availability
Why It Matters
For infrastructure engineers and DevOps teams: compute scarcity is now a structural constraint, not cyclical. This validates multi-cloud strategies and explains why even trillion-dollar companies can't self-supply. For founders and investors: the marginal dollar of AI capacity comes from neoclouds financed by private credit and sovereign funds, reshaping infrastructure economics and creating regional compute champions.
Source: www.datagravity.dev