AI GPU neocloud (CoreWeave) — purpose-built GPU hyperscaler renting NVIDIA capacity to AI labs
AI exposure 100% — derived from reported segment revenue, not asserted.
AI Platforms → Hyperscalers & Cloud Platforms · figures rebased to Q2 2026, reported 2026-08-11
Pure-play AI GPU cloud operator — operates the platform layer AI is delivered through; Core Scientific DC-ownership deal terminated Oct 2025 so no B2 colocation tag
CoreWeave ($2.575B Q2 2026, +112% YoY; FY2026 guidance raised to $12.4-13.2B) — AI GPU neocloud. Specialized cloud renting NVIDIA GPU clusters for AI training/inference under long-term, largely take-or-pay contracts. Primary B6 Hyperscalers & Cloud Platforms. Revenue backlog $104B at Jun 30 2026 rising to $129.2B by the Aug 11 call; active power 1.5 GW (+~500 MW in the quarter) against ~4.2 GW contracted, targeting 8 GW+ by 2030; FY2026 capex raised to $35-39B. NICHE: AI GPU neocloud — purpose-built GPU hyperscaler (vs general AWS/Azure); NVIDIA holds an equity stake and a $6.3B take-or-pay…
Moat. Scarce inputs, not lock-in: preferential NVIDIA allocation (equity stake plus a $6.3B take-or-pay GPU backstop to 2032), ~4.2 GW of contracted power and 1.5 GW active after adding ~500 MW in a single quarter. Real assets — but the advantage is capital access and speed to deploy, and Nebius, Lambda and the hyperscalers are contesting it with the same playbook and, in Nebius's case, cheaper money.
Bottleneck / pricing power. Pricing power is real and improving — management disclosed that new contracts land at contribution margins 5-10 points above recent quarters, and a DDTL 5.5 facility now lets CoreWeave underwrite shorter two-to-three-year enterprise deals that price better than five-year hyperscaler contracts. It is still scarcity rent rather than structure: the $129.2B backlog fixes revenue, not returns, and Q2 net interest expense of $640M absorbed ~42% of adjusted EBITDA.
Role in the AI stack. The purpose-built GPU-neocloud layer — highest-beta AI-capex play.
All 174 companies · What moved this quarter · How the map is built