CoreWeave

CoreWeave (NASDAQ: CRWV) is the largest pure-play AI cloud infrastructure company in the United States, purpose-built to power the training and inference workloads of the world's leading AI labs and enterprises. Founded in 2017 and headquartered in Livingston, New Jersey, CoreWeave went public in March 2025 in the largest U.S.

tech IPO since 2021, raising $1.5 billion at $40 per share. The company generated $5.13 billion in revenue in 2025 and is guiding for $12–13 billion in 2026 revenue, backed by a contracted backlog exceeding $99 billion.

Nine of the ten largest AI model providers run on CoreWeave's platform, including Microsoft (approximately 67% of FY2025 revenue), OpenAI (a $6.5 billion multi-year expansion signed in 2026), Anthropic (a multi-year deal for production-scale Claude inference), and Meta. CoreWeave specialises in NVIDIA GPU clusters—H100, H200, and Blackwell B200 systems—with full-stack software including Kubernetes-native orchestration, high-speed networking, and distributed storage purpose-built for AI.

Its 2026 planned capital expenditure of $31–35 billion underscores its position at the centre of the global AI infrastructure build-out. CoreWeave serves AI labs, foundation model providers, and enterprises requiring dedicated, high-throughput GPU compute for large-scale model training, RLHF fine-tuning, and production inference at scale across data centres in the U.S.

and Europe. Within this category the useful distinction between AI clouds is procurement shape rather than silicon, since most of them rent the same NVIDIA parts.

CoreWeave sits at the contracted end: it does publish self-serve rates — an eight-GPU NVIDIA HGX H100 node was listed at $49.24 per hour on demand and $19.71 as spot capacity when checked on 30 July 2026 — but it reserves its deepest pricing, up to roughly 60% below on-demand, for committed contracts arranged through its sales team, which is how its largest customers actually buy. Note that CoreWeave quotes per node while Lambda and RunPod quote per GPU, so those lists are only comparable once normalised.

Two caveats belong on any shortlist. That Microsoft share of FY2025 revenue is the counterparty risk in the business: the backlog is diversifying faster than recognised revenue, so the near-term financial profile still rests on a handful of customers.

And the build-out is capital-hungry — against an exit-2026 annual run-rate target of $18–19 billion, the shares fell about 10% after the 7 May 2026 first-quarter report on softer revenue guidance and higher spending, which is the market pricing execution risk on the capex rather than doubting demand. For a buyer that matters mainly as a question to ask about delivery dates on new capacity.

Best fit: labs and enterprises that can sign a multi-year, multi-megawatt commitment. Teams wanting a card for a weekend should look at Lambda or RunPod instead.