CoreWeave is a cloud company that rents Nvidia GPUs at cluster scale to AI labs and enterprises, and one of the companies grouped as “neoclouds”: specialist AI clouds built around GPU capacity rather than a full range of cloud services. CoreWeave said its contracted power reached approximately 4.2 GW as of August 11, 2026, and its shares trade on Nasdaq under CRWV. Nebius, Lambda and Crusoe sell similar GPU capacity with different business models, and each publishes its own capacity figures, summarized below as of October 2026.

On this page

What a neocloud sells

All four companies rent access to Nvidia accelerators in their own or leased data centers. What a customer actually buys falls into three forms:

  • Reserved clusters. Large numbers of GPUs dedicated to one customer under a multi-year contract. Microsoft’s contracts with Nebius and Lambda, described below, are of this kind.
  • On-demand instances. GPUs rented by the hour through a console or API. List prices for these appear on each provider’s pricing page; /gpu-prices tracks them with the date checked.
  • Managed software. Kubernetes, storage, monitoring, and in some cases managed inference, sold on top of the GPUs.

Their headline capacity numbers describe power, in megawatts or gigawatts, rather than GPU counts. Those figures are company totals across many sites and combine capacity in operation with capacity contracted for future years.

CoreWeave

CoreWeave calls itself “the essential cloud for AI.” Its product list includes GPU compute (“the latest NVIDIA GPUs at cluster scale”), bare-metal servers, AI-tuned object storage, a high-bandwidth cluster network, CoreWeave Kubernetes Service and Mission Control for fleet monitoring. The GPU platforms it names include Nvidia Hopper, Blackwell, GB200 NVL72 and Vera Rubin.

On September 17, 2026, CoreWeave said it had “increased its total contracted power to approximately 4.2 gigawatts as of August 11, 2026 from approximately 3.7 gigawatts” at June 30, 2026, and that it “added more than $25 billion of net new customer commitments early in the third quarter of 2026.” The release said these contracts were signed at higher prices than earlier ones and included short-dated agreements of about three to six months.

Listing: public; Nasdaq: CRWV.

Nebius

Nebius describes itself as “the AI cloud company,” building a full-stack platform “from data and model training to production deployment.” Its main products are Nebius AI Cloud, which it describes as “a cloud engineered from silicon to API” with non-virtualized GPUs and InfiniBand, and Token Factory, a managed inference service.

In its second-quarter 2026 shareholder letter of August 12, 2026, Nebius reported group revenue of $582.3 million, up 454% year over year, and annualized run-rate revenue of $3.0 billion at the end of June. It said: “We now anticipate ending 2026 with 5 GW of contracted power, up from the +4 GW we indicated last quarter.” The letter said Nebius had delivered all capacity tranches under its Microsoft contract and that capacity under a second agreement with Meta was on track for early 2027. Its listed locations include Finland, Iceland, the UK, France, New Jersey, Missouri and Pennsylvania, among others.

Nebius also leases capacity from third parties. On September 30, 2026, AIB Data Centers said it had signed Nebius for 50 MW of critical IT capacity in the southeastern US under a 12-year initial term.

Listing: public; Nasdaq: NBIS.

Lambda

Lambda calls itself “the Superintelligence Cloud” and says it serves “tens of thousands of customers.” It was founded in 2012 and “builds supercomputers for AI training and inference,” according to its own boilerplate.

Its largest announced contract is with Microsoft: on November 3, 2025, Lambda announced a “multibillion-dollar,” multi-year agreement to deploy AI infrastructure for Microsoft built on tens of thousands of Nvidia GPUs in GB300 NVL72 systems. In 2026 Lambda has announced financing rather than capacity totals: a $1 billion senior secured credit facility on May 7, 2026, and a $926 million term loan B, priced on August 12 and closed on August 27, 2026, which backs GPU servers “dedicated to an investment-grade offtaker.” On January 9, 2026, it said Nvidia Vera Rubin NVL72 systems are coming to its cloud, and on September 27, 2026 it announced a new data center at MidAmerica Industrial Park in Mayes County, Oklahoma.

Listing: private. Lambda’s August 12, 2026 release described the term loan as the first investment-grade-rated term loan B “executed by a private neocloud.”

Crusoe

Crusoe describes itself as “the industry’s first vertically-integrated AI infrastructure provider”: it designs, builds and operates data centers and also runs Crusoe Cloud, which offers GPU capacity, managed inference and fine-tuning. On September 17, 2026, it announced a $3.9 billion Series F round “at a $30.9 billion post-money valuation” and reported “6GW+ of gross contracted capacity across data centers and cloud, including 1 GW of gross capacity delivered and operational.”

Crusoe says it designed and built the Abilene, Texas campus used by OpenAI. Its 2026 projects include a 1.0 GW campus in Childress, Texas, with Lancium, announced July 15, 2026 for a hyperscale technology company, and a partnership with Aalo Atomics announced July 30, 2026 to put a modular data center beside Aalo’s 10 MWe reactor at Idaho National Laboratory in 2027.

Listing: private.

The four neoclouds compared

Company What it sells Capacity figure (company’s own wording) Source date Listing
CoreWeave GPU clusters, bare metal, storage, Kubernetes About 4.2 GW total contracted power (as of August 11, 2026) September 17, 2026 Nasdaq: CRWV
Nebius AI Cloud, Token Factory managed inference Expects 5 GW contracted power by end of 2026 August 12, 2026 Nasdaq: NBIS
Lambda GPU cloud, dedicated clusters Builds “gigawatt-scale AI factories”; multibillion-dollar Microsoft contract November 3, 2025 Private
Crusoe Data centers and Crusoe Cloud 6 GW+ gross contracted; 1 GW delivered and operational September 17, 2026 Private

The capacity figures are not on one basis. CoreWeave reports contracted power at a date; Nebius gives a year-end expectation; Crusoe reports “gross” capacity that covers both its data centers and its cloud. Ownership details, funding rounds and filings for each company are in /ai-companies, and individual campuses are in the /ai-data-centers tracker.

Sources