The top 10 Neocloud companies in the world in 2026 include CoreWeave, Crusoe, IREN, Nebius, Lambda, Fluidstack, Vultr, Northern Data, WhiteFiber, and Nscale. These companies are among the leading providers of GPU cloud infrastructure, AI computing, and high-performance computing services, with operations and expansion plans spanning multiple global markets.

Neocloud companies have evolved from niche GPU rental providers into major players in the AI infrastructure industry. They are securing multi-billion-dollar contracts with technology companies such as Microsoft, Meta, Anthropic, and OpenAI while investing heavily in GPU clusters and data center infrastructure.

This article examines the ten leading Neocloud companies worldwide, covering their financial performance, GPU fleets, data center footprints, major customers, strategic partnerships, and infrastructure expansion plans.

What Are Neocloud Companies?

Neocloud companies are cloud providers built specifically to rent GPU computing power for artificial intelligence, a model often called GPU-as-a-Service (GPUaaS). Instead of offering hundreds of general cloud services like AWS or Azure, a neocloud focuses on large GPU clusters, high-density data centers and long-term contracts for AI training and inference.

The name simply means "new cloud": a generation of providers that emerged around 2022-2023, when demand for NVIDIA GPUs outstripped what the established clouds could supply. Many began life as cryptocurrency miners with spare power and GPU-operating experience; CoreWeave, IREN, Crusoe, Northern Data and Nscale all trace their roots to mining.

Most neocloud companies share four traits:

• GPU-first fleets: NVIDIA H100, H200, Blackwell (B200, GB200, GB300) and Vera Rubin systems, with some running AMD Instinct GPUs or Google TPUs.

• Purpose-built data centers: liquid-cooled, high-density sites designed for tens of thousands of GPUs working as one cluster.

• Long-term, contracted revenue: multi-year take-or-pay agreements with hyperscalers and AI labs, alongside on-demand GPU rental.

• Speed to capacity: faster access to scarce chips and power than many enterprises can arrange themselves.

Well-known examples of neocloud companies include CoreWeave, Nebius, Lambda, Crusoe and Together AI.

Neocloud vs Hyperscaler: What's the Difference?

The difference between hyperscalers and neoclouds is breadth versus specialization. Hyperscalers sell a full menu of cloud services to every kind of customer; neoclouds concentrate on GPU compute for AI and compete on availability, performance and price.

List of Top 10 Neocloud Companies in the World 2026

CoreWeave

CoreWeave is the largest pure-play AI cloud, renting NVIDIA GPU capacity to AI labs, hyperscalers and enterprises, mostly on multi-year contracts. It is headquartered in Livingston, New Jersey, and was founded in 2017 by Michael Intrator, Brian Venturo, Brannin McBee and Peter Salanki, who began by mining Ethereum on GPUs. It listed on Nasdaq (CRWV) in March 2025.

GPU Infrastructure

CoreWeave had over 250,000 NVIDIA GPUs deployed by end-2025. It was among the first clouds to offer HGX H100 and H200 at scale and runs GB200 Grace Blackwell systems on Quantum-2 InfiniBand, with Vera Rubin planned for new sites such as India. Beyond GPUs, its storage, CPU, networking and software products passed USD 400 million ARR in Q2 2026.

Data Center Footprint

As of 30 June 2026 it ran 51 data centers with more than 1.5 GW of active power and about 4.2 GW of contracted power. Most sites are in the US, with others in the UK and continental Europe, including a 22.5 MW deployment in Copenhagen, Denmark. It has announced entry into Indonesia and India.

Major Customers and Partnerships

Microsoft accounted for about 67% of 2025 revenue. Other major customers include OpenAI and Meta. NVIDIA is both supplier and investor, adding a USD 2 billion investment in January 2026. Facility partners include Core Scientific and AdaniConneX.

Expansion Plans

CoreWeave guides 2026 capex of USD 35-39 billion and expects more than 1.85 GW of active power by end-2026, with a target of 8 GW by 2030. Announced projects include a 240 MW campus in Navi Mumbai with AdaniConneX (three 80 MW buildings), three facilities in Indonesia, a USD 6 billion Pennsylvania data center, and further UK investment. It also launched a global AI network backbone linking its sites.

Crusoe

Crusoe is a vertically integrated AI infrastructure company that develops power, builds data centers and runs its own GPU cloud. It is headquartered in Denver, Colorado, and was founded in 2018 by Chase Lochmiller (CEO) and Cully Cavness. It began by using stranded natural gas for Bitcoin mining and sold that mining business to NYDIG in March 2025. 

It earns money in three ways: leasing data center space to customers who bring their own GPUs, renting its own GPUs, and selling inference.

GPU Infrastructure

Crusoe cloud runs NVIDIA GPUs, and it has strategic partnerships with both NVIDIA and AMD. Products include Managed Inference and AutoClusters, and its Spark line of factory-built modular data centers can be trucked to a power source.

Data Center Footprint

Crusoe reports more than 1 GW operational out of over 6 GW of contracted capacity. Its flagship is a 1.2 GW campus in Abilene, Texas and by June 2026 two buildings were live and six under construction. It operates mainly in the US (Texas, Wyoming, Missouri) and is developing gas-powered sites in Alberta, Canada.

Major Customers and Partnerships

Abilene is used by Oracle to host OpenAI workloads as part of the Stargate project. Other customers include Microsoft, Meta and, per the Wall Street Journal, Google. In 2026 it reportedly signed a USD 13 billion, five-year cloud contract with Jane Street. NVIDIA is an investor and technology partner, and Brookfield has provided credit.

Expansion Plans

Crusoe says its projects carry more than USD 140 billion in contracted value. Underway are a second, 900 MW Abilene campus for Microsoft, a 1.4 GW site in Texas, two Missouri sites and the Alberta projects. The Series F will also scale Spark modular units, and Crusoe has reportedly held IPO talks with Goldman Sachs and Morgan Stanley.

IREN

IREN is a vertically integrated data center operator that is converting from Bitcoin mining to AI cloud services. It was founded in Sydney, Australia, on 6 November 2018 as Iris Energy by brothers Daniel and William Roberts, who are co-CEOs. It sells bare-metal and managed GPU cloud to hyperscalers, frontier labs and AI data center developers.

GPU Infrastructure

IREN says its Microsoft contract scales its footprint to about 100,000 GPUs. That contract uses NVIDIA GB300 systems bought from Dell for about USD 5.8 billion. A further USD 1.6 billion Dell order covers air-cooled Blackwell systems, and about 74,000 GPUs are planned across its British Columbia and Childress sites. It is an NVIDIA Preferred Partner.

Data Center Footprint

At 30 June 2026 IREN ran about 40 MW of AI Cloud capacity and about 380 MW of Bitcoin mining capacity it is converting to GPUs. It holds about 3 GW of secured grid-connected power. Its sites are the 750 MW Childress campus and the planned 2 GW Sweetwater hub in Texas, plus several sites in British Columbia, Canada, including Prince George.

Major Customers and Partnerships

Microsoft signed a five-year, USD 9.7 billion GPU cloud contract in November 2025. NVIDIA holds a five-year, USD 3.4 billion managed AI cloud contract and has a 5 GW strategic partnership with IREN. Other customers include Cohere, Perplexity, Figure AI, Fal AI and a frontier AI lab signed in August 2026. Dell supplies its GPU systems.

Expansion Plans

IREN is building four liquid-cooled facilities, Horizon 1-4, at Childress with 200 MW of IT load for Microsoft and the first 50 MW has been delivered. The NVIDIA contract's Blackwell systems are due to commission in early 2027. At Sweetwater 1 the substation is energized and about 300 MW is under construction for delivery in Q4 2027. IREN aims to finish converting its mining sites by 31 December 2026. Texas grid rules and an ERCOT energization pause could delay parts of this.

Nebius Group

Nebius is a full-stack AI cloud offering GPU compute plus software for training and inference. It is headquartered in Amsterdam and is led by founder and CEO Arkady Volozh, a Yandex co-founder. It took its current form in 2024, when Yandex N.V. sold its Russian business and kept the Finnish data center, the Nebius AI unit, Toloka, TripleTen and Avride. It is listed on Nasdaq (NBIS).

GPU Infrastructure

The Microsoft contract alone covers more than 100,000 NVIDIA GB300 GPUs. It runs B200 clusters in Kansas City, a 4,000-GPU B300 cluster at Ark's Longcross Park in the UK, and Blackwell systems in Finland. Vera Rubin is currently planned for Meta.

Data Center Footprint

Nebius had about 170 MW of active power at the end-2025 and sells out new capacity as it comes online. It operates in Finland (Mäntsälä, 75 MW), the UK, Iceland, France, Israel and the US (Vineland, New Jersey, and Kansas City, Missouri). Vineland has a design capacity of up to 300 MW with an option for 400 MW more.

Major Customers and Partnerships

Microsoft signed a five-year contract worth USD 17.4 billion, rising to USD 19.4 billion with options, served from Vineland. Meta signed a USD 3 billion deal in November 2025 and a further contract of up to USD 27 billion in March 2026. NVIDIA is a technology partner, and site partners include DataOne, Patmos, Verne, Ark Data Centres and AIB Data Centers.

Expansion Plans

Nebius raised its year-end 2026 contracted-power target to 5 GW and plans to add more than 1 GW a year from 2027, backed by 2026 capex of USD 16- 20 billion. Announced projects include a USD 10 billion, 310 MW campus in Lappeenranta, Finland, a second Mäntsälä site of up to 70 MW from 2027, a 240 MW facility in France and 50 MW with AIB Data Centers in the southeastern US.

Lambda

Lambda is an NVIDIA-backed GPU cloud for AI developers, enterprises and frontier labs. It was founded in 2012 in San Francisco by brothers Stephen and Michael Balaban. For a decade it sold GPU workstations and servers; it shut that hardware business in August 2025 and dropped its inference API to focus on GPU cloud and long-term training contracts.

GPU Infrastructure

Lambda’s virtualization software splits clusters of 16,000+ GPUs into rentable slices. It offers NVIDIA H100, H200 and B200, and is deploying tens of thousands of GB300 NVL72 GPUs for Microsoft. NVIDIA leases back about 18,000 GPUs under a USD 1.5B, four-year deal.

Data Center Footprint

Lambda's website lists 15 data centers, all in the US. Its total operational capacity in MW is not disclosed. It states an ambition of 3 GW under management by 2030.

Major Customers and Partnerships

Anthropic is now its largest committed customer: its S-1 shows agreements with Lambda worth about USD 41.3 billion through 2034, including a USD 34.9 billion deal from August 2026. Microsoft signed a multibillion-dollar, multi-year agreement in November 2025. NVIDIA is an investor, supplier and customer, and Hudson River Trading has a multi-year supply agreement.

Expansion Plans

The Anthropic deal centers on a 350 MW data center in Nueces County, Texas, developed by Hut 8, with NVIDIA holding the lease. Lambda's contracted backlog rose to about USD 50 billion by September 2026. It closed a USD 1 billion credit facility in May 2026 and a USD 926 million loan, and an IPO is expected late 2026 or early 2027.

Fluidstack

Fluidstack builds and operates custom AI data centers and GPU/TPU clusters for frontier labs. It was founded in 2017 as a University of Oxford spinout by Gary Wu (CEO) and Jamie Cox. It began as a marketplace for spare GPU capacity and now runs dedicated private clusters. It moved its headquarters from London to New York in December 2025.

GPU Infrastructure

Fluidstack is the first publicly known operator of Google TPU data centers outside Google, building the sites that will house Anthropic's planned 1 million TPUs. It also runs NVIDIA GPU clusters for other customers.

Data Center Footprint

The investor memo puts Fluidstack at up to 1.3 GW across more than ten sites in 2026, all in the US after it withdrew from its French projects in March 2026. Key sites are 378 MW leased at TeraWulf's Lake Mariner campus in New York and a 168 MW joint venture at Abernathy, Texas. It also has about 244 MW of hosting agreements with Cipher Mining in Texas.

Major Customers and Partnerships

Anthropic is the anchor customer, under a USD 50 billion program announced in November 2025 to build custom data centers in Texas and New York. Google backstops the debt on several sites and supplies the TPUs. Facility partners include TeraWulf, Cipher Mining and Hut 8, and earlier customers included Meta and Mistral.

Expansion Plans

The Anthropic sites are due online throughout 2026, with more locations to follow. The Abernathy campus targets delivery in the second half of 2026 under a 25-year lease. The investor memo targets more than 17 GW under management by 2030.

Vultr

Vultr calls itself the world's largest privately held cloud infrastructure company. It offers cloud compute, bare metal, storage and, since 2023, a fast-growing GPU cloud. It was founded in 2014 by David Aninowsky, was self-funded for over a decade, and is led by CEO J.J. Kardwell. It is headquartered in West Palm Beach, Florida.

GPU Infrastructure

Vultr is AMD's flagship GPU-cloud partner, running Instinct MI300X, MI325X and MI355X GPUs. Its two largest clusters are 24,000 MI355X GPUs in Springfield, Ohio, and more than 10,000 MI355X GPUs in Chennai, India. It also offers NVIDIA GPUs, including GB300 systems through HPE, and is an NVIDIA Exemplar Cloud on Blackwell.

Data Center Footprint

Vultr operates 34 cloud data center regions in 17 countries across six continents. Its two AI superclusters draw about 50 MW (Ohio) and 20 MW (Chennai). Total operational MW across all regions is not disclosed. Chennai is its largest deployment outside the US.

Major Customers and Partnerships

Vultr says it serves hundreds of thousands of customers in 185 countries, named AI customers include Chai and LiquidMetal AI. AMD is both an investor and chip partner. In September 2026 Vultr placed a USD 1.2 billion order with HPE for AMD Helios rack-scale systems.

Expansion Plans

Vultr aims to reach 1 GW of capacity by 2029. Chennai is set to expand in 2027 with AMD Instinct MI455X GPUs and Helios racks, and the Ohio cluster is planned to move to Helios. Its new funding round and a possible IPO are intended to finance this build-out.

Northern Data

Northern Data runs one of Europe's largest independent GPU clouds through Taiga Cloud, plus data center design and operation through Ardent Data Centers. It is headquartered in Frankfurt, Germany, and took its current form in 2019 from predecessors Northern Bitcoin and Whinstone. It sold its Peak Mining business in November 2025, and Rumble (now RUM Group) took 85.2% control on 17 June 2026, folding it into a new unit, Quake AI.

GPU Infrastructure

The fleet is about 22,400 NVIDIA GPUs: 20,400 H100s and 2,000 H200s. About 85% were generating revenue in March 2026, up from 62% in December 2025. Taiga Cloud lists H100 from USD 2.60 and H200 from USD 3.00 per GPU-hour.

Data Center Footprint

Northern Data has about 250 MW of energized and planned power across ten data centers, four of them owned, with more than 200 MW not yet earning revenue. It operates in Sweden, Norway, Portugal, the Netherlands, the UK and the US, with corporate offices in Germany.

Major Customers and Partnerships

Tether is a long-standing investor and GPU financier and is linked to Rumble. NVIDIA is its chip supplier, and it earlier partnered with Core42 for access to up to 10,000 GPUs. Under Rumble, the combined Quake AI unit signed a USD 270 million multi-year GPU cloud contract with Together AI on NVIDIA B300 systems.

Expansion Plans

Its largest project is the Maysville, Georgia site, planned for up to 180 MW. Ardent aims to have about 250 MW deployed or coming online across ten sites by 2027. Rumble has signalled a squeeze-out of the remaining minority shareholders.

WhiteFiber

WhiteFiber combines a GPU cloud with AI colocation in data centers it owns or retrofits. It is headquartered in New York and was formed in 2024 as a spin-out of Bitcoin miner Bit Digital, which had launched the cloud business in January 2024. It listed on Nasdaq (WYFI) in August 2025. CEO Sam Tabar previously ran Bit Digital.

GPU Infrastructure

WhiteFiber has about 5,000 NVIDIA GPUs contracted, mainly H200 and H100. In July 2026 it used DriveNets AI Fabric to link two separate data centers into one logical GPU cluster.

Data Center Footprint

About 11 MW was online in 2025, with about 76 MW targeted by end-2026. Sites are MTL-1 (4 MW) and MTL-3 (7 MW) in Montreal, Canada, and NC-1 (99 MW gross) in Madison, North Carolina and MTL-2 is on hold. It also has a presence in Iceland.

Major Customers and Partnerships

Nscale is the anchor colocation customer under a 10-year, 40 MW, ~USD 865 M contract at NC-1. Cerebras uses MTL-3 under a five-year, 5 MW agreement. WhiteFiber reports 20+ customers; power partners include Duke Energy and Hydro-Québec.

Expansion Plans

The 40 MW for Nscale was scheduled to be ready for service by 31 May 2026. Expanding NC-1 from about 54 MW to about 100 MW of gross power depends on a later utility upgrade, expected around late 2027. WhiteFiber reports a pipeline of about 1.5 GW and follows a retrofit-first strategy to cut build time.

Nscale

Nscale is a vertically integrated AI infrastructure company that builds data centers and sells GPU capacity, mostly as dedicated infrastructure for hyperscalers and AI labs. It is headquartered in London and was founded in 2024 by Josh Payne (founder and CEO), spun out of Australian crypto miner Arkon Energy.

GPU Infrastructure

Nscale has about 25,000 active GPUs and about 461,000 active and contracted, so only about 5% were live at 31 August 2026. Deployments include NVIDIA B200 (2,304 at Glomfjord), GB300 Blackwell Ultra and Vera Rubin, and it also offers AMD accelerators.

Data Center Footprint

Nscale runs five active sites, with about 1.37 GW of active and contracted capacity. Sites include Glomfjord (30 MW) and Narvik (230 MW planned, plus 290 MW expansion) in Norway, Sines in Portugal, Ward County in Texas, and the Monarch campus in West Virginia. The Loughton, UK site is on hold because grid power is not expected until the 2030s.

Major Customers and Partnerships

Microsoft has contracts worth up to about USD 43.8B through 2033 across Norway, Portugal and Texas, but reportedly exited Loughton on 18 October 2026. Anthropic signed a USD 45 billion deal for the West Virginia campus. NVIDIA is an investor, and partners include Aker, Singtel, Start Campus and WhiteFiber.

Expansion Plans

Narvik will add more than 30,000 NVIDIA Rubin GPUs for Microsoft in 2027, though the site has run about six months late. Monarch's first phase is designed for 1.35 GW by early 2028. Nscale is buying Anyscale for about USD 1.65 billion and says it is scaling toward 10 GW. Microsoft and Anthropic hold about 85% of its contract value.

How the Neocloud Business Model Works

The neocloud business model is simple to describe and hard to execute: secure GPUs and power, sign long-term customers, and borrow against those contracts to build more capacity.

Neocloud companies earn money in three main ways:

• GPU cloud rental: customers rent GPUs by the hour or reserve them for months, often with software for orchestration, storage and inference on top.

• Dedicated capacity contracts: a hyperscaler or AI lab commits to a cluster for several years, such as Microsoft's USD 9.7 billion contract with IREN or Meta's contract with Nebius worth up to USD 27 billion.

• Colocation and hosting: some, like WhiteFiber and Crusoe, also lease data center space to customers who bring their own hardware.

Signed contracts let neoclouds raise GPU-backed debt, which buys more GPUs and builds more capacity, which earns more revenue. It is capital-intensive: CoreWeave's total debt reached about USD 21.6 billion by the end of 2025, and it plans USD 35-39 billion of capital spending in 2026.

Do neoclouds build their own data centers? 

Some do, and most mix approaches. Crusoe and IREN build their own campuses; Lambda and Fluidstack often use partner sites developed by companies such as Hut 8, TeraWulf and Cipher Mining.

Who are neocloud customers? 

Hyperscalers (Microsoft, Meta, Google, Oracle), AI labs (Anthropic, OpenAI), AI-native startups and, increasingly, trading firms such as Jane Street and Hudson River Trading.

The main risks are customer concentration, delays in power and grid connections, and the speed at which GPUs lose value as new chip generations arrive.

Neocloud Companies by Region

United States

The US is home to most of the world's largest neocloud companies and most of their capacity. CoreWeave, Crusoe, Lambda, Fluidstack, Vultr and WhiteFiber are US-headquartered, and Texas has become the industry's busiest build market, with campuses for Crusoe (Abilene), IREN (Childress and Sweetwater), Lambda (Nueces County), Fluidstack (Abernathy) and Nscale (Ward County).

Europe and the UK

Nebius (Amsterdam), Nscale (London) and Northern Data (Frankfurt) are Europe's leading neocloud companies. The Nordics are a magnet for AI capacity thanks to cheap renewable power: Nebius operates in Finland and Iceland, Nscale in Norway, and Northern Data in Sweden and Norway. CoreWeave also runs sites in the UK and continental Europe, including a new deployment in Copenhagen.

India and Asia-Pacific

IREN is headquartered in Sydney, though its capacity is in North America. India is an emerging neocloud market: CoreWeave plans a 240 MW campus in Navi Mumbai with AdaniConneX, and Vultr has opened an AMD-powered supercluster in Chennai. CoreWeave has also announced entry into Indonesia.

Other Neocloud Providers to Watch

Several fast-growing neocloud providers are not ranked here because they publish only run-rate or ARR figures, not annual revenue:

• Together AI (San Francisco): an AI cloud for open-source models that combines token-based inference with GPU cluster rental. Its annualized revenue was reported at about USD 1 billion in early 2026, and it raised USD 800 million at a USD 8.3 billion valuation in mid-2026.

• RunPod (Mount Laurel, New Jersey): a developer-focused GPU cloud that said it passed USD 325 million in ARR in October 2026, up from USD 120 million in January.

• Lightning AI (merged with Voltage Park in January 2026): combines PyTorch Lightning software with Voltage Park's 35,000+ NVIDIA GPUs; the merged company reported more than USD 500 million in ARR.

Neocloud Industry Trends in 2026

• Contracts are running far ahead of capacity. Backlogs have ballooned, CoreWeave's to about USD 104 billion, while much of the contracted capacity is still being built.

• Frontier labs are now the anchor buyers. Anthropic alone has committed tens of billions of dollars to Lambda, Nscale and Fluidstack in 2026.

• An IPO wave is building. Nscale has filed, Lambda and Crusoe are preparing, and private valuations have climbed sharply.

• Alternatives to NVIDIA are gaining ground. Vultr is built around AMD Instinct GPUs, and Fluidstack is building Google TPU data centers for Anthropic.

• Power is the bottleneck. Grid delays have hit projects on both sides of the Atlantic, from an ERCOT energization pause in Texas to Nscale's Loughton site in the UK.

Conclusion

CoreWeave is the world's largest neocloud company in 2026. Crusoe, IREN, Nebius and Lambda form a second tier of companies with roughly USD 500 million to USD 1 billion in annual revenue, while Fluidstack and Nscale are smaller today but sit on some of the largest contracted pipelines in the industry. The defining question for 2026 and 2027 is whether these companies can turn tens of billions of dollars in signed contracts into powered, revenue-generating capacity on time.

Track neocloud data center projects with Blackridge Research. 

Blackridge's Data Center Projects Database and Global Project Tracker follow AI and neocloud data center developments worldwide, from announcement to construction to operation. Contact Blackridge Research to explore the data.