Neoclouds explained: the GPU cloud providers

What a neocloud is

A neocloud is a cloud company built specifically to rent out AI computing power, mostly Nvidia GPUs (graphics processing units, the chips used to train and run AI models). Unlike Amazon Web Services, Microsoft Azure or Google Cloud, which sell hundreds of services from storage to databases, neoclouds focus on large clusters of accelerators, fast networking between them, and the software to schedule AI workloads.

The category grew out of the GPU shortage that followed the launch of ChatGPT in late 2022. Several of today’s leaders started elsewhere: CoreWeave in cryptocurrency mining, Nebius from the international assets of Russian search company Yandex, Lambda as a seller of AI workstations, and Crusoe by powering computing with otherwise flared natural gas.

How the business model works

The core product is GPU rental, priced per GPU-hour (one chip running for one hour). Small customers pay on demand. The bulk of revenue, however, comes from large customers that reserve whole clusters for several years.

Those reservations are usually take-or-pay contracts: the customer commits to pay for the capacity whether or not it uses it. The contracts are what make the model financeable. A neocloud signs a multi-year commitment, then borrows against that contract and the GPUs it buys to fulfil it, and repays the debt from the customer’s payments.

The economics depend on three numbers: the price per GPU-hour, the share of capacity that is actually paid for (utilization), and how long the chips stay commercially useful. For reference, Silicon Data’s neocloud H100 rental index stood at $2.78 per GPU-hour on October 4, 2026. The H100 is Nvidia’s previous-generation flagship; newer Blackwell chips rent for more.

The main players and their numbers

CoreWeave and Nebius are publicly listed and file results with the SEC. Lambda and Crusoe are private, so their figures come from company announcements and press reports.

Customer concentration

A handful of buyers account for most neocloud demand. CoreWeave disclosed that Microsoft generated about 67% of its 2025 revenue. In its June 2026 quarterly filing, its largest customer accounted for 36% of quarterly revenue and the next two for 26% and 10%, showing diversification but still a narrow base.

The contracts behind that backlog are large and long. CoreWeave’s filings describe OpenAI commitments of up to $11.9 billion through October 2030 and up to a further $6.5 billion through May 2031, and Meta order forms of about $14.2 billion (September 2025) and about $21 billion (March 2026) running to 2032. Nebius signed a Microsoft agreement valued at $17.4 billion, potentially $19.4 billion, in September 2025, and in March 2026 a Meta deal worth up to $27 billion over five years, of which $12 billion is committed capacity. Lambda has a multibillion dollar, multi-year Microsoft contract and a $1.5 billion arrangement under which Nvidia rents back GPU capacity.

The upside is visibility: much of the next few years’ revenue is contracted. The downside is that a dispute, renegotiation or slowdown at one buyer can move results sharply.

How neoclouds are financed

Buying GPUs ahead of revenue requires huge amounts of capital. CoreWeave spent $6.4 billion on capex in Q2 2026 alone and Nebius about $5.7 billion. Much of this is funded with GPU-backed debt, loans secured by the chips and by the customer contracts they serve.

• CoreWeave pioneered the structure with a $2.3 billion loan backed by H100s in August 2023. By June 30, 2026 its gross debt was about $35.6 billion. In March 2026 it closed an $8.5 billion delayed draw facility tied to its Meta contract that Moody’s rated A3, an investment grade rating.
• Nebius closed its first asset-backed facility, $775 million at SOFR (a benchmark short-term rate) plus 2.50%, and has raised equity through an at-the-market share program.
• Lambda announced a $1 billion credit facility in May 2026 and a $926 million term loan at SOFR plus 3% in August 2026, both secured by GPU servers and related cash flows.

How neoclouds differ from hyperscalers

Hyperscalers (Amazon, Microsoft, Google, Oracle and, for its own use, Meta) fund data centers largely from operating cash flow, sell a wide range of services, and design some of their own AI chips. Neoclouds are narrower: they depend on Nvidia supply, borrow heavily, and sell one main product. Their advantages are speed in standing up large GPU clusters and willingness to take on dedicated, single-tenant builds. Hyperscalers are also among their largest customers, renting neocloud capacity to supplement their own buildout.

Risks to understand

• Depreciation. CoreWeave depreciates its GPU equipment over six years. If newer chips make older ones uneconomic sooner, earnings and collateral values would be overstated.
• Utilization and repricing. Capacity not covered by long contracts must be resold at market rates, which can fall quickly as supply grows.
• Leverage and interest costs. CoreWeave’s Q2 2026 interest expense of $640 million was more than double a year earlier, and it reported a $626 million net loss despite a 59% adjusted EBITDA margin.
• Counterparty concentration. Backlog is only as strong as the few buyers behind it, some of which are themselves loss-making AI labs.
• Execution and power. Revenue starts only when sites are energized and delivered on schedule. Delays push out cash flows that debt service depends on.

What to watch next

• Q3 2026 results from CoreWeave and Nebius, expected in November, especially backlog, customer mix and interest costs.
• Whether Lambda or Crusoe file for public listings, which would bring audited disclosure.
• GPU rental indices and the pricing of new GPU-backed debt as signals of supply and lender appetite.

Sources and further reading

• CoreWeave second quarter 2026 results (August 11, 2026)
• Nebius Q2 2026 results, Form 6-K exhibit (SEC) (August 12, 2026)
• DCD: CoreWeave secures $8.5 billion loan for GPU purchases (March 31, 2026)
• The Next Web: Meta commits up to $27 billion to Nebius (March 16, 2026)
• Bloomberg: Crusoe raises over $3 billion at $30 billion valuation (September 3, 2026)
• Silicon Data H100 rental price index (October 4, 2026 reading)