Home / Research / Neocloud Market Statistics

Market research · September 17, 2026

Neocloud Market Statistics: 2026 Report

This report looks at specialist companies whose business is renting accelerated compute as of 2026. Those companies are called neoclouds: they operate between chip vendors and model developers, buying GPUs at scale, wrapping them in power, cooling, and high-speed networking, and selling the result by the GPU-hour or under multi-year contract.

We tracked 23 specialist providers using quarterly filings, shareholder letters, published rate cards, and announced power contracts. Where a company is private, we built figures from disclosed funding rounds, announced customer contracts, and capacity filings with utilities and planning authorities.

Leading Neoclouds by Revenue and Contracted Power

The table ranks the seven largest specialist providers by annualized revenue from third-quarter 2026 results, with the power they have energized and the power they have under contract. Contracted power is the better forward indicator, since a gigawatt under contract converts to revenue only as it comes online.

The Leading Neoclouds by Revenue and Contracted Power, 2026

ProviderAnnualized RevenueYear-Over-Year GrowthActive PowerContracted Power
CoreWeave$12.2B+108%1.6 GW3.9 GW
Nebius$3.3B+298%0.9 GW4.8 GW
Crusoe$2.6B+166%0.25 GW4.6 GW
Lambda$1.4B+98%0.18 GW0.9 GW
Together AI$0.9B+112%0.07 GW0.3 GW
Groq$0.6B+175%0.06 GW0.25 GW
Nscale$0.6B+240%0.12 GW1.3 GW

Insights

Neocloud Quarterly Revenue, Q1 2024 to Q3 2026

The category's growth is easier to read quarter by quarter than in annual totals, because the largest contracts step up as capacity energizes rather than ramping smoothly. In the table below, we separate CoreWeave and Nebius from the rest of the field, then plot all three series.

The Neocloud Quarterly Revenue, 2026

QuarterCoreWeaveNebiusAll Other Tracked Neoclouds
Q1 2024$330M$24M$215M
Q2 2024$470M$31M$262M
Q3 2024$655M$39M$335M
Q4 2024$890M$51M$398M
Q1 2025$1,040M$74M$465M
Q2 2025$1,255M$121M$572M
Q3 2025$1,465M$205M$688M
Q4 2025$1,720M$348M$845M
Q1 2026$2,110M$405M$1,020M
Q2 2026$2,610M$590M$1,235M
Q3 2026$3,050M$815M$1,530M
Line chart of quarterly neocloud revenue from Q1 2024 to Q3 2026 for CoreWeave, Nebius, and all other tracked neoclouds.
The Neocloud Quarterly Revenue, September 2026

Two things stand out in the shape of these curves. The first is that CoreWeave's lead has held in percentage terms while widening in absolute dollars, so the category is neither consolidating toward a single winner nor fragmenting toward many. The second is that the rest of the field, taken together, has grown faster than CoreWeave since the middle of 2025, which is what a maturing supply market looks like: the second and third tier providers are now winning contracts that would previously have gone to whoever had capacity first.

Customer Concentration and Contract Structure

The defining financial risk in this category is that a small number of customers account for most of the revenue, and those customers are also building their own capacity. In the table below, we report concentration and contract structure for the five providers with sufficient disclosure to measure it.

The Customer Concentration and Contract Structure, 2026

ProviderLargest Customer ShareTop Three ShareRevenue Under Committed ContractMedian Contract Length
CoreWeave61%82%95%4.2 years
Nebius44%71%88%3.5 years
Crusoe38%66%91%4.8 years
Lambda29%54%73%2.6 years
Together AI18%39%51%1.8 years

Concentration of this order is normally a warning sign, and here it is partly the opposite. A four-year take-or-pay contract with a creditworthy counterparty is what allows these companies to raise the debt that buys the GPUs in the first place, so the concentration and the financing are the same fact viewed from two sides. The larger exposure arrives when the contract ends: the customer has by then built its own capacity, and the provider is left holding three-year-old accelerators in a market where the replacement generation is twice as efficient. The providers with the shortest contracts and the least concentration are also the ones with the lowest exposure to that scenario, which is a trade most of the category has chosen not to make.

Neocloud Unit Economics and Leverage

In the table below, we compare the accounting assumptions and balance sheets behind the revenue figures above. The depreciation schedule matters more than any other line: every year added to the assumed useful life of a GPU moves cost out of the current period and into a future one.

The Neocloud Unit Economics and Leverage, 2026

ProviderGPU Depreciation LifeAdjusted EBITDA MarginNet MarginDebt Outstanding
CoreWeave6 years58%-24%$23.8B
Nebius4 years33%-11%$4.2B
Crusoe5 years41%Not disclosed$6.1B
Lambda5 years36%-8%$1.4B
Together AI4 years29%-6%$0.4B

Insights

Neocloud Share of the AI Cloud Market Through 2030

In the table below, we set our estimate of total AI cloud spend against the share captured by specialist providers, and against the share of neocloud workloads that are inference rather than training.

The Neocloud Share of the AI Cloud Market, 2026

YearAI Cloud MarketNeocloud RevenueNeocloud ShareInference Share of Workloads
2024$41B$4.6B11%22%
2025$72B$11.8B16%31%
2026$118B$21.6B18%44%
2028 (projected)$196B$43.1B22%63%
2030 (projected)$274B$57.5B21%78%

The share figure peaks before the end of the forecast period, which is the single most interesting number in this report. Specialist providers win share while capacity is scarce and while the buyers are model developers who care about interconnect quality and nothing else. As the workload mix shifts toward inference, the buyer changes: an enterprise serving a product wants the compute next to its data, its identity system, and its compliance posture, which is the hyperscalers' home ground. Our projection assumes the neoclouds answer that by moving up the stack into managed inference rather than by competing on price per GPU-hour, and the providers already doing so are the ones carrying the shortest contracts and the least debt.

Requesting a Copy of This Report

If you would like a PDF copy of this report, or the provider-level dataset behind it, you can reach out here.

Sources