Published:
Nvidia has invested $2 billion each in CoreWeave and Nebius, two “neocloud” companies whose primary business is renting out Nvidia GPUs, creating a circular financing loop that analyst Beth Kindig of I/O Fund documented in an analysis published in July 2026.
Nvidia supplies the GPUs, holds equity in the buyers, and receives a share of the cloud revenue those GPUs generate. CoreWeave’s IPO registration statement (Form S-1) names Nvidia as both its primary hardware supplier and an equity stakeholder. Some neoclouds pay Nvidia a percentage of cloud revenue as the return on that financing, according to reporting by Spheron Network. Neither CoreWeave nor Nebius is profitable, and both carry soaring debt loads, according to the I/O Fund analysis. Kindig’s finding, in one sentence: “The same money flows from NVIDIA to the neocloud and partially back to NVIDIA.”
What Is a Neocloud and Why Nvidia Funds Them
A neocloud is a GPU-specialist cloud provider that sits between the hyperscalers — Amazon Web Services, Microsoft Azure, and Google Cloud — and the AI companies that need training and inference capacity. CoreWeave and Nebius both buy Nvidia hardware and rent it out at a margin.
CoreWeave went public in early 2025 and supplies AI training capacity to customers including Microsoft and Meta. Nebius is a newer entrant in the same market. Nvidia’s $2 billion positions in each company give the chipmaker an equity claim on the profits its own hardware generates downstream.
How the Nvidia, CoreWeave, and Nebius Financing Loop Works
The loop runs in 5 steps: Nvidia sells GPUs to the neocloud, Nvidia takes an equity stake in that neocloud, the neocloud rents the GPUs to AI companies, the AI companies pay for compute, and the neocloud returns a share of that revenue to Nvidia. Each turn of the loop funds the next GPU order.
The structural concern analysts raise is demand visibility. Nvidia’s investment income depends on neocloud success, and neocloud success depends on Nvidia continuing to sell them GPUs — so a portion of Nvidia’s reported GPU demand originates from companies Nvidia itself capitalized. The arrangement is disclosed and legal; the I/O Fund analysis characterizes it as unprecedented in scale rather than improper.
Why CoreWeave’s S-1 Disclosure Matters
CoreWeave’s S-1 filing states Nvidia’s dual role — supplier and shareholder — in the same document that asks public investors to price the company. Financial analysts, including Kindig, flag that dual role as a conflict of interest without a close precedent in public markets.
The disclosure matters because the risk is concentrated, not distributed. A single supplier sets the input price, holds equity in the buyer, and collects a revenue share on the output. Published revenue-share percentages do not exist, which limits how precisely outside investors can model the arrangement.
What Circular GPU Financing Means for Businesses Buying AI Tools
Compute economics set the floor under AI software pricing. Every business paying for AI tools — such as ChatGPT Business, Claude Team, and Gemini for Workspace — pays a price built on top of GPU rental costs at companies like CoreWeave and Nebius. Our comparison of AI subscription pricing across ChatGPT, Claude, and Google tracks where those per-seat costs currently land.
The pressure point is substitution. Large buyers are building their own silicon: Apple committed $30 billion to Broadcom for US-made chips, and DeepSeek is developing an in-house inference chip. AMD’s MI355X already runs top models at roughly half the inference cost of Nvidia Blackwell. Each of those shifts removes GPU demand from the loop.
Our Take
Nvidia has built the most elegant business model in tech: sell the shovels, finance the miners, then take a cut of every gram of gold they find. The question is what happens when the miners start smelting their own shovels. For business software buyers, the practical read is simpler — today’s AI tool pricing rests on a compute market whose demand signal is partly self-generated, so treat multi-year AI budget projections as provisional.
For Context: WithO2’s AI Compute Financing Coverage
- Anthropic’s $35B Compute Deal: How Apollo and Blackstone Are Financing Claude’s Future — the same debt-financed compute pattern, applied to a model developer rather than a GPU renter
- Anthropic Is Paying SpaceX $1.25B/Month for 220,000 GPUs — what GPU capacity costs an AI company that rents rather than owns
- NVIDIA Nemotron 3 Ultra Is Now Live — First Look at Benchmarks — Nvidia’s own model work, the software layer above the hardware business
Sources: Beth Kindig, I/O Fund — Nvidia, CoreWeave, Nebius circular financing analysis · Spheron Network — Nvidia neocloud backstop financing
