OpenAI advances Ohio data center project with reported Nvidia backing as part of major infrastructure buildout.

At its simplest, circular financing describes a deal structure in which a chipmaker or hyperscaler takes an equity stake in, or extends credit to, an AI lab or neocloud provider, and that same company then commits to multi-year purchases of chips or computing power from the firm that just funded it. The money moves in a loop: invest, then buy back, then invest again.

Nvidia is the company most associated with the pattern. It has backed neocloud providers including CoreWeave, Nebius and Nscale through funding rounds and supply commitments, while those same providers use the capital, and often debt raised against Nvidia chips as collateral, to buy more Nvidia hardware. The arrangement has been described as turning a single dollar of Nvidia investment into several dollars of Nvidia purchases.

The scale of these deals has grown sharply over the past year. OpenAI alone has committed to purchasing $250 billion worth of cloud services from Microsoft, deploying tens of billions of dollars' worth of chips from AMD while becoming one of AMD's largest shareholders, and receiving up to $100 billion in investment from Nvidia to help stock its data centres with Nvidia hardware. Anthropic has similarly seen commitments from both Nvidia and Microsoft.

The pattern extends beyond direct chip purchases into the data centre layer itself. Construction is increasingly outsourced to third parties that lease the completed facilities back to hyperscalers on long-term contracts, often with exit clauses built in. The Bank for International Settlements used its 2026 Annual Report to name this dynamic, alongside a potential AI capex bust and sovereign debt fragility, as one of the three biggest risks to global financial stability—a notable escalation from what had largely been an analyst-community concern to a central-bank one.

Neoclouds sit at the centre of the mechanism. The term refers to a new class of cloud providers that specialise almost exclusively in renting out GPU compute for AI workloads, a category that barely existed before late 2024 and has since become one of the fastest-growing parts of the infrastructure market. Somewhat counterintuitively, the hyperscalers are among the neoclouds' biggest customers. Microsoft has struck commitments worth around $60 billion with CoreWeave, Nebius and Nscale, in part because such contracts are booked as operating expenses over their lifetime rather than as capex on Microsoft's own balance sheet. Meta has signed comparable deals, including one worth $35.2 billion with CoreWeave and up to $27 billion with Nebius.

This has direct consequences for how neoclouds price their services. Providers carrying vendor-financed debt from GPU purchases face a different cost structure to those that do not, and that difference filters through into what customers actually pay, including capacity rationing and pricing changes when utilisation runs below the level needed to service that debt.

Power, not chips, is increasingly the constraint that determines who can actually deliver on these financing commitments. CoreWeave is targeting 1.7 gigawatts of active power by the end of 2026, while Nebius is aiming for up to 1 gigawatt of connected capacity against more than 4 gigawatts of already contracted power—a gap between what has been committed and what is actually online that matters enormously to any operator negotiating colocation, interconnection or power-sharing arrangements with a neocloud tenant.

That gap is now shaping how capital treats the sector. Investors underwriting data centre projects increasingly favour traditional hyperscale tenants over neocloud, GPU-as-a-service tenants precisely because hyperscale credit is easier to underwrite, and are hedging neocloud tenant risk through parent guarantees or hyperscaler credit "wrappers" rather than relying solely on the tenant's own balance sheet. For operators, that means the identity and financing structure of a prospective tenant is no longer a side consideration. It is now a core part of the credit conversation before a lease is signed.

There is also a broader market-confidence angle. Meta's $12 billion El Paso bond exposed AI debt repricing pressure; circular financing concerns are a significant part of why capital markets are re-examining the credit quality of AI infrastructure debt more broadly, rather than treating it as a uniformly safe hyperscaler-backed asset class.

Not everyone treats circularity as a red flag. Supporters of these deal structures argue the criticism misses a basic point: building AI infrastructure is extraordinarily expensive and the most advanced chips remain hard to secure, so pairing long-term buying commitments with financing is simply how supply gets locked in during a capacity-constrained market.

The more measured reading is probably that circular financing is not inherently unsound, but it does concentrate risk in ways that are harder to see using traditional balance-sheet analysis. A neocloud's revenue, its debt, and its supplier's investment can all trace back to the same handful of counterparties, which makes conventional credit assessment less reliable than it looks.