AI Policy Wiki
Dashboard

Circular Financing in AI

high confidence · updated 2026-08-03

2024-2026 structural pattern: Nvidia ↔ OpenAI ↔ Microsoft ↔ CoreWeave ↔ Oracle funding loops; cross-investment and cross-purchase commitments that inflate apparent market size.

Circular financing in AI refers to a 2024–2026 pattern in which AI infrastructure spending and investment flow in closed loops among a small number of firms, such that the same dollars are counted as revenue, investment, or capital expenditure depending on a firm's position in the loop. Observers including Senator Elizabeth Warren and several financial outlets have argued that the pattern amplifies apparent market size while concentrating risk across interlinked firms.

The core loops

Nvidia and OpenAI

On September 22, 2025, OpenAI and Nvidia announced a 10 GW partnership under which Nvidia committed up to $100B in investment in OpenAI. Under the announced structure, Nvidia's investment funds OpenAI's purchase of Nvidia GPUs at scale, so that revenue flows from Nvidia to OpenAI and back to Nvidia. On September 23, 2025, The Information reported on the financing mechanism under the headline "In OpenAI Megadeal, Nvidia Discusses a New Business Model: Chip Leasing." On January 30, 2026, the WSJ and Reuters reported that Nvidia's $100B stake had stalled; Jensen Huang publicly denied being unhappy with the arrangement on January 31, 2026.

Microsoft and OpenAI

Microsoft's stake comprises more than $13B in investment plus IP rights, against which OpenAI committed $250B back to Microsoft Azure. On November 12, 2025, The Information published "Here's How Much OpenAI Spends On Inference and Its Revenue Share With Microsoft." On October 29, 2025, The Register reported that Microsoft earnings suggested more than $11.5B in quarterly OpenAI losses flowing through Microsoft's books.

OpenAI, Oracle, and CoreWeave

OpenAI committed $300B to Oracle Cloud, with Oracle taking on debt to build the capacity, and $11.9B to CoreWeave, which likewise took on debt to build capacity for OpenAI. In a January 28, 2026 letter, Senator Elizabeth Warren wrote that OpenAI had "leverage[d] other people's balance sheets," with more than $100B borrowed by partners to build for OpenAI while keeping OpenAI's own debt load low (Sen. Warren Letter to OpenAI (2026-01-28)).

Amazon, Anthropic, and OpenAI

Amazon committed up to more than $4B to Anthropic, with more planned as of July 2025. On October 31, 2025, Bloomberg reported that Alphabet and Amazon stakes in Anthropic boosted those firms' profit by billions through mark-to-market gains. On January 29, 2026, Amazon was reported to be in talks for a $50B OpenAI investment, which would make Amazon simultaneously the largest external investor in both Anthropic and OpenAI.

Inflated market signals

A single dollar flowing through Microsoft to OpenAI to Nvidia can be counted simultaneously as Microsoft's AI revenue (Azure consumption), OpenAI's investment intake, and Nvidia's product revenue, tripling apparent AI-market size. Warren's January 28, 2026 letter framed the arrangement as "systemic risk" (Sen. Warren Letter to OpenAI (2026-01-28)).

Concentrated risk

The interlinked structure means that the failure of any one node — OpenAI slowing growth, Microsoft reconsidering its commitment, Nvidia's own customers slowing, or Oracle debt becoming unserviceable — transmits a shock to all connected nodes.

The debt-funding leg of the loop drew attention in 2026. On April 24, 2026, the WSJ reported that JPMorgan and other banks had struggled to syndicate billions of dollars in loans tied to Oracle data-center leases in Texas and Wisconsin, described as a signal that the debt-funding leg of the loop was hitting market constraints (Source: wsj.com). Two days earlier, on April 22, 2026, The Information reported that OpenAI and Oracle had formed DeployCo, a $1.5B joint venture for AI data-center deployment that formalized the prior Oracle-OpenAI compute relationship into a co-owned vehicle (Source: theinformation.com).

Coverage and analysis

Financial outlets covered the pattern through late 2025 and into 2026. On October 22, 2025, the WSJ ran "Is the Flurry of Circular AI Deals a Win-Win — or Sign of a Bubble?" On November 10, 2025, the NYT published "Why Debt Funding Is Ratcheting Up the Risks of the A.I. Boom." On November 18, 2025, TheStreet ran "Nvidia, Microsoft deal takes 'circular' financing to entirely new level."

In July 2026 Nvidia formalized a revenue-share backstop program: it agreed to rent back neocloud customers' unused GPUs at a fixed rate — using its balance sheet to help customers finance purchases of its own chips — in exchange for a share of those customers' cloud revenue, with Firmus and Sharon AI the first adopters after 2025 backstop deals with CoreWeave ($6.3B) and Lambda ($1.5B) (Sources: theinformation.com; datacenterdynamics.com).

On May 9, 2026, CNBC reported that Nvidia had committed more than $40B to AI-infrastructure equity in 2026, including $30B to OpenAI, $3.2B to Corning, and $2.1B to IREN, drawing analyst critique over the circular vendor-financing pattern. The 2026 equity-bet total recast the prior $100B OpenAI commitment as one component of a broader Nvidia-as-AI-investor posture (Source: cnbc.com).

The pattern was also reported outside the United States. On April 22, 2026, Bloomberg reported that Tencent and Alibaba together committed more than $20B to DeepSeek-centric infrastructure builds, a Chinese parallel to the US circular pattern (Source: bloomberg.com).

Matt Stoller argued on August 2, 2026 that the financial reporting of the largest AI spenders cannot be checked from public filings, pointing to cross-subsidization at Google, Amazon, Microsoft and Nvidia and to the use of shell companies and off-balance-sheet instruments. He proposed repealing parts of the JOBS Act of 2012 to restore the earlier SEC registration thresholds of 500 shareholders and $10 million in assets, stricter SEC enforcement of disclosure on material lines of business, revenues and risks, and tighter Public Company Accounting Oversight Board inspection of the hyperscalers' auditors, framing the approach as what New Deal regulators did with 1920s financiers (Source: thebignewsletter.com). The argument is a disclosure-regime critique rather than a measurement of the loops described above, and is offered as an attributed position. See AI Bubble Debate.

A public-market valuation reference point for non-Nvidia compute emerged in May 2026. On May 11, 2026, Cerebras Systems upsized its IPO to 30M shares at $150–$160, targeting up to a $4.8B raise at a $34.4B valuation, with trading expected May 14. The offering represented a partial public-market exit for a non-Nvidia compute provider and a non-Nvidia frontier-compute IPO valuation reference point (Source: bloomberg.com).

Relationships