AI Circular Financing: Inside the $800 Billion Bubble Risk Reshaping Infrastructure Deals

Direct answer: AI "circular financing" describes arrangements — now estimated at more than $800 billion across the industry — where chipmakers and cloud providers invest in AI companies that then spend much of that money back on the investors' own chips and cloud capacity, including Nvidia's $100 billion OpenAI partnership, AMD's roughly $200 billion in deals involving 10% equity warrants, and Oracle's $300 billion cloud infrastructure commitment. The Bank for International Settlements named this dynamic one of the three biggest risks to global financial stability in its 2026 Annual Report, and commentators including CNBC's Jim Cramer have compared the pattern to dot-com-era financing. It matters now because these deals make AI infrastructure demand look larger and more self-sustaining than the underlying, external revenue may actually support.

How the Loop Actually Works

Strip away the scale and the AI framing for a moment, and the basic mechanic behind circular financing is simple enough to describe in one sentence: a supplier invests in, or extends credit to, one of its own customers, and that customer uses the money to buy more from the supplier. Money that started on the supplier's balance sheet ends up back on the supplier's income statement, having taken a detour through a customer's books along the way.

In the current AI infrastructure buildout, that loop shows up in a few recurring shapes. A chipmaker takes an equity stake in an AI lab, or extends it favorable financing terms, and the lab uses that capital — alongside its own fundraising — to buy the chipmaker's GPUs and the cloud capacity built around them. A cloud provider commits to buying massive amounts of compute capacity from a hardware vendor, while that same hardware vendor commits to buying cloud capacity, or backstopping financing, from the cloud provider. A GPU maker grants equity warrants to a customer as part of a supply agreement, which gives the chipmaker upside if the customer's business succeeds — success that is measured, in no small part, by how much hardware that customer continues to buy from the chipmaker itself.

None of these structures are illegal, undisclosed, or even particularly novel in isolation — vendor financing has existed in one form or another across capital-intensive industries for decades. What's drawing scrutiny in 2026 is the scale, the concentration, and the interconnectedness: a relatively small number of very large companies are transacting with each other in overlapping directions, at a size — collectively estimated at more than $800 billion — large enough to materially shape how healthy the broader AI infrastructure market appears, independent of how much real, external, non-circular demand actually exists underneath it.

The Deals Behind the $800 Billion Number

It's worth naming the specific arrangements driving this figure, because the abstract description of "circular financing" undersells just how large and how directly interconnected these individual deals are.

Nvidia and OpenAI. Nvidia's reported $100 billion partnership with OpenAI sits at the center of this story. In broad structure, financing and investment commitments from Nvidia support OpenAI's ability to build out compute infrastructure, and that infrastructure buildout in turn represents enormous future demand for Nvidia's own chips. Beyond that headline figure, a further arrangement has been discussed involving a roughly $250 billion Nvidia backstop tied to a planned 10-gigawatt AI data center campus in Ohio for OpenAI — a single project, tied to a single chip supplier's financial backing, at a scale that rivals the annual infrastructure budget of entire industries.

AMD's equity-warrant deals. AMD has structured roughly $200 billion in customer deals that involve handing over equity warrants worth approximately 10% of the company to major customers. In practice, this means AMD is giving large AI customers a stake in AMD's own future upside as part of the arrangement that secures their chip purchases — a structure that tightly links AMD's future profitability to the very customers whose purchasing volumes are propping up AMD's near-term revenue growth.

Oracle's cloud infrastructure commitment. Oracle has committed $300 billion to cloud infrastructure buildout, a figure large enough to represent a fundamental bet on sustained AI compute demand continuing to materialize at the scale currently being projected across the industry — and Oracle's customer base for that capacity overlaps significantly with the same small set of AI labs and hyperscalers involved in the other deals described here.

The broader web. Beyond these headline arrangements, the more than $800 billion total reflects a wider pattern: Bloomberg's reporting on how Microsoft, OpenAI, and Nvidia "keep paying each other" describes a genuinely circular set of cash flows, where money moving from any one of these companies to another has a real chance of eventually flowing back to where it started, or to one of a small number of other firms in the same interconnected group. That's the structural detail that has moved this from an interesting financing footnote to a subject of Bank for International Settlements-level scrutiny: this isn't one unusual deal, it's a pattern spanning a "handful of interconnected firms," in the words used to describe it, transacting at a scale that now measures in the hundreds of billions of dollars.

Why Regulators and Market Watchers Are Worried Now

The Bank for International Settlements — the Basel-based institution that functions as a bank for central banks and tracks systemic risk to the global financial system — named AI circular financing, alongside a possible AI capital-expenditure bust, as one of the three biggest risks to global financial stability in its 2026 Annual Report. That's a notably specific and unusual thing for an institution of the BIS's stature and global remit to call out by name. Central banking institutions tend to speak in careful, hedged language about emerging risks; naming a specific financing pattern in a flagship annual report signals a level of concern that goes well beyond routine market commentary.

The concern isn't that any single deal described above is fraudulent or improperly disclosed — these are, by and large, publicly reported, negotiated commercial arrangements between sophisticated counterparties. The concern is what circular cash flows do to the legibility of the market from the outside: when a meaningful share of reported AI infrastructure revenue and demand is generated by money moving between a small number of interconnected companies, it becomes much harder for outside investors, analysts, and even the companies' own boards to distinguish "real," externally sourced demand from revenue that is, in effect, the same dollars completing another lap around the same small circuit. Circular cash flows can make demand look organic and revenue look robust precisely when a meaningful share of it is the same money going around the same circle.

This is also the exact comparison CNBC's Jim Cramer drew in July 2026 when he warned that AI's circular financing frenzy echoes the dot-com bubble. The dot-com era had its own version of this pattern: telecom and infrastructure companies extended vendor financing to customers who used it to buy telecom equipment and services, inflating reported demand for a buildout that, in several well-documented cases, outran the real, sustainable end-customer demand underneath it. When that gap became visible, the unwind was fast and severe. The comparison Cramer and others are drawing isn't that AI infrastructure spending is worthless — it's that the financing structure has a specific historical precedent for how it can mask an emerging supply-demand mismatch until the mismatch becomes undeniable, at which point repricing tends to happen abruptly rather than gradually.

The Historical Echo Behind the Dot-Com Comparison

Cramer's comparison lands because the dot-com era's vendor-financing pattern is genuinely well documented, and it's worth walking through in a bit more detail than a one-line comparison usually gets, because the mechanics — not just the headline analogy — are what make it relevant here. In the late 1990s, a wave of telecom and networking equipment vendors extended financing, credit lines, and in some cases direct equity investment to newer telecom carriers and internet infrastructure companies, on the understanding — implicit or explicit — that much of that capital would flow back to the vendor as equipment purchases. For a period, this created a genuinely self-reinforcing growth story: vendor revenue grew, which supported vendor stock prices, which made it easier to raise the capital needed to extend even more financing to even more customers, who bought even more equipment.

The arrangement worked, in the sense of generating reported growth, for as long as new capital kept entering the system faster than the underlying end-customer demand needed to be validated. What eventually broke it wasn't a single event so much as a slow accumulation of evidence that real consumer and business demand for bandwidth, while genuinely growing, wasn't growing anywhere near fast enough to justify the pace of network buildout being financed. When that gap became broadly visible to investors and lenders, financing dried up quickly, several heavily-financed carriers couldn't refinance their obligations, and the vendors who had extended that financing took substantial write-downs on both the loans and the equipment revenue that depended on them being repaid. The unwind, once it started, moved fast — considerably faster than the multi-year buildout that preceded it — because credit and confidence, unlike physical infrastructure, can be withdrawn far more quickly than they were extended.

The relevance to today's AI infrastructure financing isn't that history has to repeat exactly — the AI industry has real, rapidly growing paid usage in a way the dot-com telecom buildout, at a comparable stage, largely didn't yet have. It's that the specific financial mechanism — vendor-extended capital that flows back to the vendor as revenue, at a scale that makes reported growth partly self-referential — has a well-documented failure mode under one specific condition: a sustained gap between financed buildout and real, externally validated end-demand. Whether that gap exists today, and how large it is, remains the genuinely open, contested question at the center of this entire debate — which is exactly why serious institutions like the BIS are watching it as a live risk rather than dismissing the comparison outright or treating it as settled fact in either direction.

Circular Financing vs. Ordinary Vendor Financing: What's Actually Different

It would be an oversimplification to treat every form of vendor-adjacent financing as inherently a bubble signal — plenty of legitimate, healthy industries have used supplier financing, equity partnerships, and long-term purchase commitments to build real, durable infrastructure. Aircraft manufacturers have long offered financing support to airlines buying their planes; telecom equipment vendors have extended credit to carriers for decades without every instance being a systemic risk. The distinction worth drawing carefully is between financing that accelerates genuinely independent, externally validated demand, and financing that is substantially propping up demand that might not otherwise exist at the scale being reported.

A few practical signals tend to separate the two, and they're the same signals sophisticated analysts are watching for in the current AI buildout. Does the customer receiving financing have a credible, growing base of paying end-users generating revenue from outside the financing relationship itself? Is the customer's spending diversified across multiple suppliers, or concentrated almost entirely with the same firm providing its financing? Would the underlying business plausibly survive, at a smaller scale, if the specific financing arrangement disappeared tomorrow? Deals that pass those tests look more like ordinary vendor financing supporting real growth. Deals that don't — where the financing relationship and the purchasing relationship are so tightly intertwined that it's hard to imagine one continuing without the other — are the ones drawing the BIS-level scrutiny described above.

Applied to the current AI landscape, this is genuinely uncertain territory rather than a settled verdict. OpenAI, for instance, has a large and rapidly growing base of paying consumer and enterprise subscribers generating real, external revenue — that's a meaningfully different position than a company whose only revenue comes from recycled financing-partner dollars. At the same time, OpenAI is reportedly projected to lose approximately $14 billion in 2026 while pursuing a $100 billion revenue target by 2029, which is a large gap between current reality and the scale of spending commitments being made against a multi-year projection. Reasonable, well-informed observers currently disagree about whether that gap represents normal, aggressive growth-stage investment or a sign that some of the underlying demand is more fragile than the headline infrastructure commitments suggest — and that disagreement is precisely why this has become a live, contested question rather than one with an obvious answer either way.

Reading These Deals Skeptically: A Practical Framework

Most people encountering this story secondhand — through a headline number rather than the underlying filings and reporting — don't have an easy way to tell a healthy infrastructure commitment from a fragile one. A few practical questions help cut through the headline figure to what's actually being committed, by whom, and under what conditions.

Is the money a firm commitment or a contingent one? Many of the largest figures reported in this space, including elements of the arrangements described above, combine firm spending commitments with options, warrants, and contingent backstops that only convert into real cash flow under specific conditions. A $100 billion "partnership" headline can bundle equity investment, chip supply agreements, and infrastructure financing support into one number that reads as a single, unconditional commitment when the reality is considerably more conditional and staged over years.

Whose balance sheet actually carries the risk first? In a circular arrangement, it's worth asking who absorbs the loss first if projected demand doesn't materialize — the chipmaker that extended financing, the cloud provider that built the capacity, or the AI lab that signed the purchase commitment. The answer isn't always obvious from a press release, and it matters enormously for understanding where the real exposure sits versus where it's merely reported.

Does the deal's size make sense relative to the counterparty's independent revenue? A useful sanity check is comparing the size of a financing or purchase commitment to the customer's actual, externally generated revenue, rather than to its total funding raised or its projected future revenue. When a commitment dwarfs current independent revenue by an order of magnitude or more, that's not automatically a red flag — early-stage, high-growth companies often look this way — but it does mean the commitment is a bet on future growth materializing, not a reflection of demand that already exists today.

Is the same capital being counted more than once across public announcements? Because these companies transact with each other in overlapping directions, there's a real risk that the same underlying dollars get referenced across multiple separate headline figures — a chip supply commitment, a cloud infrastructure deal, and an equity investment might all be partially describing the same flow of capital moving through the same small set of companies, inflating the apparent total scale of "AI infrastructure investment" when added together naively.

Applying this kind of framework doesn't require access to nonpublic information or specialized financial training — it mostly requires resisting the pull of a large, round headline number and asking a few grounded follow-up questions before treating it as evidence of anything in particular. That habit is useful for investors evaluating exposure, for journalists covering the space, and just as much for an ordinary business trying to decide how much weight to put on a vendor's own infrastructure announcements when choosing a technology partner.

Who's Exposed If the Loop Breaks

The most direct exposure sits with the companies named throughout this piece — Nvidia, AMD, Oracle, OpenAI, and the broader set of hyperscalers and AI labs party to these arrangements — but the risk doesn't stop there, and this is exactly the systemic-risk framing that pushed the BIS to flag it at the level of global financial stability rather than treating it as a company-specific concern.

Public market investors are exposed through equity valuations that, in several of these companies, now embed assumptions about sustained, multi-year AI infrastructure spending continuing at or near current levels. If circular financing has been inflating the appearance of demand, a repricing event wouldn't be contained to one company's stock — it would likely affect the entire cohort of companies whose valuations are benchmarked against the same growth narrative.

Debt and structured-finance markets carry a specific and growing form of exposure. Bank of America has projected that the data center asset-backed securities (ABS) and commercial mortgage-backed securities (CMBS) financing market could top $35 billion in new issuance and refinancing in 2026 — meaning a meaningful and growing share of AI data center buildout is now financed through structured debt instruments that get distributed across a wide range of institutional investors, pension funds, and credit markets, not held solely on the balance sheets of the companies involved. That distribution is exactly the mechanism that turned localized problems into systemic ones in prior financial crises: risk that looks contained when concentrated on a few balance sheets becomes a much broader problem once it's sliced, packaged, and sold into markets with far more numerous and less informed counterparties.

Employees, suppliers, and adjacent industries — data center construction firms, power infrastructure providers, specialty real estate developers, and the broader ecosystem that has grown up around the AI infrastructure buildout — carry indirect but real exposure. A slowdown or repricing in AI capital expenditure would ripple through all of these adjacent businesses, most of which have made their own capacity and hiring decisions based on the current trajectory of AI infrastructure spending continuing roughly as projected.

Smaller AI vendors and startups building on top of the major platforms carry a subtler exposure worth naming separately. A company building its product on a specific cloud provider's AI infrastructure, or pricing its own roadmap around a specific chip generation's expected availability and cost, is implicitly betting on the continued health of the exact commitments described throughout this piece. That's not a reason to avoid building on major AI platforms — most viable alternatives ultimately depend on similar infrastructure somewhere upstream — but it is a reason to build with enough architectural flexibility that a pricing shock or availability change at one layer of the stack doesn't require a full rebuild, a consideration that belongs in technical architecture decisions and vendor contracts alike, not just in a CFO's risk register.

Law firms are already positioning around this risk. Quinn Emanuel has published a client alert specifically on emerging litigation risks in financing the AI data center boom — a signal that sophisticated legal counsel is actively preparing for disputes that could arise if financing structures, disclosures, or counterparty commitments come under stress. That kind of pre-positioning by major litigation firms doesn't happen around risks considered purely theoretical.

The Global Picture

United States. This is unambiguously the epicenter of the phenomenon. Every headline arrangement described in this piece — the Nvidia-OpenAI $100 billion partnership, AMD's roughly $200 billion in warrant-linked deals, Oracle's $300 billion cloud commitment, and the discussed $250 billion Nvidia backstop tied to the 10-gigawatt Ohio data center campus — is a US-based arrangement between US-headquartered companies. Bank of America's projection of a data center ABS/CMBS market potentially topping $35 billion in 2026 is likewise a US structured-finance story.

United Kingdom. No distinct UK-specific reporting on circular financing arrangements was found in this research. UK-based investors and pension funds may well hold indirect exposure through global equity and credit markets, but there's no UK-specific circular financing deal or regulatory statement to report honestly here.

UAE / Dubai. No source found in this research directly ties UAE entities into the specific circular financing arrangements described above. It's worth noting, without overstating the connection, that UAE-linked investment vehicles such as MGX are active co-investors in other large-scale AI infrastructure joint ventures reported elsewhere — but that's a separate, adjacent story about sovereign investment in AI infrastructure generally, not evidence of UAE involvement in this specific circular financing pattern, and it would be inaccurate to blur the two.

Australia. No distinct regional-specific reporting was found. As with the UK, Australian capital markets carry indirect global exposure through investment portfolios, but no Australia-specific circular financing arrangement or regulatory response was identified.

Germany. No distinct regional-specific reporting was found. Germany's substantial industrial and financial sectors are plausibly exposed to any broader AI-related market correction through general portfolio exposure, but that's a reasonable general inference rather than a Germany-specific sourced finding.

Europe / France. No France-specific or broader Europe-specific reporting on this particular financing pattern was found, with one important nuance: the Bank for International Settlements itself is Basel, Switzerland-based, and its 2026 warning was framed as a global financial stability concern rather than a Europe-specific one. It's included here as an international institutional voice weighing in on a largely US-centered phenomenon, not as evidence of a distinct European circular financing story.

China. No distinct regional-specific reporting was found. China's own AI infrastructure buildout and financing patterns are a significant and separate story, but this research pass didn't surface evidence connecting Chinese entities to the specific US-centered circular financing arrangements detailed here.

The pattern across every region outside the US is consistent: this is currently reported and analyzed as an overwhelmingly American phenomenon, with a single global institutional voice — the BIS — flagging it as a worldwide financial stability risk precisely because interconnected US company balance sheets, and the global capital markets that finance and invest in them, don't respect national borders when something goes wrong.

What This Means for Businesses Evaluating AI Infrastructure Bets

None of this means AI infrastructure spending is fake, wasteful, or destined to collapse — real compute demand exists, and some meaningful share of current spending is unambiguously funding infrastructure that will be used productively for years. What it does mean is that any business making decisions based on the appearance of AI infrastructure market health — whether that's a vendor selecting a cloud or compute partner, an investor evaluating exposure, or a company deciding how aggressively to build its own AI roadmap based on "everyone else is spending at this pace" — should look past the headline commitment figures and ask a more specific question: how much of this specific counterparty's demand is externally generated versus financing-partner-generated, and what happens to my plan if that changes.

For businesses actually building AI-powered products and features, rather than participating directly in these large infrastructure deals, the practical implication is more mundane but still important: don't let macro-level infrastructure hype substitute for your own grounded, realistic technical scoping. A vendor's or platform's enormous headline infrastructure commitment doesn't guarantee stable pricing, availability, or company survival at the specific tier or service you depend on. Our AI agents and automation team scopes AI projects around realistic, defensible assumptions rather than the momentum of industry headlines, and our case studies page shows delivered, real-world outcomes rather than projected ones — a useful comparison point when evaluating any vendor's pitch in a market this exuberant.

It's also worth building a basic habit of counterparty awareness into any significant AI vendor relationship: understanding who owns equity or financing stakes in a platform or infrastructure provider you depend on, and what that provider's own balance sheet exposure looks like, is no longer an exotic due-diligence question reserved for institutional investors. In a market where the Bank for International Settlements is naming financing structure itself as a top-three global risk, a basic version of that question belongs in ordinary vendor evaluation too.

This doesn't need to slow down a genuinely good AI initiative — it mostly means separating two decisions that often get bundled together by default: whether to adopt a given AI capability at all, and which specific vendor or infrastructure commitment to build that capability on top of. The first decision can move quickly when the business case is real. The second deserves the same grounded scrutiny you'd apply to any other multi-year technology commitment, regardless of how confidently a vendor's headline funding numbers are being reported in the press that week.

What Businesses Want to Know About AI Circular Financing

What is AI circular financing?

AI circular financing describes arrangements in which a chipmaker, cloud provider, or other AI infrastructure supplier invests in, extends credit to, or otherwise financially backs an AI company that then spends a substantial share of that money back on the investor's own products or services — chips, cloud capacity, or data center infrastructure. The money effectively moves in a loop: it starts on the supplier's balance sheet, flows to the customer, and returns to the supplier as revenue. Individually, these arrangements often resemble ordinary vendor financing, something capital-intensive industries have used for decades. What's drawn scrutiny in 2026 is the scale — over $800 billion in identified arrangements — and the concentration of these deals among a small number of deeply interconnected companies, which makes it harder for outside observers to tell how much underlying AI infrastructure demand is genuinely external versus a byproduct of the financing structure itself.

How does the Nvidia-OpenAI-Oracle financing loop actually work?

In broad structure, Nvidia's reported $100 billion partnership with OpenAI provides financing and investment support that helps OpenAI build out compute infrastructure — infrastructure that, in turn, represents substantial future chip demand for Nvidia. Oracle's $300 billion cloud infrastructure commitment intersects with this same ecosystem, since Oracle's cloud buildout serves overlapping AI lab customers, including arrangements connected to OpenAI's infrastructure needs. Bloomberg's reporting on how Microsoft, OpenAI, and Nvidia "keep paying each other" describes this as a genuinely circular set of cash flows rather than a simple linear supplier-customer relationship — money moving from any one of these companies has a real chance of eventually flowing back to one of the others in the same small, interconnected group, which is the structural feature at the heart of the concern.

Is AI circular financing a sign of a bubble?

It's a contested question rather than a settled one, and reasonable, informed people currently disagree. The case for concern rests on the historical precedent of the dot-com era, when vendor financing helped inflate reported telecom infrastructure demand beyond what real end-customer usage could sustain, followed by a fast, severe unwind once the gap became visible. The case against treating it as automatically a bubble rests on the fact that some AI infrastructure demand is genuinely being used productively, and that vendor financing has supported real, durable industries in the past without triggering a crisis. The honest answer is that circular financing is a risk factor that makes an eventual bubble more likely and potentially more severe if broader AI demand disappoints — not, on its own, definitive proof that a bubble currently exists.

Why do analysts worry about circular financing deals in AI infrastructure?

The core worry is that circular cash flows can make demand look organic and revenue look robust even when a meaningful share of it is the same dollars completing another lap around a small circuit of interconnected companies. That distorts the signals investors, analysts, and even company boards typically rely on to judge whether AI infrastructure spending is sustainable — reported revenue growth, contracted demand, and capital expenditure commitments all look stronger when circular flows are included than an "external demand only" view might show. This concern is significant enough that the Bank for International Settlements named circular financing, alongside a possible AI capital-expenditure bust, as one of the three biggest risks to global financial stability in its 2026 Annual Report — a strong signal from an institution not prone to overstating emerging risks.

What financial signals would indicate that AI circular financing arrangements are actually sustainable?

A few practical signals separate healthier arrangements from more fragile ones: a growing base of paying end-users generating revenue from genuinely outside the financing relationship; purchasing and revenue that's diversified across multiple counterparties rather than concentrated with the same firm providing financing; and a business that would plausibly survive, even at a smaller scale, if the specific financing arrangement in question disappeared tomorrow. Applied to the current landscape, OpenAI's large and growing base of paying subscribers is a genuinely positive signal on the first count, while its reported roughly $14 billion projected loss in 2026, set against a $100 billion revenue target by 2029, illustrates the scale of the gap that still needs to close before the arrangement's sustainability is beyond serious debate.

What is circular financing in AI infrastructure, and should telecoms and data centre operators be worried?

Industry coverage, including analysis from Capacity aimed specifically at telecoms and data center operators, frames circular financing as a legitimate operational concern for that sector because data center operators sit downstream of exactly the financing arrangements described throughout this piece — many rely on long-term capacity contracts with the same hyperscalers and AI labs whose spending is partly propped up by circular financing arrangements with their chip and cloud suppliers. If that spending slows or reprices, data center operators holding long-term capacity commitments or debt tied to expected future demand would feel it directly. The reasonable, practical response for operators in this position is the same one recommended for any business evaluating a counterparty in this environment: understand how much of a customer's demand is externally generated versus financing-linked before signing long-duration capacity agreements.

How do Microsoft, OpenAI, and Nvidia "keep paying each other" in these circular deals?

Bloomberg's reporting on this dynamic describes a repeating pattern where capital, credit, and infrastructure commitments flow among these three companies in overlapping directions rather than in one clean, linear direction. Microsoft has its own deep financial and infrastructure ties to OpenAI; Nvidia's chips and financing support OpenAI's compute buildout; and cloud infrastructure spending flows back toward the companies supplying the underlying hardware and platforms. The practical effect described in that reporting is that a dollar entering this system has a meaningfully higher-than-normal chance of eventually appearing as revenue for one of the other two companies in the loop, which is precisely the dynamic that makes analysts uneasy about how much of the reported growth across all three companies is mutually reinforcing rather than independently validated by outside demand.

Why did Jim Cramer compare AI's circular financing frenzy to the dot-com bubble?

Cramer's July 2026 comparison, made on CNBC, draws a direct historical parallel to the dot-com era, when telecom and infrastructure vendors extended financing to customers who used it to buy telecom equipment and services — a pattern that inflated reported demand for network buildout well beyond what real end-customer usage ultimately justified. When that mismatch became undeniable, the resulting unwind happened quickly and severely, rather than as a gradual, manageable correction. Cramer's warning isn't a claim that AI infrastructure spending is worthless; it's a claim that the financing structure itself has a specific, well-documented historical failure mode, and that today's circular arrangements share enough structural similarity with that earlier pattern to warrant real caution rather than dismissal.

How large is Nvidia's financial commitment to OpenAI, and what does it fund?

Nvidia's reported partnership with OpenAI is valued at roughly $100 billion, and in broad terms it supports OpenAI's ability to build out the compute infrastructure needed for frontier AI model training and deployment — infrastructure that in turn depends heavily on Nvidia's own chips. This is one of the clearest examples of the circular financing pattern described throughout this piece: capital that flows from Nvidia toward supporting OpenAI's buildout has a strong likelihood of flowing back to Nvidia as chip revenue, which is exactly the structural feature that has drawn regulatory and analyst scrutiny even though the arrangement itself is a publicly disclosed, negotiated commercial partnership between two sophisticated companies.

What was the reported $250 billion Nvidia "backstop" for OpenAI, and what data center project was it tied to?

Reporting has discussed a roughly $250 billion Nvidia backstop connected to a planned 10-gigawatt AI data center campus in Ohio being built for OpenAI. A backstop of this kind generally functions as a financial guarantee or support commitment that gives lenders, partners, or other stakeholders in the project greater confidence that it will be completed and funded even if other financing pieces run into trouble. At $250 billion tied to a single data center campus, the figure illustrates just how large individual pieces of the current AI infrastructure buildout have become, and how directly a single chip supplier's financial backing can be woven into the completion risk of one specific, enormous physical project.

What do AMD's roughly $200 billion in customer deals involving 10% equity warrants actually mean for AMD's balance sheet?

AMD's roughly $200 billion in customer deals include arrangements where AMD hands over equity warrants worth approximately 10% of the company to major customers as part of the agreements securing their chip purchases. Practically, this means AMD is giving up potential future ownership dilution and upside in exchange for locking in large-scale purchasing commitments from customers whose own success is, in part, defined by how much AMD hardware they continue buying. It ties AMD's future equity value directly to the fortunes of a small number of very large customers — a structure that can look attractive when those customers are growing quickly, but concentrates AMD's risk profile around the same handful of counterparties whose demand is itself partly a product of circular financing dynamics.

How large is Oracle's cloud infrastructure commitment tied to these circular financing arrangements?

Oracle has committed $300 billion to cloud infrastructure buildout — a figure that represents a major, multi-year bet that AI compute demand will continue materializing at or near currently projected levels. Oracle's customer base for this capacity substantially overlaps with the same small set of AI labs and hyperscalers involved in the other arrangements detailed in this piece, which means Oracle's own revenue realization from this commitment is, to a meaningful degree, tied to the same underlying question the Bank for International Settlements raised: how much of the demand behind this spending is genuinely external, and how much reflects the same money moving through an interconnected set of counterparties.

Why did the Bank for International Settlements name AI circular financing as one of the top three risks to global financial stability in 2026?

The BIS named circular financing, alongside a possible AI capital-expenditure bust and sovereign debt fragility, as one of the top three risks in its 2026 Annual Report because of how directly it can distort the visibility that markets, regulators, and even company boards have into real underlying demand. An institution focused on global financial stability is less concerned with any single company's fortunes and more concerned with contagion — how a problem in one tightly interconnected part of the financial system spreads to others. Circular financing, combined with a growing web of structured debt instruments financing the same infrastructure buildout, creates exactly the kind of interconnected exposure that has historically turned company-specific setbacks into broader financial stability events.

What total dollar value of circular financing arrangements has been estimated across the AI infrastructure sector?

More than $800 billion in circular financing arrangements has been identified across the AI infrastructure sector, spanning the Nvidia-OpenAI partnership, AMD's warrant-linked customer deals, Oracle's cloud commitment, and the broader web of interconnected financing and purchasing relationships among a relatively small number of major AI infrastructure companies. That figure is large enough, relative to the overall scale of the AI infrastructure buildout, to represent a genuinely significant share of total reported AI infrastructure spending — which is precisely why regulators and market analysts are treating it as a structural feature of the current AI investment cycle rather than a minor financing curiosity.

How is circular financing different from ordinary vendor financing that has historically helped build real industries?

The distinction is more about degree and structure than about any bright legal or accounting line. Ordinary vendor financing — aircraft manufacturers financing airline purchases, telecom equipment vendors extending credit to carriers — has legitimately helped build durable, real industries for decades, and remains common and generally healthy when the financed customer has meaningful revenue and demand from outside the financing relationship itself. What makes today's AI arrangements draw the "circular" label and the accompanying scrutiny is the scale (over $800 billion), the concentration among a small number of deeply interconnected firms, and the difficulty outside observers face in cleanly separating externally generated demand from financing-driven demand within the reported numbers. It's a spectrum, not a binary, and today's largest AI deals sit further along that spectrum than most historical vendor financing arrangements did.

What would indicate that AI infrastructure demand is "organic" rather than an artifact of circular financing?

The clearest indicator would be substantial, growing revenue generated from customers and use cases entirely outside the small cohort of companies currently financing each other's growth — enterprise AI adoption paying real subscription or usage-based revenue, consumer products with durable engagement and monetization, and industries outside tech incorporating AI tools into revenue-generating workflows at meaningful scale. If that kind of broad-based, externally sourced revenue continues growing and eventually dwarfs the revenue attributable to the interconnected financing cohort, it would substantially validate the current infrastructure buildout as demand-driven rather than financing-driven. The current uncertainty exists precisely because that broader validation is still an open, ongoing question rather than an already-proven fact.

How much is OpenAI projected to lose in 2026 while pursuing a $100 billion revenue target by 2029?

OpenAI is reportedly projected to lose approximately $14 billion in 2026, even while pursuing a longer-term target of $100 billion in annual revenue by 2029. That combination — a large near-term loss alongside an ambitious multi-year revenue target — is a common growth-stage pattern for companies making enormous upfront infrastructure investments, but the scale here is unusually large by almost any historical comparison, which is part of why the sustainability of OpenAI's spending commitments, including its financing relationships with Nvidia and Oracle, has become such a closely watched question rather than a routine one.

How large could the data center asset-backed securities (ABS) and CMBS financing market become in 2026?

Bank of America has projected that the data center asset-backed securities and commercial mortgage-backed securities financing market could top $35 billion in combined new issuance and refinancing in 2026. This matters beyond the headline number because it represents a growing channel through which AI data center financing risk gets distributed into broader credit and institutional investment markets — pension funds, insurance portfolios, and other investors who hold structured debt products — rather than remaining concentrated solely on the balance sheets of the technology companies directly involved in the circular financing arrangements described elsewhere in this piece.

What emerging litigation risks are law firms flagging around the financing of the AI data center boom?

Quinn Emanuel has published a client alert specifically addressing emerging litigation risks tied to financing the AI data center boom — a notable signal, since major litigation firms generally publish this kind of forward-looking client guidance when they anticipate real, near-term disputes rather than purely theoretical ones. While the specific content of anticipated disputes wasn't detailed further in the sources reviewed for this piece, the existence of this kind of alert reinforces the broader picture: sophisticated legal and financial actors are treating the current financing structure around AI infrastructure as carrying genuine, actionable risk, not simply as an interesting market narrative.

How does circular financing affect the way AI infrastructure demand is reported and perceived by investors?

Reporting on this dynamic makes the core distortion explicit: circular cash flows can make demand look organic and revenue look robust when much of it is the same dollars going around the same circle among a small group of companies. For an investor evaluating headline growth figures, contracted demand, or capital expenditure commitments, this means the reported numbers may overstate how much independent, external validation actually exists behind the current AI infrastructure buildout. It doesn't necessarily mean the numbers are wrong or misleading in an intentional or improper sense — but it does mean a more careful investor needs to look past the headline figures and ask how much of the underlying activity would still exist without the financing relationships connecting the parties involved.

What distinguishes an AI capex "bust" scenario from a circular financing "unwind" scenario in the systemic-risk analyses being published in 2026?

The Bank for International Settlements frames these as related but distinct risks. An AI capital-expenditure "bust" scenario describes a broader slowdown in AI infrastructure spending generally — companies simply pulling back on new data center, chip, and compute commitments because expected returns don't materialize as projected. A circular financing "unwind" scenario is more specific: it describes what happens if the interconnected web of financing and purchasing commitments among the small cohort of companies involved starts to break down, potentially triggering a faster, more contagious repricing across multiple companies simultaneously, precisely because their financial fates are so tightly linked. A capex bust could happen without a circular financing unwind, and vice versa, but the two risks could also compound each other — a capex slowdown could be exactly the trigger that exposes how load-bearing the circular financing relationships actually were.

Why might vendor-financed circular deals be harder to unwind than traditional bank lending if AI demand disappoints?

Traditional bank lending typically involves a lender with a diversified loan book, standardized collateral and workout procedures, and a degree of separation from the borrower's day-to-day commercial relationship with its suppliers. Vendor-financed circular deals tangle those roles together — the financier is also the supplier, and often also a customer, of the same company it's financing. If demand disappoints, unwinding the position isn't simply a matter of restructuring a loan; it potentially requires renegotiating the underlying supply relationship, the equity stake, and the financing terms all at once, often among companies whose other business dealings with each other continue regardless. That interconnectedness — the same feature that made the arrangements attractive when growth was strong — is exactly what analysts and the BIS worry could make an unwind messier and more contagious than a standard credit event.

How many distinct companies are actually involved in the "handful of interconnected firms" driving most reported circular financing volume?

Available reporting describes the phenomenon as cash looping "among a handful of interconnected firms" without specifying an exact, definitive count — and it would be inaccurate to invent a precise number where the sourced reporting doesn't provide one. What can be said accurately, based on the specific deals detailed throughout this piece, is that Nvidia, OpenAI, Microsoft, AMD, and Oracle are the entities most consistently named across the largest individual arrangements — a genuinely small group relative to the more than $800 billion in total value attributed to circular financing across the sector, which is itself part of what concerns analysts: an enormous amount of value concentrated among very few counterparties.

If you're evaluating an AI infrastructure or platform partner in this environment, our AI agents and automation practice can help scope a build around realistic, defensible assumptions rather than headline momentum, and our compliance page covers how we think about governance and risk in technical partnerships more broadly.