The Rope With Which Tech Giants Will Hang Themselves

The International Monetary Fund (IMF) has just dropped a bombshell on financial markets, and the shockwave promises to be historic. In an official warning that starkly contrasts with the prevailing euphoria on Wall Street, the international institution points the finger at a silent, complex, yet formidably effective mechanism: circular financing arrangements linked to artificial intelligence have crossed the staggering threshold of $1.4 trillion. This dizzying figure represents half the gross domestic product of a global power like France, built entirely on a web of accounting entries and cross-promises.
Although concealed beneath thick layers of technological jargon and opaque strategic agreements, the fundamental mechanism remains disarmingly simple. Tech giants and semiconductor manufacturers are investing massively in their own clients, acting as guarantors for their colossal debts in a ballet where this money circulates and then reappears on their own balance sheets as revenue. This is the essence of authentic perpetual motion applied to contemporary finance. It is a closed-loop system where the expense of one guarantees the income of the other, all while inflating the valuation of the entire ecosystem.
Let us illustrate the scale of this frenzy through seventeen major operations made public to date, whose amounts defy comprehension and outline the contours of a systemic bubble.
At the center of this whirlwind of capital is OpenAI, the figurehead of the generative revolution. Indeed, the relationship between Sam Altman’s startup and chipmaker Nvidia perfectly illustrates this vertigo. Nvidia did not hesitate to inject $30 billion into OpenAI while acting as a guarantor for its Ohio data center lease for a maximum amount of $105 billion. The objective being for OpenAI to scrupulously fill these infrastructures with Nvidia processors. The guaranteed money thus instantly transforms into an order book for the guarantor.
This logic of financial symbiosis extends to all the titans of the global cloud. Oracle signed a $300 billion cloud contract with OpenAI, taking on massive debt to build the physical infrastructure required by this pharaonic agreement. A historical player, Microsoft holds about 27% of OpenAI’s capital, securing in return a colossal spending commitment of $250 billion on its own Azure cloud. Fearing being sidelined, Amazon is contributing $50 billion to OpenAI. In exchange, the creator of ChatGPT is increasing its orders with AWS, Amazon’s cloud hosting subsidiary, by $100 billion. Not wanting to be left out, the Japanese conglomerate SoftBank is offering $64.6 billion to OpenAI, which logically leases data centers in exchange from SB Energy, a SoftBank subsidiary. AMD is granting OpenAI warrants allowing it to acquire up to 160 million shares at the derisory price of $0.01, ensuring through this nearly free gift that OpenAI stocks its servers with AMD chips.
However, OpenAI is not the sole center of gravity in this financial galaxy. Anthropic, its great rival, finds itself at the heart of equally staggering financial engineering, weaving a web of cross-obligations. Google has promised to invest up to $40 billion in it, on the express condition that Anthropic commits to spending at least $121 billion on Google Cloud. The exact same pattern is repeating with Amazon, with $33 billion in investments against $110 billion in guaranteed spending on its infrastructure. Microsoft and Nvidia have even forged an unprecedented alliance to jointly inject up to $15 billion into Anthropic, locking in a $45 billion spending commitment on the Redmond firm’s servers along the way.
It must be said that the Anthropic ecosystem pushes the concept of circular financing to its most creative limits. A banking consortium is lending $63 billion to a Special Purpose Company (SPC) whose sole purpose is to buy Broadcom chips to lease them to Anthropic. Broadcom went so far as to guarantee the residual value of these used chips out of its own pocket to reassure creditors. Even more surprisingly, Elon Musk’s space company, SpaceX, is taking out a massive $40 billion loan to buy Nvidia chips and lease this computing power to Anthropic. The crowning detail: Nvidia is among SpaceX’s shareholders.
The apotheosis of this incestuous system manifested during Anthropic’s recent IPO. The main players in the value chain—Nvidia, Google, AMD, Micron, Broadcom, and Microsoft—effectively united to acquire $26 billion in shares. While the press sees this as a mark of unwavering confidence in the future of AI, it is in reality a closed loop: to survive, Anthropic will in any case be forced to pay back hundreds of billions of dollars to these very same companies to lease the computing power (servers and chips) essential to run its AI.
Beyond model designers, this financial IV drip is force-feeding the entire hardware infrastructure. Aware that its hyper-growth depends on the market’s absorption capacity, Nvidia took an 11% stake in CoreWeave while committing to buy back $6.3 billion of unused capacity from it. Naturally, CoreWeave uses these secured funds to acquire new Nvidia chips. It is a model the manufacturer has replicated on a large scale by committing to buy back $36 billion in capacity from mid-sized cloud providers (the “neoclouds”). Armed with these guaranteed purchase revenues, these players take out loans to… supply themselves from Nvidia.
As for Microsoft, it is replicating this expansionist strategy with infrastructure operators and reconverted former Bitcoin miners like Core Scientific, Nebius, and IREN, signing contracts reaching $70 billion. Backed by this guarantee, these operators borrow and immediately reinvest every cent to buy the famous GPUs, the chips essential for AI. Google is following suit by guaranteeing leases for TeraWulf and Cipher Mining data centers to the tune of several billions—in exchange, of course, for equity stakes. Finally, Meta is not immune to this fever, structuring a joint venture with Blue Owl to build a $27.3 billion data center, while providing its support via an overwhelming $28 billion guarantee.
This mirage is perfect—as long as the money is flowing and debt accumulates without macroeconomic friction. Tech giants, semiconductor manufacturers, and data center operators all boast dazzling balance sheets, revenue growth that seems to defy the laws of gravity, and quarterly results consistently exceeding expectations.
The stock market applauds, the loop is closed, everyone is happy.
The breaking point is written into the very source code of this financial architecture, because this colossal $1.4 trillion orgy rests on a single, disarmingly fragile premise: massive and lasting adoption by the end consumer. Indeed, this oversized model can only survive if traditional businesses, developers, and individuals spend limitlessly on artificial intelligence services.
This is where the IMF’s warning resounds like a thunderclap, because if monetization at the final link of the chain proves disappointing, and if the true utility of AI struggles to generate the hoped-for subscriptions and productivity gains, the house of cards will be swept away. Gorged on debt to finance stratospheric capital expenditures (CAPEX), AI model companies like the neoclouds will be the first to break. It is a contagion that will not stop at these weak links because—bound hand and foot by their capital investments and gigantic lease guarantees—the tech giants will have no alternative but to absorb these cascading defaults onto their own balance sheets.
As long as the liquidity pump is working, we are spectators to the most spectacular perpetual motion machine in the history of modern capitalism. If the record skips, the loop will transform into a domino effect of unprecedented violence.
Dear readers,
This blog is yours: I maintain it diligently, with both consistency and passion. Thousands of articles and analyses are available to you here, some dating all the way back to 1993!
What were once considered heterodox views on macroeconomics have, over time, become widely accepted and recognized. Regardless, my positions have always been sincere.
As you can imagine — whether you’re discovering this site for the first time or have been reading me for years — the energy and time I dedicate to my research are substantial. This work will remain volunteer-based, and freely accessible to all.
I’ve made this payment platform available, and I encourage you to support my efforts through one-time or recurring donations.
A heartfelt thank you to all those who choose to support my work.