In recent weeks, you’ve probably come across a meme on social media that simulates several identical superheroes, pointing at each other in bewilderment. Superimposed on their faces, the names of firms such as OpenAI, Nvidia, Oracle, Microsoft o AMD. However, the reason for this comical image should give us reason to reflect, if not to worry.
The current awakening of generative artificial intelligence is being based on losses and numbers that do not add up. It is normal, after all, that investments in technology that is not yet mature take a while to turn into benefits. Up to this point it may even seem logical to us that companies like OpenAI are valued at $500 billion, with just $13 billion in annualized revenue and estimated losses for 2026 of $14 billion.
We are talking about OpenAI currently having a price-sales ratio of 38x, unheard of for companies in any sector other than technology. So much so that, to justify this assessment, Sam Altman’s brand would need to reach 225 billion in revenue by 2030 with a free cash flow margin of 27%. Something that, except for a surprise in the form of the advent of a general AI that seems quite unlikely with current models, seems difficult to happen.
But let’s assume, for the sake of this reflection, that these feet of clay do not exist and that the future of OpenAI is based on firm foundations. The real problem, and hence the meme that gives rise to this column, lies in the unprecedented circular financing structure that exists in the artificial intelligence market.
Let’s be clear: the AI funding ecosystem is characterized by a pattern whereby Suppliers invest in their own customers, who then use that capital to purchase products and services from those same suppliers.
I am not the first, far from it, to notice this textbook dissonance: analysts are increasingly concerned about this phenomenon. The Bank of England warned last month of “risks of sharp corrections in AI.” Morgan Stanley directly alludes to the fact that “providers are financing clients and sharing revenue; there is mutual ownership and increasing concentration. Increasingly intricate transactions make it difficult to assess real demand for AI and raise the risks associated with the success of this technology.”
The clearest, and who offers us a historical perspective of this paradigm, is Jeremy Grantham, from GMO. He compares the current situation with the 2008 financial crisis and the bubble dotcom. He even warns that this is similar to what happened with Cisco in the late 90s, when the company lent money to startups to buy its routersrecording those sales as income. When the bubble burst, Cisco lost 78% of its value.
The dangerous circularity
OpenAI is the epicenter of this interdependent financial scheme. At the moment, maintains infrastructure purchase commitments that exceed one billion dollars (trillion Anglo-Saxons), distributed among Oracle (300,000 million), Microsoft Azure (250,000), Nvidia (100,000), AMD (90,000), AWS (38,000) and CoreWeave (22,400).
First example of dangerous circularity: Nvidia agreed to invest up to 100 billion in OpenAI to finance the construction of data centers. Sam Altman, in turn, pledged to use that capital to purchase millions of Nvidia chips and deploy at least 10 gigawatts of infrastructure based on Nvidia technology.
In turn, Oracle signed a 300 billion contract with OpenAI to provide cloud computing services for five years, starting in 2027. If we take into account that the outstanding obligations of Larry Ellison’s multinational amount to 455 billion, this means that 66% of its future order book depends on the good performance of ChatGPT. In parallel, Oracle uses this revenue to purchase Nvidia chips, which in turn power the infrastructure that OpenAI uses.
Microsoft’s situation is even clearer: Satya Nadella’s team has invested 13 billion in OpenAI (it owns 27% of the company, valued at 135 billion). On the way back, OpenAI committed to purchasing 250 billion in Azure services from Microsoft.
AMD, Nvidia’s great rival, is also no stranger to this wheel. The chip giant granted OpenAI an option to acquire up to 160 million shares (roughly 10% of the company) in exchange for OpenAI commits to deploying 6 gigawatts of AMD GPUs. We are talking about approximately 90,000 million in potential income. The deal is just as circular as all the previous ones: AMD provides funding to OpenAI, who then uses those funds to purchase its silicon.
And, closing the most immediate circle, CoreWeave. This AI infrastructure provider is more than 5% owned by Nvidia, but at the same time maintains contracts worth $22.4 billion with OpenAI. With the support of this demand, it is purchasing 6.3 billion in chips from Nvidia, who has also committed to buying any residual capacity left over from this supplier. The scheme has no loss: the money enters and leaves through the same pocket.
But don’t think that this financial practice is exclusive to Altman and his acolytes. Anthropic -one of OpenAI’s big rivals- received an investment of $3 billion from Google and another $8 billion from Amazon in exchange for committing to using one million TPU chips from the search giant and another million Trainium chips from AWS. Call it coincidence, call it whatever you prefer.
Elon MuskOf course, I had to participate in this party. Its scorned alternative to OpenAI, xAI, raised $20 billion in a funding round in which Nvidia contributed a tenth. And the compensation, oh surprise!, is the purchase of Nvidia chips worth 18 billion.
Debt and more debt
It is clear at this point that the money never leaves the pockets of the technology providers who finance the AI actors who in turn buy their infrastructure to train and run their models. But any avid reader will be wondering where so much money comes from in any case.
The answer is as obvious as it is alarming due to its systemic risk: debt.
Meta, Google, Microsoft and Amazon alone have spent around $200 billion on AI infrastructure in 2025. As if that seems like a small amount, Google has revised its 2025 capital spending forecast from $75 billion to $93 billion. Microsoft expects its investment in this plot to increase by 74%, to 34.9 billion this year. And Meta has adjusted its spending forecast to a minimum of 70,000 million by the end of the year.
Bank of America has immediately warned that 75,000 million in debt of US investment grade linked to AI-focused technology companies occurred in just two months, September and October 2025, more than double the sector’s annual average of 32 billion between 2015 and 2024. This includes debt issues of 30,000 million by Meta and another 18,000 million by Oracle.
It is not the only bank to show signs that something is not right. UBS estimates that the AI-related private credit loans may have almost doubled in the last twelve months. Meanwhile, Morgan Stanley predicts that debt markets could account for more than half of the $1.5 trillion required for data center expansions through 2028.
The essential question is not whether AI is disruptive – it is, and to an excellent level – but whether the way we are financing its deployment is sustainable. The companies that embody the cutting edge of artificial intelligence They are simultaneously raising their own pedestal and the trap that lurks beneath it.. Circularity has become the new structural obstacle: I invest in you so that you buy from me, and thus, like a financial palimpsest that is rewritten without completely erasing what came before, capital is recycled in a closed cycle that does not always create new value, but rather redistributes the existing one, like an accounting twist worthy of a compendium of financial engineering.