OpenAI announced that it has raised 110 billion dollars (mM$) in its most recent round of financing; Amazon invested 50mM$, Nvidia 30mM$ and SoftBank 30mM$. OpenAI buys GPUs from Nvidia. OpenAI invests in CoreWeave. CoreWeave leases GPUs from Nvidia and sells compute back to OpenAI. Meta makes a 14 million$ agreement to guarantee capacity cloud CoreWeave GPU. Microsoft puts $13 billion into OpenAI and $5 billion into Anthropic. OpenAI commits to investing 300bn$ [ao longo de cinco anos] to finance Oracle’s construction of a data center of 4.5 gigawatts (GW) as part of its megaproject Stargate.
Oracle takes on $105 billion in debt plus $248 billion in off-balance sheet leases to build the Stargate. 57% of your pipeline of revenue goes back to OpenAI. OpenAI order 10bn$ to Broadcom to build chips AI solutions tailored to address the projected exponential growth in demand for generative Artificial Intelligence (AI) tools and products.
Let’s look, in more detail, at a concrete example involving Nvidia, OpenAI and AMD. In September 2025, Nvidia and OpenAI announced a historic partnership to deploy 10GW of AI supercomputing capacity using Nvidia systems. To support this, Nvidia has announced its intention to invest up to $100mn in OpenAI – essentially $10mn/GW of compute delivered, with Nvidia directly funding the expansion of OpenAI with phased investments as each piece of infrastructure comes online.
Shortly after, OpenAI and AMD revealed an agreement in which OpenAI commits to purchase 6GW of chips of AMD’s AI, valued at approximately $90bn, granting AMD to OpenAI warrants for up to 160 million AMD shares (c. 10% of the share capital, for US$ 0.01), to be acquired in tranches as OpenAI buys more chips of AMD AI – until reaching 6GW.
According to the Reuters“the first installment will be released after the initial shipment [1GW] of the chips MI450, expected in the second half of 2026. Remaining milestones include specific AMD stock price targets reaching $600 per share for the final tranche to be unlocked.” In other words, OpenAI is literally betting on the future price of AMD shares as part of its payment.
One report noted that these acquisition thresholds reach an ambitious value of 600 dollars per share for AMD – a reflection of how much share price momentum is incorporated into the structure of the deal.
Microsoft and Meta are also in this network – Microsoft with its large shareholding and multi-year partnership in cloud with OpenAI, and Meta as a major buyer of hardware of AI – effectively funding much of the underlying infrastructure that Nvidia, OpenAI and AMD rely on.
It is [e outras] companies became intertwined in a circle of huge financing and investments [essencialmente em infraestrutura de IA]enhancing the capabilities (and valuations) of interconnected companies, in a kind of AI financial Ouroboros: a closed circuit in which the same capital circulates through different companies, each accounting for this input as revenue, investment gain or strategic partnership value.
In mythology, an Ouroboros is a snake [ou um dragão] which swallows its own tail, forming a virtuous circle.
Many analysts consider that there is a convincing strategic logic – AI is the new platform, and it makes sense that the companies involved in its development are establishing solid partnerships and investing heavily so that a visionary and collaborative leap guarantees them a good share of a future market with enormous potential.
In any case, these investments contribute to building a real, cutting-edge infrastructure. In this sense, the cycle of interconnected investments and financing can be rational and mutually beneficial for all these companies.
This AI financial Ouroboros raises several concerns. The main one, from the outset, consists of determining how many of these announced operations will correspond to effective transactions and how many are essentially financial mechanisms aimed at facilitating the financing of these companies and increasing the value of their shares.
These capital flows boosted the valuations of companies in the sector to very high levels, evoking the ghost of past crises in the financial markets. The S&P 500’s projected price-to-earnings (P/E) ratio is elevated at around 25, reflecting high earnings growth expectations. While this value is high by historical standards (the long-term average is around 18), it is still slightly below the peak of the internet bubble, when projected P/E ratios exceeded 30x.
The market context is also different – the current “Magnificent Seven” [empresas gigantes] of technology collectively represent 33% of the market capitalization of the S&P 500, a greater concentration than the 27% at the height of the internet bubble; However, today’s big tech companies are much more profitable than companies [startups] from was dot-com.
The high extent of debt-financed leverage underlying this unprecedented expansion of AI infrastructure remains concerning; Even the richest companies are tapping the credit markets and stretching their balance sheets.
While this financial Ouroboros of American AI companies continues to spin – driven by technological advances and a good dose of euphoria-driven capital – and inflate the value of the technology companies involved, it needs to continue to increasingly devour its own tail to survive. Equity-based deals that depend on achieving extreme share prices, debt-financed bets that require exponential growth, and high P/E and P/E multiples echo the characteristics of classic bubbles.
Although it is possible that this serpent of capital will catalyze a virtuous cycle with the AI revolution living up to expectations, the risk analyzes carried out on some of OpenAI’s (and Oracle’s) most recent financing operations show that it is also quite possible that we will see an Ouroboros gradually eating more and more capital, ending up, in the end, devouring the dreams of those responsible for technology companies and their investors and financiers.
Leave a Reply