Are we all living in an AI bubble? Inside circular trading and the silent financial loop it has created

Three years ago, generative AI was hardly on anyone’s radar. Three months ago, circular trading was only being whispered about in closed-door meetings at major banks. A few weeks, a few days ago, these loops were still a niche topic. Today, those are all things anyone on Wall Street can talk about.As investment grows, from $100 billion chip deals to multi-gigawatt data center projects, so do the questions. The first question is: What does that mean? And are we among them? AI bubble? More and more AI companies and their partners are working on “circular” financing. This is a loop in which the same companies that pour money into AI startups also sell them the infrastructure they need to survive.It has created what could be described as an ouroboros ecosystem, a self-feeding financial creature that devours its own tail. And as the numbers grow, from $100 billion chip deals to multi-gigawatt data center projects, questions are growing about how sustainable this loop really is.

In the Loop: What is circular trading?

Circular trading is simple in theory, but very powerful in practice. AI companies take huge investments from tech giants and promise to return that money to those same investors by investing in cloud computing, chips, and infrastructure.Consider one of the most obvious examples. Microsoft and Nvidia’s multi-billion dollar effort for Anthropic. The startup received about $15 billion from both giants, while also pledging to spend $30 billion on Microsoft’s Azure cloud and Nvidia GPUs. On paper, this looks like a huge show of confidence for Anthropic. In reality, much of that investment will flow directly into Microsoft and Nvidia’s pockets through long-term infrastructure contracts.

OpenAI, a flagship product in this space, takes this circularity even further. Nvidia has reportedly committed to investing up to $100 billion. OpenAI is one of the largest purchasers of Nvidia GPUs on the planet. AMD struck a similar deal, offering next-generation chips while also granting OpenAI warrants to buy AMD stock at a symbolic price. In other words, OpenAI will pay AMD for the hardware and at the same time hold a stake that will benefit if AMD is successful.Meanwhile, Oracle is deeply integrated as OpenAI’s cloud backend. The two companies are building a new AI super data center in a partnership valued at more than $300 billion. OpenAI is currently being provided computing power, but it has committed to paying Oracle roughly the same amount over the next five years.The deal between SoftBank and OpenAI will take place in the same area. SoftBank will inject capital into the company and at the same time build the infrastructure for OpenAI’s Stargate project, with OpenAI pledging to pay for it over the next 10 years.This loop is tight, profitable, and increasingly large.

But why do they exist?

The most limited resource of the AI ​​boom is not people or data, but computers. Training frontier AI models requires hundreds of thousands of GPUs and tons of power.This is why many transactions are measured in gigawatts (GW) rather than dollars.One gigawatt is an incredible unit.1 GW ≈ 750,000 household electricity consumption.Next-generation AI data centers will increasingly require sites of 1 to 4 GW each. OpenAI’s planned megasite in Texas is expected to consume as much power as a small country. These sites require dedicated substations, transmission lines, cooling facilities, and, in some cases, on-site power generation.

With significant capital expenditures, AI companies need partners who can provide cash and infrastructure. Cloud providers and chipmakers fit the bill, but they recoup their investments by locking startups into large, long-term spending commitments.The economics are daunting. McKinsey estimates that meeting global AI computing demand will require $7 trillion in capital investment by 2030. Bain issued a similar warning, noting that the difference between current AI revenues and the future revenues needed to sustain that spending is hundreds of billions of dollars a year.

So are we in a bubble?

An easy way to answer whether there is an AI bubble is to start with the generative model itself. For example, when ChatGPT is asked, “Are we in an AI bubble?” we answer, “Yes (in the short term).”“In the short term, yes, there is too much momentum. Some of the funds are going to speculative play and not all AI startups will survive,” ChatGPT opined.“Medium to long term: We don’t see AI going away, even if there are ‘pops’ and fixes. The infrastructure (data centers, chips, cloud) and real-world use cases are likely to persist and mature,” the company said in its long-term forecast.Elon Musk’s Grok agreed that we are in an AI bubble, but expressed optimism that it is not “the kind that will inevitably end in a complete disaster like the dot-com crash or 2008.”“Yes, I think we’re in an AI bubble, but not the kind that inevitably ends in complete disaster like the dot-com crash or 2008. Rather, it’s more like a frothy speculative groundswell built on real technological expectations, where valuations are ahead of immediate reality and corrections feel premature,” Musk’s chatbot said.

Gemini described the current moment as a small bubble of investor excitement riding on top of much larger, longer-term technology change. In its view, the underlying technology remains stable, so only weak players and short-term bets are at risk.But to really understand whether there is a bubble or not, First you need to check the size of your investment.

bigger market pharmaceutical

Nothing better represents the scale of speculation than Nvidia. The company’s market capitalization is now larger than the entire global pharmaceutical industry combined. This comparison is surprising. The pharmaceutical industry is a 100-year-old industry with regulated demand, stable profits, and essential products. Nvidia is working on one topic, AI computing, for which long-term commercial benefits have yet to be proven.Investors are embracing the idea that AI will reshape the global economy. But these valuations are so large that some analysts warn that even small disappointments in AI adoption could cause a disproportionately large correction.This is one reason why prominent investors like Michael Burry of “Big Short” fame have reportedly shorted companies closely tied to the AI ​​boom, including OpenAI-related investments and Palantir. His bearish stance shows that not everyone is convinced the sector can maintain its current speed.

OpenAI revenue and pricing issues

OpenAI reportedly generated approximately $12 billion in annual revenue. While the numbers represent significant traction for the eight-year-old company, operating costs still exceed revenue. Frontier models like GPT-5 cost billions of dollars to train, and inference costs (the cost of running the model) increase with user demand.This raises uncomfortable questions such as:Will raising prices solve OpenAI’s financial pressures?It’s possible, but it’s expensive.OpenAI’s mainstream scaffolding relies heavily on consumer adoption and developer usage. Its subscription tiers, API pricing, and enterprise agreements are designed to drive volume, not maximize profits. Significant price increases could slow the growth on which its entire valuation depends.Additionally, the company faces strong competition from Anthropic, Google DeepMind, Meta, Mistral, and a rapidly emerging open source ecosystem. In a market shaped by hype and experimentation, raising prices too aggressively risks driving developers elsewhere.OpenAI thus finds itself in a classic innovator’s dilemma.Charge extra and risk losing scale, or keep your prices low and burn cash.

How Oracle became a surprise winner

Oracle may seem like an unusual company to benefit from the AI ​​boom, but its association with OpenAI is one of the most strategic in the industry.Oracle’s cloud business has traditionally lagged behind AWS, Azure and Google Cloud. Enter AI: The OpenAI partnership repositions Oracle as a key provider of GPU capacity, filling data centers with high-margin compute workloads.The $300 billion-plus partnership is effectively a circular loop.

This arrangement is lucrative, but it also comes with risks. If AI growth slows, Oracle could repeat the communications overcapacity of the dot-com era, leaving the company with an overbuilt infrastructure.

Echoes of the dot-com bubble – and what made it burst

The parallels to the technology boom of the late 1990s are more than just rhetoric. The system was similar, with companies inflating each other’s valuations through investment loops and customer financing schemes.But a more important lesson comes from why the dot-com bubble burst.It was not due to one factor, but a series of external shocks.

What destroyed the dot-com boom was not a failure of technology, but a collapse of financial infrastructure.AI is now building something even bigger and more interconnected.

Has AI faced similar shocks?

So far, the AI ​​sector has survived all potential disruptions.

• Regulatory oversight is much talked about, but it is not existential.
• OpenAI’s internal governance turmoil didn’t derail the company.
• Supply chain shortages have reduced GPU availability, but not stopped it.
• High interest rates won’t dry up funding for AI megaprojects.

In other words, the external shock has not yet arrived.But history suggests that when investment loops are so tightly intertwined, a single disruption, regulatory blow, liquidity crunch, accounting scandal, or sudden slowdown in AI adoption can lead to a violent re-pricing of the system.

So is this sustainable?

Investor opinions are divided.bullish case

1. AI is transformative
2. Computing demand is increasing
3. Initial investment pays dividends for decades
4. Circular trading is just a modern way to bootstrap hyperscale infrastructure

bearish case

1. Valuations of this sector rely on uninterrupted optimism
2. Cash flow is hidden by an inner loop
3. Capital expenditure requirements are too high relative to current revenues
4. One external shock can cause cascading losses

In this sense, the AI ​​boom is at a crossroads similar to 1999-2000. Dazzling technological possibilities are intertwined with fragile financial engineering.

A revolution built on loops

Circular AI finance is great, aggressive, and extremely dangerous. This shortened the fastest infrastructure build in the history of technology to just a few years. But it has also created a tightly coupled financial ecosystem in which companies invest in each other, buy from each other, and are highly dependent on each other’s success.If AI delivers the productivity spikes predicted by its advocates, the investment loop will become visionary. If it stagnates, ouroboros may finally bite hard and unravel the most expensive technology cycle in a generation.Over time, and perhaps an external shock, it will become clear in which direction the loop bends.