There may be no better image for the strange new economics of artificial intelligence.

I got reminded of the tail-devouring serpent when I read the news of Google-parent Alphabet committing to invest up to $40 billion in arch-competitor Anthropic, maker of Claude. On the face of it, this is just another mega-deal in the AI arms race, following similar deals by Amazon and Microsoft. But, on reflection, it begins to look more like the AI Ouroboros.

Google makes Gemini, a direct competitor to Claude. Reportedly, Google is also trying hard to build stronger coding capabilities inside Gemini, with founder Sergey Brin leading the charge. There have been reports of a divide between Google employees who have access to Claude and those who do not, with some engineers believing Claude performs better for coding than Gemini. And yet, here is Google putting tens of billions behind Anthropic.

Anthropic is the most fascinating creature in this circus. Amazon has already invested heavily in the company and has now planned up to $25 billion more. Google is investing in Anthropic and providing cloud and TPU capacity. Microsoft and Nvidia were also reported last year to be part of a deal in which Anthropic would buy $30 billion of Azure compute powered by Nvidia, while receiving investments from both.

We are witnessing a bizarre new economic loop. The investee (Anthropic) takes the investor (Google)’s money and immediately hands it back to the investor to pay for the massive cloud compute and specialized AI chips required to train their models. The money never truly leaves the ecosystem; it simply rotates. The investor gets to report massive growth in their Cloud division, and the investee gets the compute they need to survive another year.

In the old world, competitors fought. In the AI world, competitors invest in each other, sell to each other, buy from each other, host each other, and then announce victory over each other! In the $350 billion AI economy of 2026, it represents a new, high-stakes financial architecture that we might call Circular Capitalism. Anthropic is not alone. AI model pioneer and ChatGPT creator OpenAI created circular capitalism, with investments from Nvidia, Microsoft and others; Anthropic has just gone one step further by taking money from everyone!

There is a perfectly rational explanation for all this, especially for Google and Amazon, who want to be invested in AI’s current superstar, sell their cloud compute, their AI chips, and spread their risks around. Google does not want to put all its eggs in the Gemini basket. It wants Anthropic to buy its cloud compute and TPUs. For Anthropic, it needs the money, and even more than that, it needs the compute. Its astronomical growth has meant that capacity has not kept up with demand, throttling its demanding, high value users of Claude.

Microsoft seems to be a loser here. It was once seen as the clear early winner of the AI boom because of its OpenAI partnership. But the alliance has visibly loosened. OpenAI wants independence, more compute, more partners and a trillion dollar IPO. Microsoft still has a powerful position, but it has lost its first-mover advantage, and is no longer the only enterprise doorway into frontier AI. Its own internal efforts are lagging the Big Three of OpenAI, Google and Anthropic, and the stock market has acknowledged that.

The bigger question, however, is not who wins this quarter. It is whether these circular deals create real value or merely the appearance of value. It seems that money leaves one pocket, returns through another, and in the meantime everyone’s valuation rises. This is not entirely new. During the late-1990s telecom bubble, equipment makers such as Lucent and Nortel extended billions in financing to telecom operators, who then used that money to buy networking equipment from the same suppliers. Sales boomed. Revenues looked strong. Wall Street cheered. Then many of the customers failed, and the circular magic turned into circular pain.

That does not mean AI is a replay of telecom. The demand for AI is real. People are using ChatGPT, Claude, Gemini and coding agents every day, enterprises are rebuilding workflows, and developers are producing software differently. Unlike some fibre networks built before customers existed, AI already has users. But real demand does not automatically justify every valuation, data centre, gigawatt and cheque.

These deals are not necessarily fraudulent, but they are certainly opaque - it sometimes becomes hard to know where genuine customer demand ends and financial engineering begins. The IMF warned that circular AI financing could create systemic risk by inflating revenues and valuations and increasing interdependence between a small set of powerful firms.

And yet, the Ouroboros has a second meaning. It is not only self-consumption, it might also be renewal. The chemist August Kekulé famously claimed that the ring structure of benzene came to him after a dream of a snake eating its own tail. The ring became the basis for modern organic chemistry. Perhaps AI’s circular deals will create the same kind of ring: a self-reinforcing structure that funds the infrastructure, models and applications needed for the next technological age. Or perhaps the snake will discover, too late, that it has swallowed itself whole.

History does not repeat itself, as Mark Twain may or may not have said, but it does rhyme. In every age, the builders of the future first fight over who invented it, then who controls it, and finally who gets to profit from it. The Oakland trial is AI’s version of this ancient battle.