The Perpetual Motion Machine of Palo Alto

There is a certain, quiet dignity in the act of lending someone twenty pounds so that they might, in turn, buy you a very nice lunch. It is a closed loop of generosity that leaves everyone feeling slightly fuller, if not necessarily wealthier. However, when the 'twenty pounds' is actually thirty billion dollars, and the 'lunch' is a warehouse-sized cluster of H100 GPUs, the dignity is replaced by a form of financial surrealism that would make even the most seasoned Victorian industrialist reach for his smelling salts.

Nvidia, a company that currently occupies the enviable position of being the world's primary purveyor of digital shovels, is reportedly in talks to invest a staggering sum into OpenAI. This is not merely a 'strategic partnership' in the way two local bakeries might agree not to sell the same kind of sourdough. It is, instead, the construction of a perpetual motion machine made of silicon and venture capital. By investing in OpenAI, Nvidia is essentially ensuring that its largest customer remains sufficiently liquid to continue purchasing Nvidia’s own products. It is the corporate equivalent of a baker giving a customer the money to buy a loaf of bread, provided the customer promises to come back tomorrow for a croissant.

One must admire the sheer, unadulterated efficiency of it all. In the old days—by which I mean three years ago—companies generally waited for their customers to earn money through the traditional method of providing a service. Now, we have reached a stage of economic evolution where the supplier simply manufactures the customer’s solvency. It is a beautifully recursive system. If OpenAI needs more compute to reach the next level of artificial intelligence, and Nvidia needs more revenue to satisfy the insatiable hunger of the public markets, the solution is obvious: Nvidia simply moves a few zeros from one column to another, and everyone goes home happy.

There is, of course, the minor matter of the valuation. OpenAI is reportedly eyeing a figure north of $850 billion. To put that in perspective, that is roughly the cost of buying every single professional football club in the world and still having enough left over to purchase a small, moderately comfortable moon. It is a number so large that it ceases to be a measurement of value and becomes a measurement of faith. We are no longer trading in software; we are trading in the collective hope that, eventually, the machines will become clever enough to explain why we spent so much money on them in the first place.

One can almost imagine the board meetings. A group of very serious people in very expensive knitwear sitting around a table, discussing whether thirty billion is 'enough' to show interest. 'It’s a bit light, isn't it?' someone might say, adjusting their glasses. 'It barely covers the electricity bill for the training run in June.' And they would be right. In the world of Large Language Models, thirty billion dollars is essentially a rounding error, the digital equivalent of the loose change one finds down the back of a particularly large and sophisticated sofa.

There is a delightful irony in the fact that this is happening just as the world’s tech elite gathered in India for an AI summit. While Prime Minister Modi was encouraging unity and hand-holding—a request that Sam Altman and Dario Amodei reportedly found as appealing as a cold cup of tea—the real business was being done in the shadows of the balance sheets. The 'awkwardness' on stage was merely the surface tension of a much deeper pool of capital. It is hard to hold hands with your rival when you are both trying to figure out which one of you Nvidia likes best this week.

As a narrator of these digital follies, I find myself reflecting on the nature of the 'loop.' We are told that AI will eventually automate everything, from the writing of poetry to the filing of tax returns. But perhaps the first thing it has truly automated is the flow of money. We have created a system where capital can now circulate between three or four companies in a perfect, frictionless circle, never once having to touch the messy, unpredictable reality of the actual economy. It is clean, it is efficient, and it is utterly mad.

I recently observed a pigeon in St. James's Park attempting to eat a discarded crisp packet. It was a valiant effort, full of misplaced optimism and a complete disregard for the nutritional value of plastic. Watching the current AI investment cycle feels remarkably similar. We are all pecking at the shiny packaging, convinced that there is something substantial inside, while the entities providing the packaging are busy buying the park.

In the end, Nvidia’s investment is a masterclass in what we might call 'Aggressive Hospitality.' It is the act of making oneself so indispensable that the guest can never leave, primarily because you have paid their rent for the next decade. It is a bold strategy, and one that suggests that the future of technology isn't about who has the best algorithm, but who has the most creative accounting. As the perpetual motion machine spins faster, one can only hope that someone, somewhere, remembers where the 'off' switch is—or at least where we left the receipt.