An Unusually Quiet Trillion-Dollar Evaporation

It is a generally accepted rule of arithmetic that a trillion is a very large number indeed. It is the sort of number that usually requires several planets, or at least a very long tape measure, to comprehend. Yet, in the span of less than two months, the financial markets have managed to misplace precisely this amount of valuation from the books of the world's premier graphics processor manufacturer. It has not been left down the back of the sofa, nor has it been left on a late-night bus to Croydon. It has simply ceased to exist, leaving behind only a slightly embarrassed silence and a large number of very warm graphics cards.

For the past two years, the graphics processing unit has been treated with the sort of reverence usually reserved for fragments of the True Cross or particularly successful sourdough starters. Entire corporate strategies were built on the assumption that if one piled enough of these small, green rectangles into a room and fed them enough electricity to power a medium-sized European principality, something resembling divine wisdom would eventually emerge. The processors were housed in vast, air-conditioned cathedrals, tended to by high priests in black turtlenecks who spoke in a dialect consisting entirely of three-letter acronyms and vague promises of a post-labor paradise.

The difficulty with divine wisdom, of course, is that it rarely submits a quarterly revenue report. Eventually, a gentleman in a grey suit from the accounts department will knock on the door and ask, with the polite persistence of a man who has a train to catch, what exactly the green rectangles have been doing to justify their keep. The answer—that they have been writing slightly better emails and generating pictures of cats wearing bowler hats—has proved to be somewhat unsatisfying to the people who write the cheques. The market, having spent twenty-four months in a state of breathless anticipation, has suddenly asked to see the receipt.

I once spent an afternoon observing a senior vice president of a major retail bank attempt to explain to his board why they had spent forty million dollars on a cluster of processors that were currently occupied with predicting which of their customers was most likely to buy a second-hand lawnmower. The board listened with the glazed, polite expressions of people who had long since accepted that their lives were being governed by forces they could neither see nor influence. When the vice president finished his presentation, the chairman paused, took a sip of water, and asked if the bank might not have achieved a similar result by simply asking the branch managers.

This brings us to the current state of affairs, which might be described as the Great Silicon Sobriety. The market has not so much panicked as it has simply sat down, taken off its spectacles, and rubbed its eyes. The realization is dawning that artificial intelligence is not a cathedral to be worshipped, but rather a utility—like municipal sewage or the supply of reasonably reliable paperclips. It is useful, certainly, but one does not generally trade at a hundred times earnings for the privilege of owning the pipes. A pipe is a pipe, regardless of how quickly the water flows through it.

This shift is being accelerated by the sudden appearance of much cheaper, more sensible alternatives. When one can purchase a perfectly adequate, if slightly less poetic, piece of logic for half the price of the premium brand, the incentive to maintain a trillion-dollar altar begins to fade. The high-end models are beginning to look like those extremely expensive, multi-functional Swiss Army knives that include a fish-scaler and a magnifying glass; marvelous in theory, but ultimately less useful than a simple pair of scissors when one merely needs to open a parcel. The world, it turns out, does not need a supercomputer to draft a polite rejection letter to a job applicant.

There is a quiet corner of a data center in Frankfurt where, I am told, a row of top-tier processors sits entirely idle, having been programmed to wait for a query of sufficient intellectual weight to justify their power consumption. They have been waiting since April. In the meantime, a much smaller, cheaper model next to them is happily processing thousands of requests regarding the optimal temperature for brewing Earl Grey tea. The smaller model is not particularly wise, but it is remarkably efficient, and it does not require its own dedicated substation to function.

The evaporation of Nvidia's trillion dollars is therefore not a tragedy, but a return to the comfortable, boring reality of ordinary economics. The world is a place where things must eventually pay for themselves, or at least have the decency to fail in a way that doesn't require a bailout from the IMF. As the silicon dust settles, we are left with the comforting knowledge that while the future may indeed be automated, it will at least be billed at a reasonable hourly rate. The high priests may have to trade in their turtlenecks for something a little more breathable, but the rest of us can look forward to a world where a graphics card is once again just something one uses to play video games in the dark.