A Different Sport

Yesterday, The Decoder reported on a striking divide in how US businesses are spending on artificial intelligence.¹ The headline from fintech company Ramp’s latest AI Index was modest: AI spending rose 1.65 per cent in October, reaching a new peak. What lay beneath it was not: the top one per cent of firms now spend a median $10,200 per employee per month on AI. The typical firm spends just $16.32 – roughly 625 times less. That same top one per cent accounts for 80 per cent of the revenue going to OpenAI and Anthropic combined, a concentration that Ramp’s lead economist, Ara Kharazian, calls »unseen in any other software category we track«.²

»The track remains the same, but the cars can change«

The artist Andy Warhol once argued that money’s limits become visible at the vending machine. »America started the tradition where the richest consumers buy essentially the same things as the poorest,« he wrote. »A Coke is a Coke, and no amount of money can get you a better Coke than the one the bum on the corner is drinking.«³ Some things simply cannot be improved by spending more. So when one firm spends 625 times more per worker on AI than another, is it buying a better Coke, or something else entirely?

In 1976, economist Fred Hirsch described two kinds of things that people spend money on.⁴ Some have a natural limit: once you have a Coke, buying a more expensive one doesn’t get you a better drink. Others are valuable precisely because few people can have them: a house in the one neighbourhood everyone wants, or an Olympic medal. A six-hundred-fold gap in spending should produce either a great deal of waste or a decisive winner. Ramp’s figures show neither. Perhaps the difference lies not in AI itself, but in what companies use it to do.

Picture two drivers: one racing in Formula 1, the other driving to the shops. Give the racer a car 600 times »better« and there’s no race left. Is it still even the same sport? Give the second driver the same car and nothing changes once it arrives. In 1981, economists Edward Lazear and Sherwin Rosen described this mechanism: where rewards depend on relative performance, a small capability edge can determine the outcome. Where a task requires only clearing a threshold, capability beyond it may be worth nothing.⁵

This is close to what the firms Ramp analysed already do. Engineering is the closest thing in commercial life to a race: better performance can mean a better result, so there is value in reaching for a higher capability ceiling. Customer support is different. A ticket is either resolved or not; once a model is good enough, a more expensive one adds nothing the customer can perceive. The top one per cent, then, may not be buying a uniformly better can of everything. They may be paying for a higher capability ceiling where performance matters, and a sufficient capability floor everywhere else. The median firm and the support desk of the richest firm may both be operating at that floor, even though only one of them is rich.

This reframes equality itself. In 1979, economist and philosopher Amartya Sen argued that equal resources don’t produce equal capabilities – what matters is not how rich you are, but what you can do with those riches.⁶ Two engineers with »access to AI« could differ radically in what that access lets them do in an afternoon. Spending measures the input, not the capability that comes with it.

The same shape reappears at the size of the planet: only a small fraction of humanity has ever touched a frontier AI product, and today’s leaders are no permanent caste. SpaceXAI’s adoption rose to 6.3 per cent in October, on an »immense compute advantage« of its own. The track remains the same, but the cars can change.

The theoretical biologist Robert Rosen offers one last distinction.⁷ An anticipatory system acts on a model of what may happen; a reactive system responds to what is happening now. The deeper divide between firms may be less about how much they spend than what they are preparing for. Some are reorganising around capabilities they expect AI to acquire. Others are using it to handle the work that’s already in front of them. The difference isn’t just in their budgets, but in how far ahead they’re looking.

There’s no cause for alarm. A firm spending heavily on a race it’s actually running isn’t wasting money. But an AI budget, seen from the outside, tells us little about what the firm is trying to achieve. The same technology can be used to win the race or just get someone to the shops. The difference is not how much you spend, but what you’re trying to win.

References