McKinsey’s clients have started refusing to pay by the hour, and the reason they give is blunt. With AI in the room, billing by time spent has become a kind of theater. If the slide deck that took a junior associate three days now takes ninety minutes, what exactly is the client paying for when the invoice still reads forty hours? The Financial Times has covered the pushback against the billable hour as a story about consulting economics, and it is, but the more useful reading is wider than one industry. The hour was never the thing anyone was buying. It was a proxy, and AI has just made every proxy in the economy visible at once.
The billable hour stood in for expertise, for judgment earned over years, and for the risk a firm absorbs when it puts its name on your decision. Time was simply the unit everyone agreed to stop questioning, because it was easy to count and hard to argue with. You could not measure wisdom, so you measured the hours that wisdom sat in a chair, and both sides of the invoice found the arrangement tolerable. The fiction held for decades because nobody had a reason to look underneath it, and looking underneath it would have required proposing something harder to administer.
AI gave everyone a reason. It did not break the pricing model. It revealed that the model was already a story people had been telling each other, and that the story and the value had quietly come apart somewhere along the way. That is the pattern worth carrying into the rest of the week’s news, because the consulting invoice is only the first place it surfaced.
Anthropic and OpenAI Are Repricing Too
Anthropic’s filing to go public gets framed, reasonably enough, as a market event. CNBC calls it the first big test of AI boom valuations, and that framing is right as far as it goes. Strip away the drama and the question underneath is precisely the one McKinsey’s clients are asking: are we pricing this by what it earns, or by what we hope it becomes? A listing is the moment a private story meets a public ledger. The market gets to decide whether the number people have been repeating to each other in funding rounds survives contact with people who want their money back on a schedule. It is the billable-hour conversation, scaled to eleven figures and conducted in public.
Set that next to what OpenAI is doing with the product that made it famous. The Financial Times reports the biggest ChatGPT overhaul since launch, and the friction between those two stories is the interesting part. One company is asking the public to price its future at a very large number. The other is rebuilding the thing that established the category in the first place. You do not rebuild from the studs because the product is working perfectly. You rebuild because the ground under it moved, and the version that defined a category three years ago is now one option among several that do roughly the same job. The product has not gotten worse. The thing it is priced against has gotten cheaper, which is a different problem and a harder one, because no amount of internal excellence fixes it.
Then there is Apple, walking into its developer conference with the quiet pressure of a company that built its fortune on polish and is now being asked to ship intelligence. CNBC framed it as Tim Cook’s AI legacy at stake at his final developer conference, which puts a personal frame on a structural problem. For decades Apple priced itself on the part you could feel: the weight of the hardware, the smoothness of the glass, and the sense that someone cared about details nobody would ever name. That was real value, and it commanded a real premium for a very long time. But you cannot sand and polish your way to an assistant that actually understands you. The market has started pricing a different thing, and a company that spent forty years perfecting the old thing has to decide what it is actually selling now.
The Amplifier Has a Power Cord
Underneath all of it sits the least glamorous signal of the week, which is the growing fight over data centers. The compute that makes every one of these stories possible has to live somewhere and draw power from somewhere, and increasingly the somewhere is pushing back. Set aside the question of who is funding the resistance and the physical fact stays simple. The intelligence everyone is busy repricing runs on electricity, land, and water that someone in a town meeting has opinions about, and those opinions arrive on a timeline that no product roadmap controls.
This is the part the valuations keep forgetting. Computing is a genuine amplifier. It makes individual people capable of work that used to require institutions, and that is not hype; it is the actual engine of the last twenty years of progress. An amplifier is not free, though. It has a power cord, the cord runs back to a grid, and the grid runs back to a county that gets a vote on whether the next substation gets built. The cheapest input in the model on the slide turns out to be the one with a body, a location, and neighbors.
Hold that against the repricing stories and the shape of the problem gets clearer. Every one of them is about a price that drifted away from what it was supposed to measure. The consulting hour drifted from expertise. The valuation drifted from earnings. The premium on polish drifted from what a customer now wants a device to do. And the assumption that compute is an abstract, elastic input drifted from the physical reality of where it plugs in. Cheap to produce, expensive to value correctly: the two move in opposite directions, and AI is the force pulling them apart faster than anyone’s pricing committee can meet.
What Fast Feedback Does to a Price
When you build fast feedback into any system, the great gift is that you see the effect of your choices immediately, with no lag and no place to hide a bad assumption. That is exactly what is happening to the whole economy right now, except that nobody asked for the feedback loop and nobody can turn it off. The cost of making things has collapsed, and the collapse is showing us, in real time, every place where we had been pricing the proxy instead of the thing itself.
The consultants felt it first because their proxy was the most naked: time, billed by the hour, in a job that was never really about time. Their clients could do the arithmetic without any special access, which is why the pushback came from the buyer’s side rather than from some internal reckoning. Everyone else’s proxy is buried deeper and will take longer to surface, which is a reprieve rather than an exemption.
Because the hour was always a proxy. So was the valuation, so was the polish, and so was the assumption that the power would simply be there when the racks arrived. We are about to find out, all at once and in public, which of our prices were measuring value and which were measuring a habit. The bill for the difference is already in the mail.
Leave a Reply