Governments are buying AI at introductory prices: stress-test business cases at the unsubsidised price

Something has been nagging at me about the way governments are buying AI, and two reports published in the same week helped me put a finger on it.

Bain says the compute being built needs about US$6 trillion in annual revenue by 2031, and roughly US$4.2 trillion of that has to come from markets and products that do not exist yet. Read that as a buyer, not an investor. Somebody has to pay for a very large bill, and the sellers know it.

Then McKinsey's work on US government buyers, as Semafor reports it. The price of a token is collapsing, yet the total bill is rising, because agents burn far more compute than a chatbot ever did. A McKinsey partner says that for many governments usage has been close to free, and very soon it won't be. The detail behind that, in a summary I could only find second-hand, is that the introductory federal deals priced frontier models at a dollar or less per agency, and they run out across 2026 and 2027.

Put those side by side and I think the interesting point is not that prices will go up. McKinsey says the opposite about unit prices. It is that pilots were priced as an introduction, sized for curiosity and approved on that basis, while the seller needs revenue. Those two facts meet on a renewal date.

So my instinct is that the next test for any public-sector AI program is boring and financial. Not whether it works, since plenty of it does, but what it costs at agent-scale volume once the introductory price has gone, and who has authority to stop it if the number is wrong.

I may be too gloomy. Volumes may settle and unit prices may keep falling faster than usage grows. But a business case that cannot survive the unsubsidised price is a bet, not a case.

If the subsidy ended tomorrow, which of your AI programs would you still fund?

Sources: Semafor, "AI costs are rising for organizations, reports warn" (2 October 2026), https://lnkd.in/gma4cUqA . Bain & Company, 2026 Global Technology Report release (29 September 2026), https://lnkd.in/gBVMixWg . Pivot News summary of McKinsey, https://lnkd.in/gUt7cb2w

#AIStrategy #PublicSector #Procurement