The AI pricing trap: how the hype is being used to overcharge the public sector
Every gold rush produces two things: genuinely useful tools, and pricing designed to separate excited buyers from their money. AI has produced an abundance of both, and the public sector — spending public money, under pressure to look modern — is a prime target.
I have sat on the buying side of the table as a council CEO, and I now sit on the building side. From both seats, the same pattern is obvious: a lot of AI pricing has nothing to do with the value delivered and everything to do with what the moment will bear. The technology is genuinely powerful, which makes it easy to attach a powerful-sounding price to work that, underneath, is not especially expensive to run.
The tells
A few patterns should make any public-sector buyer slow down. Per-seat licensing for software that scales at near-zero marginal cost — charging as if each user consumes a scarce resource when they do not. Per-conversation or per-token gouging, where the meter is designed to be opaque and to punish exactly the high-volume use that made you buy it. The “AI premium” — the same task you were already paying for, repriced with a multiple simply because “AI” is now in the name. And lock-in, where your data, your configuration and your history are held in a way that makes leaving expensive on purpose. None of these track the value you receive. They track how much leverage the vendor has over you.
Why the public sector is targeted
Councils and agencies are attractive marks for this, and not because anyone is naive. It is because public buyers are under real pressure to be seen adopting AI, budgets are annual and visible, procurement rewards the confident incumbent, and the cost of switching later is borne by someone who may not be in the room today. The result is contracts priced for the hype cycle, not for the decade of service that follows.
What honest pricing looks like
The alternative is not complicated, it is just less lucrative for the seller. Price to the value and the outcome, not to the buzzword. Be transparent about what actually drives cost, so a buyer can predict their bill. Offer it as a managed service with human-grade standards, where you are paying for a reliable outcome rather than renting access to a model you then have to operate yourself. And do not build the exit door shut. A vendor confident in the value they deliver does not need to trap you to keep you.
The question I would put to any public-sector buyer is the one I ask myself as a builder: does this price track the value, or the hype? If the honest answer is the hype, you are not buying AI. You are subsidising someone else’s gold rush with money that was meant for your community.