What you pay for not leaving¶

A plain-language companion to the draft working paper "Exit, Pursued by a Bill: Lock-in Rents and Portability in European Cloud and Software" (v0.12, September 2026). The paper carries the model, the proofs and the sources; this text carries the ideas.

The paper this condenses Read the full draft working paper (PDF)

The short version. A supplier that knows you cannot leave can charge you for it. The most it can charge is what leaving would cost you, so the price of staying is the cost of leaving. You can put a floor under that cost from two numbers you may already have: how much your price has risen, and how many customers like you have walked. The floor tells you about your own cost of leaving. It does not tell you whether the supplier is exploiting it, because prices also move with currency, energy, memory and bundled features; between 2022 and 2026 every price the paper could verify moved for one of those reasons. Most of the cost of leaving is rewiring what depends on the supplier's interfaces. None of the rules Europe has passed so far reaches that part.

The rent is the cost of leaving¶

Think of two suppliers. One runs your systems today. The other would run them for less. You stay as long as paying the difference, year after year, costs less than moving once. The incumbent knows this, so it prices the difference just under your moving cost. Under a posted price that everyone sees, the incumbent sets the difference at the moving cost of the customer it is just willing to lose. Under negotiated prices, which is what an enterprise agreement is, it sets the difference at each customer's own moving cost. Either way, what you pay above the alternative is your cost of leaving. Economists have said this since the 1980s. The paper's contribution is what comes next.

A floor you can compute from your own bills¶

If your supplier raised its price and you stayed, you revealed something: moving would have cost you more than the extra you now pay over the years you plan to keep the system. The argument needs nothing about the supplier. Take the price rise since the last time you could have left and multiply it by the number of years you look ahead, discounted: that is a floor under your cost of leaving. Add the switching rate, the share of customers who did leave; the floor then holds for everyone else. The customers who left give the mirror image: each of them revealed a ceiling, the extra it would have paid had it stayed.

Two complications come with the floor. It covers the cost of leaving plus whatever else makes staying worth it: features the alternative lacks, and plain inattention where nobody ever put the question on the table. In business software the list price is only where the negotiation starts, so the paper uses the vendors' announced increases as a public proxy and tells you to run the arithmetic on the price you actually pay.

What the floor says and what it cannot¶

The floor holds whatever moved the price. If your supplier raised its euro price to align with the dollar, or because its own costs rose, a buyer that stayed still revealed that leaving would have cost more than the gap. The cause changes a different question: whether the price is a rent, meaning money taken because you cannot leave, or a cost passed on. Between 2022 and 2026 that question has no answer in the prices. Microsoft's euro increase of 2023 was a currency alignment. Its dollar increases of 2022 and 2026, like Google's of 2023 and 2025, came bundled with new features. The European cloud providers raised prices with energy in 2023 and with memory in 2026, after memory contract prices doubled in a single quarter; Hetzner's mid-range plan went up 144%, which a supplier selling servers close to cost has to pass on. So the paper draws no conclusion about rent from these prices. For plain cloud computing it draws no floor at all, because both sides moved.

As an illustration only, take a buyer of Microsoft 365 E3 who stayed through the July 2026 increase. On a five-year view it revealed a cost of leaving of at least four to five months of what it spends on the suite each year, or about eleven months if the currency step counts. Three other numbers land in the same range: a 2022 survey in which CIOs put their own cost of leaving at 15% of their IT budget, a supplier-side method that gives about a year, and the migration estimates the paper discusses.

Where the evidence of rent actually is¶

The UK's competition authority found the two largest cloud providers earning returns above their cost of capital while fewer than 1% of customers switch in a year. That is the pair the model predicts: high returns and no exit. A 2022 survey found that nine in ten French companies had faced increases of 3% to 6% a year at renewal over the years before, a period of low inflation and stable currency. That is what a supplier pricing on its customers' cost of leaving looks like. Neither measures the rent a given buyer pays. Doing that needs the prices a buyer paid, the alternative's price at the same dates, and the supplier's own costs, which is a data set nobody has published.

Five parts of an exit cost, and which rules reach them¶

Leaving costs five things: moving the data, rewiring everything that depends on the supplier's interfaces, the downtime of the switch, retraining the people, and the calendar, since a contract you cut mid-term saves nothing until it ends. A rule lowers your price only through the part it touches.

The Data Act removes the fees for moving data. Those fees are under 1% of most customers' annual cloud spend, which is why the three hyperscalers could waive them for departing customers in 2024 while the switching rate stayed where it was. The commitments the UK accepted in March 2026 act on the same part and on the calendar. Nothing in force reaches the rewiring, which is where most of the cost sits. Only an interface that a second supplier serves can reach it.

Why nobody fixes it alone¶

Suppose you, alone, made your systems easier to move. Under a posted price your supplier's price does not change, because you are one customer among many; you gain only if you leave. So individual customers under-invest in their own portability, while the things that have moved prices have all been collective: statutory switching rules, exit tests imposed on a whole sector, buyers pooling their demand behind a second supplier. Portability is a public good among the customers of one supplier.

What to do with this¶

If you buy: compute the floor from your own renewal history before the next renewal, put a number on each of the five parts, treat the rewiring part as the one that decides. If you regulate: an instrument that lowers the fees for moving data will not move the switching rate, while one that gets a second supplier onto the same interface will. If you sell as the alternative: the price you can win a customer at is the incumbent's price minus that customer's cost of leaving, which is mostly the interfaces, which is where the work is.