Shifting to sovereign digital supply: a “how to” approach
Six questions European companies and their investors ask before moving their digital supply to European providers. What do we put in scope? How far does AI belong in it? How do we build the business case? How do we structure the contracts? Do we rent or build? How do we form a consortium? A seventh, for investors: what should we require of the companies we fund?
It asks a board to decide three things: which layer to move first and on what contract terms; whether to anchor a consortium; what to require of the companies it funds.
Written for the EuroStack Foundation, the industry initiative for a European digital stack. The lab’s parent company, Abilian, is a French open-source software vendor. The brief recommends no supplier.
Software you cannot leave is where prices keep rising, because the ceiling on what you pay is your own exit cost
Moving the servers to Europe while keeping the same provider’s software changes neither the exit cost nor the price
A data centre in Europe running the same provider’s proprietary software at every level leaves your switching cost exactly where it was, together with the price you can be charged.
Residency answers one legitimate question: where does the data sit? It does not answer the one that sets the bill: who controls the software your business runs on?
The seven questions and their answers
- What do we put in scope? Move the control plane first; take the commodity layers as a quick win in parallel; build nothing new on the services you cannot leave.
- How far does AI belong in it? Put it in scope at the layer where projects actually fail; budget it as a running cost.
- How do we build the business case? Build it on the bill you will receive over the years your exit cost keeps you in place.
- How do we structure the contracts? Screen suppliers on who controls them; commit volume, never architecture.
- Do we rent or build? Rent from European providers by default; build a data centre of your own only where you can keep it busy.
- How do we put together a consortium? Be the anchor yourself: bring your holdings in as the first contracted members and lock the governance before the first euro goes in.
- What should we require of the companies we fund? Write the covenants into the next shareholders’ agreement; put a yearly exit-cost disclosure on every board agenda.
What is decided today and what waits on two studies the brief proposes
- Commission the costing of the control-plane move on your own systems; start the commodity sourcing exercise in parallel.
- Adopt the contract terms of answer 4 for every renewal from now on.
- Put an internal price on bundled services; run the break-even of answer 3 on your own invoice, workload by workload.
- Write the seven covenants into the next shareholders’ agreement; put the yearly exit-cost disclosure on every board agenda.
- Decide in principle whether to anchor a consortium; if so, name the holdings that join first and the governance lock the vehicle carries.
- Until the two studies are done, commit demand through your holdings’ membership and equity into the club’s vehicle; capital into a facility waits.
Sort every service on two axes and attack where leaving is expensive and the code is shallow
Thirty-five managed services from Amazon Web Services’ catalogue were scored on these two axes. The numerals are the order of attack.
Starting where leaving is already cheap removes little: a service you can walk away from is not one you are being overcharged for.
The control plane first, the cheap layer in parallel, a rule for everything else
Why the exposure is large. Who can log in and to what, where the keys are kept, the monitoring and the deployment tooling are all configured in the provider’s own formats. Leave in a hurry and you rebuild it under pressure with the business down. Your supplier prices every renewal knowing that.
Why the fix is bounded. A planned replacement rewrites no business logic. You rebuild the record of users and rights, the custody of the keys and the tooling once, then re-point your applications. We cannot size it for you; cost it first.
Buy computing capacity, storage and databases from European providers as ordinary procurement while the control-plane work goes on.
It funds the programme without being it: these are services you could already leave.
- The proprietary tier (managed AI, pay-per-call functions) is woven into your code and no European provider sells an integrated substitute. Do not plan to replace it. Stop adding to it and confine what you have.
- The office and collaboration suite has proven European substitutes and a known sequence: back-end services, then client applications, the workstation last.
- Line-of-business systems, ERP included, stay out of joint purchasing and get the exit clauses of answer 4.
Fully, aimed at the layer where AI projects fail, which Europe’s public programmes do not fund
Build it on the bill you cannot walk away from; a three-year window is biased against any investment
Separate two kinds of move. Where a European provider already sells the same service cheaper, the case is a price comparison with the migration effort added. Where the move means rebuilding something, it is an investment: a migration buys a credible exit threat first; cost reduction, if it arrives, arrives late.
- Escalation. Use your own rate, from your last three years of renewal quotes, across the whole horizon.
- The margin you concede. A supplier can charge above the competitive price only up to what it would cost you to leave. The brief states the bound; the euro figure needs your own numbers.
- Insurance. A working alternative limits the loss when a supplier fails or cuts you off; it needs no political forecast.
Do not forecast savings. Compute the price at which the move breaks even, then put it next to your invoice
What would the incumbent have to charge you, per seat, per hour of computing or per million tokens, for the move to pay over three, five and ten years? You already hold the second number. The inputs you assume are your escalation rate and your own migration effort.
| Estate | 3 years | 5 years | 10 years |
|---|---|---|---|
| 250 seats | 401 | 269 | 170 |
| 2,500 seats | 293 | 229 | 182 |
| 25,000 seats | 207 | 160 | 126 |
Above the figure the move is cheaper over that horizon; below it the incumbent is. Migration cost is mostly fixed, so small estates face the hardest arithmetic: an argument for pooling. The 250-seat row omits three components priced only from 300 seats, so that row understates the break-even. Undiscounted, with the incumbent’s price held flat. Discounting raises the figures, since the migration pays at the start and the incumbent across the horizon: a 4% rate adds 5% to 11% to the ten-year figures, an 8% rate 10% to 22%, the smallest estate most.
The list prices are February 2024 figures for five open-source products (Bern University of Applied Sciences, for the Swiss Federal Chancellery). The EUR 1,200 day rate, the 144 to 953 person-days and the lost user-day per user are illustrative: replace them first.
Screen suppliers on control: five facts separate them where the usual requirements do not
- Data residency
- Local staff and local incorporation
- A local headquarters, in the newer requirement texts
- Generic certifications
A foreign-controlled supplier meets every one of them at the same cost as anyone else. Require them for their own sake; give them no award weight.
- Who ultimately owns the supplier
- Which law governs the contract
- Who holds the encryption keys
- What the parent’s home law lets a foreign authority compel
- What the contract says happens to your service under such an order
Every extra clause weakens the ones that matter: a supplier caught cheating loses the same contract however many clauses it broke.
Award on price at the commodity layer, where you can leave. For a supplier you will rely on and have verified, a decent margin gives it something to lose if caught cheating; award such a supplier on lowest price and you pay for the enforcement yourself.
Commit volume, never architecture: a floor the supplier can plan on, while you keep the right to move
- Exit assistance is priced and scoped in the contract.
- Data export is free, tested and retested on a schedule.
- Each bidder names who could take over from it in three years; a good answer names competitors.
- The tender is written for the service you need and weighted on the supplier’s willingness to build to it.
Rent from European providers by default. Build your own data centre only where you can keep it busy
- No usable supplier can be verified to the standard a workload needs, so custody of the hardware is the only real protection.
- The workload’s physical profile favours ownership: high utilisation, data you cannot delegate, hard latency limits, cheap energy and heat of your own.
- A consortium can contract enough pooled demand to keep the facility busy.
Be the anchor yourself: bring your holdings in as the first contracted members
Five rules for the anchor, before the first euro goes in
- Equity goes into the club’s own vehicle, against contracted demand; everywhere else, use demand. Buying cloud and software suppliers tends to buy an acquisition target. In the lab’s model, procurement and portability rules remove about as much exposure as buying them would.
- If the consortium takes a stake in a supplier anyway, require a governance lock: foundation ownership, super-voting shares, a golden share or state control. A purely financial investor, for whom a sale is an exit, is not bound.
- Give the related-party conflict its own rules: arm’s-length pricing and approval by members without a stake, since the anchor buys from a facility it finances and powers.
- Apply the supplier screen of answer 4 to the consortium’s own procurement, including members’ own bids.
- Contract your holdings’ commitments for longer than the formation phase; where the club finances a facility, they are long by design. An anchor that leaves early takes the club with it.
Seven covenants for the next investment, one board item for the companies you already hold
The cheapest instrument an investor holds works through the companies it funds, because a company that has not yet built its systems has nothing to migrate. Cloud credits, the free capacity hyperscalers give start-ups, are how the dependence usually arrives: book them as a loan repaid in switching cost.
- The login system and the encryption keys under the company’s own control
- Open interfaces named in the plan
- A stop-adding rule for the proprietary tier, with exceptions justified on the record
- Portable formats for what its AI systems accumulate
Section 7 of the brief gives the seven in full. What comes back: a company that could walk away from its suppliers concedes them less margin on every contract, which shows up in earnings. Companies built on shared interfaces can later buy together.
- A yearly disclosure to the board of what leaving each material supplier would cost
- Free and tested data export in every material contract
- An internal price on anything that arrives bundled
Three things this brief cannot give you, plus the costs a business case must not hide
We cannot size the control-plane rebuild for you. Its cost grows with the number of systems that authenticate against your provider; the size of your codebase does not enter. It is the first thing to cost on your own systems.
We do not know the committed load at which owning a facility beats renting, once pooled demand and the exit cost of the chosen design are counted. No build-or-rent model we have found carries those two terms. The anchor’s own load and energy data would turn it into a figure.
A small European supplier can only borrow against demand that is committed to it. We cannot yet write the contract that commits that demand without handing the supplier the lock-in you are leaving. We can price exit rights and premia; the contract template is the second study the brief proposes.
Who controls the software your business runs on? What would leaving it cost?
Every answer in this deck follows from those two questions. Three next steps: cost the control-plane move on your own systems, run the break-even on your own invoice, name the holdings that would join a consortium first.
The brief, “Shifting to sovereign digital supply: a ‘how to’ approach”, carries the checklists, the break-even device and the sources: eurostack.eu
Additional resources/papers: econ.lab.abilian.com