A story as old as writing
In the myths of many peoples, fire is guarded by someone powerful and jealous who will not share it. For centuries writing was the privilege of a few scribes, until the alphabet put it within reach of many more people. The history of innovation is also the history of who controls access to it.
With artificial intelligence the question returns on a global scale. Three companies, Amazon, Microsoft and Google, run almost two thirds of the planet's cloud infrastructure.1 The most advanced chips come from a supply chain concentrated in a few Asian countries, the most powerful models from a small number of labs. It is not a conspiracy: it is how markets that need enormous investment work. But it has consequences for anyone building on these foundations.
How to recognise dependence
In companies I often see the same path. A tool is chosen because it is the most convenient, then data, processes and skills take its shape. After a few years switching costs so much that nobody tries any more, and every renewal goes through without discussion. Four questions are enough to see where you stand:
- Can the data be exported in a format other systems can read?
- What happens if the provider changes prices, terms or model?
- Would the work done so far survive a change of platform?
- Inside the organisation, does anyone know how it really works?
Why AI raises the risk
With traditional software, dependence was mostly about data and processes. Artificial intelligence adds another layer: behaviour. The instructions, examples and checks a company writes for a model are tuned to that model. When the provider updates or retires it, the answers change, sometimes subtly, and something that used to work stops working.
Then there is the knowledge that flows through these systems. If every document, meeting and decision is summarised and stored by a single provider, the organisation's memory lives in someone else's house. That is no reason to give up. It is a reason to decide clearly what stays under your own control.
Sometimes the best way to turn an invention into a real innovation is not to own it, but to give it away.
From Un perché alla volta (translated)
An exit plan, before going in
Dependence is not destiny. It is avoided by designing the way out from the start. In practice:
- Write into the contract how data is recovered, in which format and how quickly.
- Keep a tested alternative for critical services, even if you use a single provider every day.
- Separate knowledge from the model. Instructions, examples and quality tests must be able to move from one system to another.
- Consider open models. There are open-weight models that can be downloaded and run on your own servers: they are not always the most powerful, but for many tasks they are enough and they keep data in-house.
- Invest in in-house skills, because the deepest dependence is not knowing how the things you use actually work.
Rules help too. With the Data Act, Europe requires cloud providers to make switching to a competitor easier and, from January 2027, to stop charging for it.2
The keys can be kept
History offers encouraging precedents. In 1993 CERN made the web software available to everyone, royalty-free, and that is partly why the web came to belong to everyone.3 The three-point seat belt and the polio vaccine followed similar paths: the people who invented them chose not to lock them away.
Not every technology will become a common good, and it does not need to. It is enough that the people who use it keep the ability to choose. A technology truly becomes ours when we can understand it, change it and, if need be, leave it.