After a cyberattack, everything hinges on your ability to rebuild fast

In its late-September 2026 letter on digital security and artificial intelligence, the Cigref describes a threat that is changing in nature. According to the Cigref, AI lowers the cost of an attack, raises its cadence and widens its targets: what once took rare expertise becomes automatable. The association sees in this a balance to be regained, between the acceleration of the offence and the means of defence.

The consequence fits in one sentence. According to the Cigref, preventing the intrusion remains necessary but is no longer enough as a strategy: you also have to be able to rebuild the information system very fast after an incident. For an IT leader in a large organisation or public body, that is a concrete shift in priority, one that touches architecture before tooling.

Rebuilding fast cannot be decreed, and it cannot be bought as a number. We do not commit to a recovery time, nor a turnkey contractual disaster-recovery plan: that would be a commitment that nothing, in practice, lets you honour identically across every situation. What we build is the set of conditions that make a fast rebuild possible rather than theoretical.

Those conditions come down to three simple questions every IT department should put to its host before an incident.

Can I leave without asking its permission? Can my software be taken over elsewhere, as is? Do my backups and my data stay under my control and in Europe?

When all three answers are yes, the speed of reconstruction stops depending on a provider's goodwill.

Reversibility, with us, is not a clause buried at the bottom of a contract. It is a documented exit procedure: you recover your data through exports and database dumps, you take back control of your domain names, and you redeploy your applications elsewhere, on your own hardware or with another host, without Bunker having to give its approval. Taking your infrastructure back in-house is planned from the outset, not negotiated under pressure.

That mechanism matters precisely on the day of an incident. If recovery depends on the intervention, the authorisation or the goodwill of a third party, every hour of waiting is an hour of business lost. An exit you can trigger yourself is an exit you can trigger fast.

You do not rebuild fast on software you can neither inspect nor redeploy. Our stack is open source end to end and rests on open standards: Kubernetes for orchestration, S3 for object storage, PostgreSQL for databases, OpenID Connect for identity. Nothing proprietary holds you back, and every layer can be taken over as is, with documentation to match.

That is the difference between a backup you own and a backup you can actually restore. An export in a closed format, readable only by the tool that produced it, is no guarantee of recovery. Our formats are open, and so are our open-source building blocks, which makes reconstruction independent of our survival as a provider.

The last condition is physical: where your data lives, and who operates it. Bunker's infrastructure is hosted in Europe, in French datacenters run by a French company. Databases rely on CloudNativePG (PostgreSQL) with replication, backups are kept on an internal sovereign MinIO object store, single sign-on runs on self-hosted FerrisKey, and the only outbound email relay goes through IONOS, a German company based in the European Union. Subprocessors are listed in the contract, under Article 28 of the GDPR.

Supervision combines automated monitoring and an SRE on-call rota, and backups are encrypted and replicated. For reconstruction, the stake is direct: backups you control, on an infrastructure whose location and composition you know, restore without depending on a distant region or an unreachable contact. This is the heart of our approach to digital resilience, and of our reading of dependence on non-European providers.

AI does not weigh only on the threat. According to the Cigref, the spread of AI agents and usage-based billing, often per token, make spending hard to anticipate for IT departments: the cost follows a consumption that is poorly controlled. Budget resilience, too, comes down to predictability.

Our model is deliberately the opposite. Each service is billed at a monthly flat fee, publicly displayed, and volume (logs, metrics, traffic) does not change the price. Our sovereign AI instance, for example, is offered at a fixed monthly rate of €160 excl. VAT. We quote no competitor's price: we simply observe that a fixed cost can be budgeted ahead, where a meter has to be endured.