National Cybersecurity
Association of Ukraine

The data centre as critical infrastructure: what 2026 showed.

Analytical note ·

A strike, the power behind it, and the trust layer above it. The three ways digital infrastructure stops are usually discussed by different people, although to an organisation they are one failure. This note puts them together and sets out what follows for preparation.

In short

  • A data centre has stopped being a room with servers. To a state it is an object of the same order as a substation.
  • It can be stopped three different ways: by a physical strike, by cutting its power, and by compromising trust in the systems that run it.
  • Those three scenarios are usually owned by three different teams — physical security, energy, and information security. To the organisation they are the same outage.
  • Artificial intelligence has not created a new class of attack. It has made preparing the old ones cheaper.

The first way: the strike

Through 2026 Ukrainian data centres were repeatedly targeted. Ukrainian media reported both direct hits and stoppages caused by damage to power supply nearby.

The consequence reaches far beyond the building. Whatever is hosted there stops with it: internal company systems, state registries, communications, backups.

This is why framing such events as slow internet moves the focus to the wrong place. Slow internet is a symptom, not the event.

The second way: power

A data centre without electricity is a warehouse of equipment. A generator holds for hours, fuel for days, and grid repair can take weeks.

So the resilience of compute is a question of energy rather than only of IT. An organisation that plans redundancy for its servers but not for its power has a plan for half the scenario.

A rarely discussed risk sits here too: concentration. The fewer sites serve a country or a sector, the cheaper one successful outage becomes.

The third way: trust

The third way needs neither a missile nor an outage. It is enough for a system to trust the wrong party.

Modern infrastructure is operated by processes rather than by people: keys, tokens, service accounts, deployment automation. Each is an identity the system trusts by default, and each can be compromised without breaching any perimeter.

NAK has published a detailed examination of one such case as a separate research report.

What artificial intelligence changed

The common claim that AI has created a new class of attack on industrial and utility systems is overstated. What it genuinely changed is the cost of preparation.

Reconnaissance of a target, drafting the wording for social engineering, translation, analysis of public information about a site: all of that used to require a skilled person's time. It now takes hours rather than weeks.

For defenders that means the assumption of being too small to be worth attacking no longer holds. The cost of an attempt has fallen, and with it the threshold at which an attempt becomes worthwhile.

What follows

  • Plan for the loss of a site, not the loss of a server. The question is not whether a backup exists but how long it takes to run somewhere else.
  • Count power separately from equipment: how many hours the generator holds, where the fuel is, who delivers it during an air raid.
  • Keep an inventory of machine identities the way you keep one of people: who created the key, what it can do, when it expires.
  • Test restoration rather than the existence of backups. A backup that has not been restored is not a backup.
  • Agree a fallback communication channel in advance and in writing, for the case where the main one is unavailable.

The limits of this note

The note rests on public reporting by Ukrainian and international media and on NAK's own research. It does not attribute any particular attack, does not assess the conduct of individual organisations, and is not an incident-response procedure.

The Binance incident report