The Cloud Is Physical and We Have Built It Too Close Together

Subsea cables and concentrated data hubs are now strategic targets. The answer is not panic but engineering: distributed, self-powered infrastructure designed to keep running when the links go down.

On the last day of 2025, a fibre-optic line on the floor of the Baltic Sea stopped carrying data between Helsinki and Tallinn. Finnish authorities seized a passing vessel and escorted it to port. It was a small cable carrying a modest load, but it sat on a much larger truth.

The digital economy, for all the talk of the cloud, rests on a startlingly physical foundation, and that foundation is increasingly contested. In the past two years, cables have been severed in the Red Sea, cloud facilities have been struck in the Gulf during the conflict with Iran, and undersea construction across whole regions has stalled because the waters are no longer considered safe.

Strip away the drama and the engineering reality is sobering. Around 95% of the world's international data travels through submarine cables, and the network is growing fast, with some 119 new cables due to be laid in 2026 alone. But those cables funnel through a handful of narrow straits and landing points, and the data centres they serve cluster in the same way.

In the UK, a large share of our cloud capacity sits in a single corridor around London and Slough. Concentration is wonderfully efficient. It is also a single point of failure dressed up as a strategy.

We build the switchgear and critical power systems that sit beneath data centres, and increasingly we build modular, self-contained data centre systems designed to operate at the edge, close to where information is created and used.

That vantage point makes one thing obvious. Resilience in this industry is not bought with a policy document. It is engineered, in the layout of power, the redundancy of supply, and the physical distribution of capacity.

Edge and modular infrastructure is, at its heart, a hedge against exactly the fragility the cable incidents expose. Distribution converts a single point of failure into many smaller ones. A compact, self-powered system placed near the institutions or operations that depend on it can keep essential workloads running when a distant hub goes dark or a long-haul link is cut.

Build in resilient on-site power and the ability to run independently of a stressed grid, and that site rides through disruption rather than collapsing with it.

None of this is glamorous. It is switchgear, backup, and the discipline of putting capability where it is actually needed. But it is the difference between an outage and a catastrophe.

I want to be equally honest about what edge does not do. It does not replace subsea cables or hyperscale campuses, and anyone who tells you it abolishes geopolitical risk is selling something. The backbone still matters, and most heavy computing will stay centralised because that is where it is most efficient. The case for distributed, hardened, self-powered capacity is narrower and more durable: it shrinks the blast radius.

A cut cable or a struck facility degrades a service rather than severing it. The right posture is a portfolio; the global backbone paired with dispersed infrastructure that can stand on its own when the backbone falters.

Governments have started to treat this seriously. A new US bill would impose sanctions on those who damage subsea cables, and European policymakers are belatedly investing in monitoring and redundancy. Those measures are useful, but they are largely about deterring and detecting harm.

The deeper task is to build a digital estate that can absorb harm and keep working. That means valuing the unglamorous layer: power, switchgear, redundancy, and the modular deployment that lets capacity be placed where it is safest and most useful rather than simply where it is cheapest.

We spent two decades optimising the cloud for scale and cost, and we built something extraordinary and brittle. The lesson of the last two years is that geography has come back, and that the most efficient design is rarely the most survivable one. The organisations and the countries that come through the next decade intact will be the ones that treated resilience as a feature to be engineered, not a cost to be trimmed.

The cloud is physical, so it is time we built it to survive contact with a harder world.