For almost eight hours on Tuesday, a power event in a Google data-centre region in London left card payments failing at Revolut and online services faltering at Lloyds and HSBC. Royal Mail and DPD reported problems, and so did NHS England, which saw disruption to some of its digital services. Nobody had been hacked. A building had lost stable electricity, and a large part of the British economy noticed.

The most credible lesson for health is structural. The NHS is being pushed, with good reason, towards a digital model. The 10 Year Health Plan places the NHS App at the centre of patient access, promises a single patient record and relies on shared data infrastructure to find productivity that wards and clinics cannot deliver alone. Each of those commitments adds load to hosted systems that sit with a handful of global providers. A trust that once ran a server room now rents capacity in a region it does not control, alongside banks, retailers and parcel firms it has never met.

That arrangement is efficient, and efficiency is what a service with a record waiting list and a tight capital settlement needs. It also concentrates risk in a way the NHS has not fully priced. Financial services have already faced this question. The UK now has a regime for designating critical third parties, which brings the largest cloud suppliers under direct regulatory scrutiny because a failure at one of them can reach many firms at once. The Financial Conduct Authority said this week it was examining the October disruption with the companies affected. Health has no equivalent designation for the suppliers that sit beneath its services. Oversight of NHS organisations stops at the contract line, and the contract line is where the dependency begins.

The reported effects on the NHS appear to have been limited. That should not be mistaken for resilience. Clinical teams are practised at downtime procedures, and the ransomware attack on a pathology supplier in 2024 taught many trusts how quickly an upstream failure becomes a cancelled operation or a diverted ambulance. But a single-region outage of under eight hours is a modest test. A longer failure, or one that struck during a winter surge, would land on services already running close to capacity, and recovery would compete with the backlog for the same staff.

The practical implications fall unevenly. NHS leaders need a clear map of which of their systems depend on which provider and which region, including the products bought from software suppliers who themselves rent infrastructure. Many boards could not produce that map today. They should also ask whether a patient-facing service such as appointment booking or prescription ordering has a tested fallback, and who answers the telephone when it fails. Public trust in the digital shift is fragile. Each visible outage gives sceptics evidence that the App is an unreliable front door, and adoption is the premise on which the plan's savings rest.

Policymakers face a harder decision. Redundancy across regions or providers costs money and technical effort, and the same regulators that urge banks to plan for failure have acknowledged how expensive that is. In a health system where capital is rationed and digital budgets compete with estates repairs, resilience is easy to defer. Central government could set minimum continuity expectations in national procurement and extend critical supplier scrutiny to the infrastructure beneath the Federated Data Platform and other national services. Without that, each trust will negotiate its own terms with providers holding far more leverage than any single buyer.

Health-tech and life sciences firms should expect these questions to move into tenders. Suppliers able to show multi-region architecture, tested recovery times and honest incident reporting will hold an advantage over those that cannot. Google has yet to publish its full technical analysis, and the detail of what failed may change the picture. The direction is already clear.

The NHS cannot take the productivity gains of concentrated cloud infrastructure and leave the resilience bill unpaid. Someone will meet that cost, whether in advance through planning and procurement or afterwards through cancelled appointments and lost confidence. Doing so in advance is cheaper, and Tuesday offered a low-cost warning.

‍