A power failure at one of Google Cloud's London data-centre zones disrupted banking apps, card payments and public services across the UK on Tuesday 20 October. No construction project is known to have stopped. The incident still carries a lesson the industry should take seriously, because it is moving more of its information, automation and artificial intelligence onto the same kind of infrastructure.
Google's official service-health record attributes the outage to a power event in the europe-west2-b zone of its London europe-west2 region, with 38 cloud products affected across compute, storage, databases and networking. The disruption began at 07:08 UK time, and Google said most services had returned to normal by around 13:10, roughly six hours later.
The most visible casualties were in banking. Lloyds, Halifax, HSBC and Revolut all reported disruption, and Revolut customers faced problems with card payments, ATM withdrawals and transfers. Court live streams, e-bike rental services and budgeting apps were also reported to have been affected.
No major UK contractor or construction software provider has publicly confirmed a direct project impact from this incident. The outage is therefore not evidence that construction work stopped on 20 October. It is a clear demonstration of a failure mode to which an increasingly automated industry is becoming more exposed.
Common Data Environments now manage approved information. Algorithms monitor programme and cost, cameras feed computer-vision systems, and sensors report asset condition. AI is beginning to read documents, classify risk, prepare reports and trigger automated workflows. Fewer decisions, checks and administrative tasks depend entirely on people. That brings real productivity gains, and it raises a practical question: what happens when the systems remain secure but the infrastructure underneath them becomes unavailable?
Construction activity in Soho, London, as the industry becomes increasingly dependent on cloud-based information, automation and digital systems. Photo: London Construction Magazine.
Key Takeaway: Construction's next digital-resilience problem is not only protecting information from attack. As cloud systems and AI take over operational tasks, firms need to know which functions can continue when the cloud, identity service, API or internet connection disappears.
The Cloud Is Physical Infrastructure
The word cloud obscures what the industry is relying on: data centres, electrical supplies, switchgear, cooling systems, fibre networks and servers. An AI agent working on a construction programme may appear entirely digital, but underneath it sits physical plant that can lose power, cooling or connectivity.
London has seen this before. In July 2022, during the heatwave, a simultaneous failure of multiple redundant cooling systems forced Google to power down part of its europe-west2-a zone. In August 2024, a substation switchgear failure cut both primary and backup power feeds to a Google network site in London, causing around 40 minutes of disruption. Google's facility partner then investigated why stored or generated on-site emergency power did not carry the load.
That makes three significant London incidents in about four years, each rooted in power or cooling. Those are the same engineering systems construction designs, installs and commissions. LCM's analysis of the £100bn UK data-centre development pipeline has already shown how grid connections, substations, cooling, long-lead electrical equipment and commissioning are becoming critical constraints on Britain's AI and cloud expansion. The industry is building the infrastructure its own digital transformation depends on.
This week's incident also shows how failures spread. The trigger was in one zone, yet Google reported elevated errors and intermittent network issues across the wider region. It advised affected customers to move workloads to another zone or region "if feasible". That qualification matters. Failover only works when systems have been designed, replicated and tested to fail over before the incident occurs.
The Single Source of Truth Can Become a Single Point of Failure
Construction has spent years eliminating fragmented project information, and rightly so. Uncontrolled drawing folders, spreadsheets, email attachments and locally saved models create serious risks of their own. Governed CDEs have improved traceability and coordination. The next question is availability. If the CDE is the project's only trusted source of current information, what is the controlled procedure when it cannot be reached?
A site engineer may need the current drawing. A subcontractor may need an approved specification. A design manager may need the latest model or RFI response. On higher-risk building work, a principal contractor may need evidence held within the golden thread. The Building Safety Regulator's golden-thread guidance requires relevant building information to be kept digitally, secure from unauthorised access and available when it is needed.
LCM has previously examined how digital systems are becoming part of building-safety assurance. Availability now deserves the same attention as integrity. A perfectly protected record that nobody can retrieve has not disappeared, but for the duration of an outage it cannot do its job.
AI Creates a Different Dependency
Automation goes further than storage because it performs work. LCM's earlier analysis of AI in UK construction examined the shift from periodic human reporting towards continuous digital control. As those systems improve, businesses will reorganise around them. Administrative roles may shrink, repetitive manual checks may stop, and knowledge once spread across several people may end up embedded in a workflow or agent.
That changes the resilience question. Losing an online filing system removes access to information. Losing an automated intelligence layer can remove the process that decides what to do with it.
Consider a chain in which a site form updates a database, an AI system classifies its contents, an API checks commercial data and a workflow sends the result for approval. No component needs to be compromised for the chain to fail. The model provider may be unavailable, authentication may fail, or the integration platform may stop accepting calls. A database can stay up while the API connecting it to another system goes down.
The vulnerability is not artificial intelligence itself. It is dependency without an alternative operating state. If an AI system normally performs a task in seconds, can a competent person still perform it in twenty minutes when necessary? If not, the organisation has not simply automated the task. It may have removed its fallback.
Cybersecurity Does Not Guarantee Availability
Construction's cyber-security investment remains essential. Multi-factor authentication, encryption, patching, secure backups and staff training all protect against serious threats. But this week's event was not a cyber attack. The confirmed trigger was a power event.
That exposes the gap between security and continuity. A contractor can have excellent cyber hygiene and still be unable to reach an application because the provider's infrastructure is down. The National Cyber Security Centre's cloud guidance treats physical resilience and availability as part of cloud security. It advises organisations to understand whether their configuration across data centres, zones or regions actually delivers the failover they expect.
Losing the primary application is only one failure mode. A working application can still become unusable if its identity provider, DNS, network route, database or external API is unavailable. A firm's digital dependency map therefore needs to go beyond a list of software licences and show the infrastructure underneath each critical workflow.
| Digital Function | Potential Dependency | Resilient Operating State |
|---|---|---|
| CDE / drawings | Cloud storage, authentication, network connectivity | Controlled read-only copy of latest published information |
| BIM coordination | Cloud models, collaboration services, licence/authentication systems | Local models for viewing and defined work that does not require live coordination |
| Site monitoring | Connectivity, IoT platform, cloud analytics | Local recording and edge processing with later synchronisation |
| AI / automation | Model API, identity, database, workflow platform | Local processing for critical bounded tasks or documented manual fallback |
| Procurement / finance | ERP, banking, supplier APIs, approval workflow | Defined emergency-authorisation and reconciliation procedure |
Should Construction Bring Some Intelligence Back Inside?
Not wholesale. Hyperscale cloud offers redundancy, security expertise and geographic spread that most contractors could never reproduce, and moving everything back on-premise could create more risk, not less. Nor is "multi-cloud" automatically the answer. Running identical systems across competing providers is expensive and complex, especially when large BIM datasets, integrations and permissions must stay synchronised, and the added complexity creates failure points of its own.
A more realistic model is hybrid and tiered. The cloud remains the primary environment for collaboration and high-capability AI. Critical published drawings and specifications are mirrored locally in controlled read-only form. Site sensors keep recording when connectivity drops, and edge devices handle essential monitoring. A major project or regional office keeps enough local capacity to continue selected operations, without pretending to reproduce a hyperscale cloud.
AI adds another option. Smaller specialised models can already run on company-owned hardware without a constant connection to an external provider. They will not match the largest cloud-hosted systems, but they may not need to for bounded tasks such as searching a controlled document set, classifying records, extracting information or checking routine forms.
The real choice is not cloud versus local AI. It is deciding which intelligence must stay available when the cloud is not. LCM's coverage of Digital Construction Week 2026 argued that digital construction is moving from standalone tools to operational infrastructure. Once that happens, availability becomes a project-delivery issue, not an IT department issue.
A Digital Continuity Plan, Not a Digital Panic Button
No contractor needs to survive the disappearance of the internet indefinitely. That would be expensive, disproportionate and, for most businesses, impossible. The useful question is what the project must still be able to do after ten minutes, two hours or one working day without its normal cloud services.
Safety-critical and regulatory information deserves the highest continuity requirement, and approved construction information should have a controlled fallback. Physical access, emergency systems and essential monitoring should not become unsafe because an external authentication service cannot respond. Commercial and administrative automation can usually tolerate longer interruptions, provided transactions are queued, reconciled and not duplicated when services recover.
This changes technology procurement too. Contractors buying a CDE, ERP, AI platform or automated workflow should ask:
- Where does the service run, and in which cloud region?
- What happens when that region or zone fails?
- Is data replicated elsewhere, and has failover actually been tested?
- How is identity handled, and does it have its own single point of failure?
- What works offline, and what can be exported?
Recovery Time Objectives and Recovery Point Objectives sound like IT terms, but they answer construction questions: how long can this function be unavailable, and how much recent information can the project afford to lose? Above all, firms adopting AI should document the human or local fallback before removing the old process.
Resilience Is the Next Phase of Digital Construction
The cloud has solved real problems for construction. Centralised information, live collaboration and remote access are major improvements over uncontrolled local files. AI can extend those gains. But greater intelligence creates greater dependency. A manual process can often continue slowly when technology fails. A fully automated one may stop entirely unless continuity was designed in.
Construction has spent decades engineering redundancy into structures, temporary works, lifting operations, power supplies and safety systems. Digital delivery now deserves the same mindset. The industry does not need to retreat from cloud computing or AI. It needs to stop treating them as weightless services that cannot fail.
Google typically publishes an incident report after major outages. When this one appears, construction technology teams should read it for more than the technical cause. The more valuable question is whether their own projects could continue safely if the next failure came during a concrete pour, a critical inspection, a design change or an automated approval.
Frequently Asked Questions
What caused the Google Cloud outage in London on 20 October 2026?
Google attributed it to a power event in the europe-west2-b zone of its London region. The disruption began at about 07:08 UK time, and Google said most services had recovered by early afternoon. Banks including Lloyds, Halifax, HSBC and Revolut reported problems.
Google attributed it to a power event in the europe-west2-b zone of its London region. The disruption began at about 07:08 UK time, and Google said most services had recovered by early afternoon. Banks including Lloyds, Halifax, HSBC and Revolut reported problems.
Should construction companies move away from cloud systems?
No. Cloud platforms provide major advantages in collaboration, scalability and information control. The issue is whether critical functions have proportionate alternatives when a provider, region, identity service or connection becomes unavailable.
No. Cloud platforms provide major advantages in collaboration, scalability and information control. The issue is whether critical functions have proportionate alternatives when a provider, region, identity service or connection becomes unavailable.
Could a construction company run AI without the internet?
Yes, for selected tasks. Smaller locally hosted models can process documents, search controlled datasets and perform bounded classification or monitoring tasks on company-owned hardware. They are not a complete substitute for the most capable cloud-hosted models.
Yes, for selected tasks. Smaller locally hosted models can process documents, search controlled datasets and perform bounded classification or monitoring tasks on company-owned hardware. They are not a complete substitute for the most capable cloud-hosted models.
Is strong cybersecurity enough to protect digital construction operations?
No. Cybersecurity protects against malicious threats, but operational continuity also requires resilience against power, connectivity, provider, identity and infrastructure failures.
No. Cybersecurity protects against malicious threats, but operational continuity also requires resilience against power, connectivity, provider, identity and infrastructure failures.
Expert Verification & Authorship
Founder & Author, London Construction Magazine | Construction Testing & Investigation Specialist