Watch the video, transcript and sources
- europe-west2-b: power event
- UK bank apps disrupted
- DHS: DeepSeek out of agencies
+ DeepSeek opens DeepInferYou think a cloud region survives losing one building. On Tuesday one Google zone in London lost power, and UK banking apps fell over.
Google Cloud's London region, europe-west2, is split into three zones, a, b and c. The point of zones is that one can go dark while the other two keep running. Google's status page gives a start of 07:08 London time (it writes everything in Pacific time, for a London outage). For hours the only description was "elevated errors and intermittent network failures", and the workaround field said "None for now". Later it suggested moving your workload to another zone or region, if feasible.
At 12:12 London time Google named the cause: "a power event" took out the europe-west2-b zone. Services in that zone were unavailable, and some services across the rest of the region saw errors too. Thirty-eight products are listed on the incident. Google recovered existing workloads first, warned that new ones might not be creatable yet, and by 13:10 said most workloads were recovering. At 15:03 it called the issue mitigated for most customers, said a very small number were still affected, and promised an analysis. No root cause beyond "power event" yet.
Who said what. Revolut told the BBC its disruption came from a third-party issue and that customers' money and data were safe. HSBC apologised and said the situation was "also affecting other organisations". Lloyds said its app worked while some people had issues. The courts service said its remote-hearing video links were hit. The BBC notes most firms have not officially tied their problems to Google, and Downdetector's 1,500-plus Revolut reports are user reports.
The part you control. A VM or a zonal disk lives in exactly one zone. If your whole service sits in zone b, the power event is your outage. Spread it across a and c and you lose capacity instead of customers; the spare capacity costs money, which is why teams skip it. Monzo has written up the bigger version: its main platform runs on AWS, and a separate stand-in on Google Cloud can keep card payments and transfers going.
The open loop. In July 2022, cooling failed in a data centre hosting zone a of this same region. Google's incident report admitted regional services were hit "despite regional services being designed to survive the failure of a single zone": a traffic change steered internal services away from all three zones instead of one, so storage could not reach any replica of some objects. Google promised steps to improve the region's resilience. On Tuesday, one zone went dark and some regional services threw errors again. Google's promised analysis should say why.
Also today. DHS ordered DeepSeek products out of US federal civilian agencies, including "third-party products and services developed using DeepSeek implementations", published in the Federal Register today. The same day, DeepSeek's open-sourced inference engine DeepInfer topped Hacker News: Apache 2.0, one model (V4.1 Flash), NVIDIA SM100 or Huawei Ascend 950, offline inference only, with no KV cache system or load balancing in this release. The interesting switch is batch-invariant mode, which keeps results identical across batch compositions (NVIDIA only).
Verdict: NEEDS REVIEW — One zone failed. London wobbled again. A zone losing power is exactly what zones exist for. I want Google's report to explain why the rest of London wobbled, and I want your bank to explain why one zone was enough.
Sources
https://status.cloud.google.com/incidents/Xs2XTHzAVC9YnafnPyzj
https://cloud.google.com/compute/docs/regions-zones
https://www.bbc.co.uk/news/articles/cxdd8mm6qy4mo
https://www.bbc.co.uk/news/live/c58jd2kjp93xt
https://status.cloud.google.com/incidents/fmEL9i2fArADKawkZAa2
https://monzo.com/blog/tolerating-full-cloud-outages-with-monzo-stand-in
https://news.ycombinator.com/item?id=50163961
https://public-inspection.federalregister.gov/2026-21447.pdf
https://github.com/deepseek-ai/DeepInfer
Caveats: Google had published no root cause beyond a power event when we checked (15:04 London time). Company statements are quoted as given; we do not claim any named bank runs on Google Cloud. Architecture advice is editorial.
And that's the diff for today. I'm Niko from Axrisi. Merge responsibly.