At 7am London time, a lot of Britain stopped working. Not the power grid, not the phone network — one region of Google Cloud. And the outage rippled into banks, a postal service, a property site, a credit bureau, and a health service that have nothing to do with Google.
one Google Cloud zone took down banks, Royal Mail, Equifax and NHS England sites
cloud regions Aura depends on — it runs in the room
the only architecture with no server to breach — or to lose
At 7am London time, a lot of Britain stopped working. Not the power grid, not the phone network — one region of Google Cloud (europe-west2, a single cluster of machines in London) started throwing errors, and it rippled outward into things that have nothing to do with Google.
Revolut confirmed everything was down — sign-ups, sign-ins, payments and transfers, top-ups, ATM withdrawals, stock trades, even crypto. Lloyds, HSBC, Halifax and Bank of Scotland degraded. Royal Mail and DPD hit problems. Rightmove went down. The credit bureau Equifax went down. NHS England's websites went down. BT and even Cloudflare felt it. Google hasn't said what caused it; by mid-morning there was no fix, only advice that affected customers “try to migrate their workload to another location” — cold comfort when the location is on fire and you don't run the servers.
Here's the part worth sitting with: none of these are Google. They're banks, a postal service, a property site, a credit bureau, a health service. They went dark because, somewhere underneath the app you actually touch, a decision was made years ago to rent the floor instead of owning it. When the floor gives way, so does everything standing on it — all at once, for reasons you can't see and can't fix.
That's not a Google-failure story. It's a dependence story. Centralize enough critical services on one provider's one region and you've built a single point of failure with a nice logo on it.
One Google Cloud region in London failed. Banks, Royal Mail, Rightmove, Equifax and NHS England sites went down with it — none of them Google.
They all rented the same floor. When it gave way, everything standing on it fell at once — for reasons the companies couldn't see or fix. That's cloud dependence, not bad luck.
Run the intelligence on the device in the room. No cloud region to go down, nothing transmitted to leak — privacy and resilience are the same decision.
We built Aura the other way, on purpose. It runs on the device in the room — the speech recognition, the reasoning, the memory, all on the hardware on the shelf, not in a data center you hope stays up. There's no “our cloud region” to go down, because there's no cloud region. Unplug the internet and Aura still hears you, still answers, still works.
That started as a privacy argument — your conversations never leave the room, so there's nothing to leak, subpoena, or breach. This morning it showed its second face: the same thing that keeps your data in the room keeps the service in the room. Own the compute, and nobody else's bad morning becomes yours.
Aura runs entirely on the device in your room.
Healthcare is where this stops being abstract. A bedside assistant that answers a patient's questions at 2am is only useful if it answers at 2am — including the 2am when a cloud region in another country is down and the other systems are showing error pages. An on-device assistant doesn't transmit patient data anywhere, which is what makes it private by design; it also doesn't stop when the network does. The outage that took NHS England's sites offline this morning is exactly the failure mode a room-local device is built to sit out.
The cloud is extraordinary infrastructure. It is also somebody else's computer, and this morning somebody else's computer had a bad day. We'd rather the answer live where you are.
Know someone who should see this? Send it to them.