October 20 Is Now a Cloud Holiday. Your exit plan still isn’t on the Calendar.

Every hyperscaler gets an October. Not every buyer gets an exit.

Three clouds, one calendar

On October 20, 2025, a race condition in DynamoDB’s DNS automation disrupted AWS us-east-1 for about 14 hours. [5] Nine days later, a configuration change degraded Azure Front Door globally, taking Microsoft 365 and Xbox along for the ride. [6] On October 20, 2026, one year to the day after AWS, a third-party facility hosting Google Cloud’s europe-west2-b zone in London lost utility power, and a controller fault kept the backup generators from taking over. [1]

Three providers. Three causes. One date. At this point, October 20 deserves to be a public holiday for SREs.

The London one is the freshest, so let’s start there. Revolut runs its app and back end on Google Cloud. In September 2025, it signed a multi-year expansion with Google that also covers its AI fraud tools. [2] On the morning of October 20, Revolut customers couldn’t use the app or pay by card, and its status page showed card payments and cash withdrawals down. Users in the UK, Ireland, France, Spain, Romania and Germany reported problems. [2] Reuters named Lloyds and HSBC among the banks affected. [3][19]

Google’s advice mid-incident was sensible and brutal: move your workloads to another zone or region if you can. [2]

If you can. That’s the whole edition in three words.

Before anyone forwards this to Google comms: Google’s outage was the shortest of the three, about six hours on its own clock. A single zone losing power is exactly the failure zones are designed for, and Google published a detailed preliminary report within about two days. [1] The AWS and Azure outages were software faults with a bigger blast radius. This isn’t a Google story. It’s a buyer story, and every buyer on every hyperscaler is in it.

Copies of your data didn’t keep anyone’s card working. As TechRadar’s live coverage put it, copies of your data don’t, on their own, keep customers paying. [4] Owning the keys and owning the exit are two different things.

Two exits nobody tested

Every buyer has two exits. One you need in an hour, when a zone goes dark. One you need in a decade, when the renewal comes due. Both were decided at the last spec.

Ten days after London went dark, the UK Home Office published a notice to renew its AWS hosting: £881.6m excluding VAT, December 2027 to December 2029. [7] The implied annual commitment jumps from about £150m to about £441m. [7] In fairness, the department is moving more systems onto the cloud, so not all of that is price. Computer Weekly’s read was simpler: AWS has become the default. [7]

That’s not an AWS sin. It’s gravity. The UK competition regulator found that fewer than 1% of cloud customers switch providers in a year, and named egress fees and Microsoft’s licensing terms as two big reasons. [8] The remedy, in March: voluntary egress and interoperability commitments from AWS and Microsoft, plus a strategic market status probe into Microsoft’s business software. [9] Microsoft stretched its free-egress switching window from 60 days to 180. [9] Generous. You now have six months to move a decade.

Thomas Allgeyer’s weekly sovereignty roundup made the point that stuck with me: the US clouds named in European tenders are mostly renewals, locking buyers in before the next spec is written. [18] Sovereignty isn’t decided when you write the sovereignty requirements. It was decided three contracts ago.

Someone did it the other way. The Dutch Ministry of Defense just signed with SAP for a sovereign cloud in the Netherlands, due live on January 1, 2028, as one piece of a multicloud strategy. [14] State Secretary Derk Boswijk’s selling point wasn’t the features. It was that Defense keeps control of its data and can move it to another cloud whenever it chooses. [14] The exit was part of the pitch on day one. Novel concept.

The 5 Pillars, graded from the buyer’s seat

Same test as always, different seat. I’m grading a regulated European firm running single-provider on any of the three hyperscalers. I’m grading the dependency, not the company.

01 Territorial

The question: Where does your data live, and can it stay there when a zone dies?

The story: The outage didn’t move a single byte across a border. Residency held. Payments didn’t.

Under pressure: Residency rules and continuity can collide. If your rule says in-country and your provider gives you one in-country region, your failover target is either another country or a prayer. Find out which before your regulator asks.

Verdict: Partial.

02 Operational

The question: Who runs the environment when it breaks, and can you run without them?

The story: Google advised moving to another zone or region. [2] Afterward, Revolut said it is reviewing the incident to reduce the impact of future third-party disruptions. [3] Translation: Whatever failover existed didn’t keep cards working that morning.

Under pressure: Since July 13, UK regulators have directly overseen AWS, Google Cloud, Microsoft and Oracle as critical third parties. [10] The FCA said it was engaging with Google. [3] None of that moves the obligation off the bank. Under EU and UK rules, the bank stays responsible for the service its customers get. [3][20] The regulator watches the provider. You still own the outage.

Verdict: Not Demonstrated.

03 Technological

The question: Can your stack move?

The story: On November 1, Vercel CEO Guillermo Rauch posted “Own your intelligence” to more than 280,000 views, telling builders to own their “harness, data, connections, runtime.” [15] He’s earned the line. On October 20, 2025, the us-east-1 outage hit Vercel’s own control plane, dashboard, API, and builds. [16] Rauch posted the impact graphs in public and said there was more to do. [17] That’s candor, and it’s the right lesson learned the hard way.

Guillermo Rauch (@rauchg) on X, November 1, 2026. [15]

Under pressure: Rauch also says anyone who opposes this has something to sell you. True. Every platform sells a runtime, Vercel included. The test isn’t whether you rent. It’s whether you can leave. Every proprietary managed service you adopt adds a line to the exit bill, and agents make the bill bigger. If your orchestration, memory, and model endpoints live in one vendor’s managed services, your agents inherit that vendor’s zone, region, and renewal. When the zone goes dark, your digital workforce does too.

Verdict: Partial.

04 Legal

The question: Who answers when it breaks?

The story: Last November, the EU designated 19 critical ICT providers under DORA, including AWS, Google Cloud, and Microsoft. [11] The UK followed in July. [10] Reuters called October 20 the first major test of those rules, and Google Cloud sits under both regimes. [3]

Under pressure: Designation buys oversight, information requests, and eventually rules for the provider. It doesn’t buy you an exit, and it doesn’t take the outage off your books. [20] And for a European buyer on any US hyperscaler, a second legal clock is always running. The US CLOUD Act lets US authorities demand any data a US provider has in its possession, custody or control, wherever it is stored. [22] London or Frankfurt residency doesn’t change the answer. Customer-held keys help, and Microsoft’s new package of customer-held root keys plus EU-partner oversight of engineer access is the best any hyperscaler has put on a public cloud. But the CLOUD Act doesn’t ask who watches the visits. It asks who has custody, and only the provider gets to argue back. [21] Same rule for AWS, Google and Microsoft alike.

Verdict: Partial.

05 Financial

The question: Can you afford to leave?

The story: Fewer than 1% switch in a year. [8] The Home Office renewal implies roughly three times the old annual rate. [7] On January 12, 2027, the EU Data Act bans switching charges outright, including egress for the switch. [12]

Under pressure: The Data Act zeroes the fee you can see. It doesn’t touch the one you can’t: re-platforming off proprietary managed services. Egress for ongoing multicloud use can still be charged at cost. [13] And if a renewal on your desk still prices switching charges after January 12, strike them. The law already did.

Verdict: Not Demonstrated.

Scorecard: zero Demonstrated, three Partial, two Not Demonstrated. That’s not a vendor failing. That’s a buyer-posture failure, across all three clouds.

The executive TCO

Three lines your CFO actually needs:

• The cost of resilience. A second zone or region, tested under real load. Known, budgetable, boring.
• The cost of not having it. A day without card payments, plus a regulator asking how you plan to stay inside your impact tolerance next time. Not budgetable. Not boring.
• The cost of exit. Priced at the last spec, not the next one. Baseline on portable components, own the crown-jewels layer, and burst into the proprietary stuff only where it earns its keep. Then the renewal is a negotiation, not a hostage situation.

Egress was always the decoy. The real exit fee is the re-platforming bill, and it grows every time someone clicks “enable” on another managed service.

Monday

1. Map every single-zone and single-region dependency behind each important business service, including the provider’s own hidden ones.
2. Run the failover under real load. If it has only ever existed on a diagram, assume it doesn’t exist.
3. Price the exit before the renewal, and write portability and exit-assistance terms into the spec.
4. Use January 12, 2027 as leverage in any renewal currently on the table.
5. Find out where your agent orchestration, memory, and model endpoints run. If the answer is “one vendor’s managed services,” your agents have the same exit problem your databases do.

The exit is the sovereignty

Allgeyer’s roundup carried a line I’d frame on the wall of every procurement office: “Holding the keys doesn’t mean holding the decision.” [18] I’d add a second one. Holding a region doesn’t mean holding an exit.

None of this is a reason to leave the hyperscalers. They run the best infrastructure on the planet, and all three recovered within hours. It’s a reason to stop pretending the exit takes care of itself. Self-awareness is the whole battle. Know which dependencies you’ve accepted, which ones you’ve tested, and which ones you’ve simply never looked at.

The Dutch Ministry of Defense built its exit into the plan before the cloud even existed. [14] You can do the same at your next renewal. The next October 20 is less than a year away. Put it on the calendar.

Amit

Disclosure and References

Amit Eyal Govrin is lead analyst for sovereign AI at theCUBE Research and CEO of Agentcy Labs, which designs and builds sovereign AI platforms for enterprises and governments. His research for theCUBE Research is conducted under theCUBE Research’s editorial standards and is not sponsored or directed by Agentcy Labs. Agentcy Labs may hold commercial relationships with vendors covered in this research.

1. Google Cloud, Service Health: europe-west2 incident and preliminary incident report, October 2026.
2. TNW, Google Cloud outage hits London region as Revolut also goes down, October 20, 2026.
3. Reuters, FCA engaging with Google; Revolut reviewing the incident, via Communications Today, October 2026.
4. TechRadar, Google Cloud went down after a power event, October 20, 2026.
5. AWS, Summary of the Amazon DynamoDB Service Disruption in the Northern Virginia (US-EAST-1) Region, October 2025, and SiliconANGLE, Hours-long AWS outage takes numerous services offline, October 20, 2025.
6. SiliconANGLE, Azure outage takes multiple Microsoft services, customer websites offline, October 29, 2025.
7. Computer Weekly, Home Office plans £881.6m AWS deal, at nearly three times spend rate, October 2026.
8. Data Centre Review, AWS, Microsoft to face competition probe as CMA concludes UK cloud market investigation, August 2025.
9. Computer Weekly, CMA to launch strategic market status investigation into Microsoft, and SDxCentral, Microsoft’s switching changes, March 2026.
10. GOV.UK, UK financial system strengthened with new safeguards for major technology providers, July 2026.
11. Reuters, Amazon, Google named by EU among critical tech providers for finance industry, November 18, 2025.
12. Deloitte Legal, Cloud switching under the EU Data Act.
13. WZ-IT, Data Act switching charges 2027.
14. ITdaily, Dutch Defence chooses SAP for sovereign cloud environment, October 2026.
15. Guillermo Rauch (@rauchg), “Own your intelligence,” X, November 1, 2026. (Add post URL.)
16. Vercel, Update regarding Vercel service disruption on October 20, 2025. (Swap for the vercel.com original.)
17. Guillermo Rauch, us-east-1 outage graphs, X, October 2025.
18. Thomas Allgeyer, Best of LinkedIn CW 43/44: Cloud Insights & Sovereignty, Frenus, November 2026.
19. AFP, UK power outage hits Google Cloud services, disrupts Revolut banking, October 20, 2026.
20. Morrison Foerster, UK announces list of first critical third parties, July 2026.

21. Amit Eyal Govrin, Europe Got Visitation Rights. Microsoft Kept Custody., theCUBE Research, October 2026.

22. US Department of Justice, Justice Department Announces Publication of White Paper on the CLOUD Act, April 2019.