The Cloud Your Business Runs On Is Funding the AI Buildout. Plan for Cloud Outages Yourself

Cloud outages are no longer rare events that hit someone else. In October 2025 two of the largest cloud providers each had a major outage, one week apart, and both started with a small change that spread faster than anyone could stop it. With the biggest providers pouring money into AI data centers, a small business should plan for the next outage instead of hoping its provider prevents it.

Two cloud outages, one week apart

On Monday, October 20, 2025, the largest cloud provider's main region in northern Virginia began failing in the early morning. CNBC reported that the provider's status page blamed "DNS problems" with a database service that many of its other services depend on. Problems spread to apps people use every day, from ride-hailing and food ordering to airlines and doorbell cameras, and the provider did not report all services back to normal until that evening.

Nine days later, on October 29, a different top-three provider went down. CNBC reported that the company suspected an "inadvertent configuration change" in the service that routes traffic at the edge of its network, and that it was "rolling back to our last known good state." An airline said the outage disrupted "key systems," including its website.

What these cloud outages had in common

Different companies, same shape:

• A small trigger. One was a name-resolution fault in a core service. The other was a configuration change. Neither was a cyberattack and neither was a shortage of servers.
• A huge blast radius. Because thousands of businesses share the same control systems, one fault became everyone's outage at once.
• No say for the customer. If your business ran there, cloud outages meant you waited, refreshed a status page and told your own customers you did not know when it would be back.

Three weeks later, the run of cloud outages reached a different layer. On November 18, 2025, a large network and security provider that sits in front of millions of websites failed for several hours. CNBC reported that a company spokesperson traced the "root cause" to an automatically generated configuration file that "grew beyond an expected size of entries," and said there was no evidence of an attack. Shopping sites, social networks and chat tools went down together, again because they shared one provider.

1. Jul 31, 2025 UK regulator finds a concentrated market. The Competition and Markets Authority publishes its final decision on cloud services.
2. Oct 20, 2025 Largest provider fails in northern Virginia. A DNS fault in a core database service spreads to apps across the internet.
3. Oct 29, 2025 Second top-three provider goes down. An inadvertent configuration change at the network edge, rolled back by the provider.
4. Nov 18, 2025 Network edge provider fails. An oversized configuration file crashes traffic handling for several hours.
5. Feb 6, 2026 Capital spending climbs again. The four biggest spenders plan to go beyond 2025's record to fund AI build-outs.

Where the money is going

The largest providers are in the middle of the most expensive build-out in their history. CNBC reported in February 2026 that the four biggest technology spenders expected to raise capital spending well beyond the record levels of 2025 to "fuel their AI build-outs," buying chips, building facilities and adding networking to connect it all.

That spending is aimed at AI capacity. The general-purpose servers, storage and networking that run your email, files and line-of-business apps are the older part of the building. Nobody announced that the old wing would get less attention, and better AI hardware may well come with better operations everywhere. But it is fair to ask where the engineering hours and attention go when one budget line is growing that fast, and the cloud outages above did not come from a lack of AI chips.

Electricity shows the scale of the build. The Department of Energy reported in December 2024 that "data centers consumed about 4.4% of total U.S. electricity in 2023 and are expected to consume approximately 6.7 to 12% of total U.S. electricity by 2028." The report behind it, from Lawrence Berkeley National Laboratory, ties the recent jump to the rapid arrival of AI hardware.

U.S. data center electricity use is highest in 2028 (high estimate) at 580.

U.S. data centers used 176 TWh in 2023, and Berkeley Lab expects 325 to 580 TWh by 2028 as AI servers spread. That growth is where the largest providers are putting their money, while the shared systems behind recent cloud outages run on as before.

	U.S. data center electricity use
2023 (measured)	176
2028 (low estimate)	325
2028 (high estimate)	580

Why concentration makes cloud outages worse

Most of the market now sits with a few providers. When the UK's Competition and Markets Authority finished its cloud market investigation in July 2025, it found the infrastructure market "highly concentrated," with the two largest providers each holding a 30 to 40 percent share of supply in 2024. It concluded that competition "is not working well," pointing to barriers such as the fees providers charge to move data out.

That concentration makes cloud outages hurt a small business in two ways. First, blast radius: when much of the internet runs on a handful of control systems, one bad change takes down a region, not a single customer. Second, few exits: the same fees and incompatibilities that the regulator flagged make it slow and costly to move, or to keep a working copy somewhere else. Spreading across two of the big providers can help, but the two October failures show that each of them can fail in similar ways, and running on both doubles the work of keeping both healthy.

What an SLA credit buys you

When cloud outages push a provider past its uptime promise, the remedy in the service level agreement is usually a credit on your bill. You get a slice of your money back. You do not get the lost day, the orders that never came in or the staff time spent waiting.

It helps to turn the promise into hours. A 30-day month has 43,200 minutes, so a 99.9% uptime promise allows about 43 minutes of downtime a month before any credit is owed. The math gets worse when your business depends on several services in a chain. If your website, your payment service and your sign-in service each promise 99.9% and any one of them failing stops a sale, the chain as a whole is up about 99.7% of the time, a little over two hours of possible downtime a month.

How a small business can plan for cloud outages

You cannot stop the next of these cloud outages, but you can decide in advance what happens when it comes. A short plan:

1. List what depends on one provider. Email, files, phones, payments, your website, your booking system. Note which ones share a provider or a region.
2. Decide what must keep working. For most businesses that is taking payments, reaching customers and reading the customer list. Everything else can wait a day.
3. Keep a copy you control. A recent export of customer data and key files, stored somewhere that does not depend on the same provider.
4. Write the four-hour script. Who tells customers what, how staff keep working on paper or phones, and who checks when service is back.
5. Know who answers. If your systems are managed, know the name of the person who picks up during an outage, not just a ticket address.
6. Review it once a year. Recheck the list after any of your providers changes, and after any cloud outages that touch you.

Where LTFI fits

LTFI runs each company on its own server that Kief Studio hosts and manages, rather than one shared pool for every customer. Our page on managed versus self-hosted AI agents explains the trade-off, and we keep access, updates and certificates current and patch and watch the server as part of the service. A server we run can still fail, which is why the accountability matters: when something breaks, a named team at Kief Studio owns it and tells you what is happening. If you want to talk through where your own systems sit, get in touch.

Frequently asked questions