AWS, Azure and Google Cloud All Went Down. Is Your Data Safe?

2026 has been a rough year for clouds

This year the major cloud providers took turns having serious accidents. A few representative ones:

• February, Azure West. A data center transformer failed. The UPS batteries held for about six minutes, and a cascading failure in the control system meant the automatic switch from mains to generators never completed. Roughly 20 hours of downtime. The irony: Azure Site Recovery, the disaster recovery service, was itself on the outage list.
• May, AWS us-east-1. The cooling plant could not carry the heat load, overheating tripped the power, and a batch of server hardware burned out. Recovery took more than 28 hours. Coinbase trading was down for nearly 7 hours. AWS itself conceded that cooling capacity and thermal load were "mismatched": a single AI server rack now draws 3-10 times what it did ten years ago, and the cooling design of older halls cannot keep up.
• June, New Delhi. A lithium battery fire in an STT GDC / Tata Communications facility burned server racks and electrical gear. Google Cloud's local nodes were shut down for nearly three weeks. The worst case was a tenant, Matrix Cellular, an international SIM company, which may have permanently lost more than 20 years of operating data. Its CEO: "It's been 20 days and the backup hasn't been restored. If there was a backup, it would have been restored by now."
• August, Phoenix. Storms caused power fluctuations, chillers failed, room temperatures soared, and hosting providers including Liquid Web disrupted service for customers for about 37 hours. The facility did not even publish a status page, so customers sitting behind several layers of resellers had no idea what was happening.

The common thread is clear. This year's big outages had almost entirely physical root causes: transformers, cooling, batteries, weather. Not software bugs. The other side of the cloud is a data center made of reinforced concrete, cable and chillers, and it is afraid of fire, heat and storms like any other piece of human infrastructure.

Three lessons for small and mid-size businesses

1. The cloud is an availability product, not a backup

Putting data in the cloud solves access-from-anywhere and single-server failure. But cloud data still faces a facility-level disaster, your own mistakes (delete a folder, empty the recycle bin 30 days later), account theft, suspension for non-payment, and vendor policy changes. Matrix Cellular's 20 years of data sat in a provider's data hall and could still go to zero after a fire, because the company had not bought a backup tier.

Tata Communications was blunt afterwards: customers who had subscribed to recovery and backup services were restored, and the rest were still waiting. Resilience in the cloud is sold by service tier. It is not included by default.

2. The AI buildout is crowding out reliability spending

The AWS cooling failure was not bad luck. Industry analysts broadly point out that cloud providers are pouring capital into GPUs and AI workloads, which thins the redundancy investment in ordinary general-purpose capacity. Forrester has gone as far as predicting multi-day cloud outages in 2026. What that means for you: putting your business systems on a single region of a single big cloud concentrates risk rather than spreading it.

3. The old server in the back room still earns its place

Many of our customers run hybrid: collaboration in the cloud, but core data such as drawings, finance and customer records kept in full on a local NAS or server, then synced to the cloud as an encrypted backup. After the fire in India, several customers called us unprompted to ask how to harden their local backup, which is the right instinct. A local copy does not care about a cloud fire; a cloud copy does not care about a flooded office. Two legs are steadier than one.

3-2-1, the small-business version

The principle is old: three copies, two media, one off-site. Translated into actual work:

Three habits matter more than the design:

1. The offline copy has to actually be offline. Unplug the backup drive and put it in a cabinet. Ransomware crawls along the network, and a backup left mounted on it is a casualty waiting to happen (see our ransomware case).
2. Restore something every month. Pick a file at random, restore it from the backup and open it. We have customers whose backup jobs had been failing silently for three months and nobody noticed.
3. Encrypt the off-site copy. Put a password on the archive you send to the cloud, which 7-Zip's AES-256 handles. Do not hand customer data to a third party in plain text.

So should you abandon the cloud? No

In fairness, cloud availability SLAs of 99.9% and up are still far better than what most small businesses achieve in a self-built server room. The incidents this year hit specific services in specific regions, and most users never felt them. The right response is not to leave the cloud but to:

• keep core business off a single region and a single availability zone, and buy multi-AZ if the budget allows;
• check a provider's status page and incident history before you sign, and its supply-chain transparency, meaning whether you hear about an incident when it happens;
• treat "one copy in the cloud, one local, one offline" as the baseline. The cloud is one of three.

Data is at risk wherever you put it. More than one place is what makes it safe, and after this year the point deserves to be taken literally.

If you do not know how to put an enterprise backup plan together, call 020-39029800. We will come and look at what you have and write the plan, and the plan itself is free. NAS selection, backup job configuration and encrypted cloud sync can all be done the same day in Panyu and Nansha.