Amazon Lost a Country. Your Data Should Survive One.

Pillar 04 · Portability

Bias declaration. I founded SkyeConnex. We sell data custody. So read the second half with that in mind. The first half is just what AWS said.

What AWS said

On September 15 AWS posted its first public update on the Gulf strikes since April. Two lines matter.

Bahrain: the damage “spanned multiple Availability Zones and exceeded what our regional and multi-AZ services are designed to withstand.”

Then: “We have determined that we are unable to restore access to the resources and data hosted exclusively” in the affected zones.

So the whole Bahrain region, me-south-1, is gone. One of the three UAE zones, mec1-az2, is gone. If your data only lived there it is not coming back. AWS is still working on the other two UAE zones and says Bahrain gets another update in early 2027.

The timeline: drones on March 1. A second Bahrain zone knocked out in April. Structural damage, power gear destroyed, then the fire suppression flooded what was left. The IRGC claimed the strikes and said they picked Amazon because Amazon supports the US military.

Sit with that. A cloud provider was the target. Not collateral. The target.

Why was the data only in one place?

That is the question I keep hearing. The textbook answer is that an Availability Zone is a failure domain, and if your only copy is in one AZ, losing it is a known risk.

Fine. Bahrain lost the whole region. Multi-AZ worked exactly as designed and it did not matter. The customers who came out of this had copies in another region, restored from backup and rebuilt. AWS said as much. Their own discipline saved them. The platform did not.

So the better question is why the default cloud model puts all of the survivability work on the customer and then bills egress when they do it.

Redundancy and survivability are different things

An AWS region is a handful of data centres a few kilometres apart. Same country, same owner, same grid, same telecom backbone. As of March, same target list. Redundancy inside that box handles floods, fibre cuts, a dead transformer. Accidents.

A drone strike is not an accident. Someone chose the box.

Every cloud resilience design I have seen in twenty-five years assumes failure is random and local. The Gulf strikes were planned and regional. When the failure model changes, the old architecture turns an outage into a write-off.

I have said for a while that residency is where the server sits and sovereignty is who controls the data. Survivability is the third piece. If you cannot get it back, you never owned it, whatever the contract said.

How we are built for this

We did not set out to build for war. We built on the assumption that no single operator, country or building should be able to lose your data, read it, or hand it over. Same thing, it turns out.

Raidr.cloud is our patent-pending custody engine. Here is what it does.

Encrypt at the client. Data is encrypted before it leaves your boundary and the keys stay with you. No host ever sees plaintext.

Adaptive dynamic sharding. Each file is erasure coded into shards. The shard count and the reconstruction threshold are set per file by policy. A Protected B file in Ottawa gets a different coding than a marketing archive in Toronto. Say seven shards, threshold of four. Any four rebuild the file. Lose three to a drone, a court order or a bankrupt host and nothing is gone. Change the policy tomorrow and every file already written keeps the coding it was written with.

Scatter across independent operators and countries. Different companies, different jurisdictions, different grids, different laws. A strike on one host takes one shard. A subpoena to one host produces one shard of ciphertext, which is noise.

Rebuild when you lose a node. The surviving shards reconstitute the missing one on a different provider. You do not phone AWS and wait for 2027.

Run the Gulf scenario again with that in place. Shards in a Bahrain host, a UAE host, and several outside the region. Two shards lost. Threshold holds. File reads. Missing shards rebuilt on a Canadian or European host the same afternoon. Nobody in the business notices.

Same strike, same rubble, different result. That is arithmetic, not marketing.

Keep the hyperscalers

This is not an anti-AWS piece. They are very good at what they do and they are already inside every enterprise and government on the planet. Nobody is ripping them out.

SkyeConnex is Bring Your Own Cloud. AWS, Azure, Google, a domestic sovereign host, a rack in your own basement. All of them are shard targets. The custody layer sits above them and decides what each one gets to hold. Any one of them can burn, get seized or get bought and your data is still yours, still readable, still recoverable.

That is what I mean by sovereignty by architecture, not by promise. A promise is a Health Dashboard post in September telling you what happened in March.

For governments

AI and cloud infrastructure are strategic assets now. Strategic assets get targeted. The IRGC said so out loud.

Any government running critical workloads on one provider's regional footprint, Canada included, is carrying a risk nobody has priced. Residency rules that force the data to sit inside the country make it worse. They pile it into a few buildings that are easy to find.

A bigger bunker does not fix that. Data that can lose a bunker and keep going does.

Ross Norrie
Founder & Architect, SkyeConnex
[email protected] · skyeconnex.com