AWS told customers on September 15 that it cannot restore access to resources and data hosted exclusively in its Bahrain region, or in one of the three availability zones of its UAE region. The notice, first reported by Reuters and covered by CNBC, reads: "After a thorough assessment, we have determined that we are unable to restore access to the resources and data hosted exclusively in this region." If your only copy lived in me-south-1, it is gone.

Six months passed between the first strike and that sentence. Most of what went wrong happened inside that gap, and most of it happened before the strikes.

What happened to the two regions

On March 1, Iranian drones hit AWS facilities in the UAE and Bahrain during the opening days of the war. Two of the three availability zones in me-central-1 (mec1-az2 and mec1-az3) went offline. EC2, S3, DynamoDB, Lambda and RDS all degraded across the region. AWS later described structural damage to buildings, lost power delivery, and water damage where fire suppression systems triggered. One Bahrain facility was hit in the same wave.

A second Bahrain availability zone was disrupted in April. The whole me-south-1 region became unavailable at that point and has stayed that way. On July 21, Iran's Revolutionary Guard claimed it had put cruise missiles into the offline Bahrain site a second time. The Register covered the claim and noted that the AWS Health Dashboard entry for the region had not been updated since April 30.

The September notice splits the two countries. In the UAE, mec1-az2 is permanently inaccessible, and AWS says the other two zones are still being recovered, with no date attached. In Bahrain the damage spanned multiple zones and the next update is promised for early 2027. AWS's own line: "The damage to our infrastructure spanned multiple Availability Zones and exceeded what our regional and multi-AZ services are designed to withstand."

Most customers left in time. AWS advised Bahrain customers to move to other regions after the March strike, and according to The National, most did so before the April disruption took the region down. UAE customers restored from backups or from data in the zones that stayed reachable. The people hurt by this week's notice had a single copy, in a single region, and either could not move it in the window or did not know they needed to.

The design envelope was published

AWS has never claimed a region survives the loss of all of its zones. The documentation defines a region as a minimum of three isolated, physically separate availability zones within 100 km of each other, and lists what that design protects against: power outages, lightning strikes, tornadoes, earthquakes. The S3 FAQ has said for years that S3 Standard is designed to sustain the concurrent loss of data in two facilities. Bahrain lost more than two.

The eleven nines of durability quoted in every S3 pitch deck is a figure for object loss inside a healthy region. It is disk failure math. It has always carried a footnote that One Zone storage classes can lose data if the zone is damaged, and nothing in that number ever covered a region being hit from the air twice in five months.

Harshwardhan Choudhary of ABN AMRO Clearing Bank said it in one sentence back in March, quoted by InfoQ: "Multi-AZ is NOT disaster recovery. It protects from hardware failures, not missiles." I have run hosting infrastructure for 25 years and I have watched that distinction get lost in every generation of tooling. High availability handles the failures the vendor planned for. Disaster recovery handles the ones nobody planned for, and it has to live somewhere else.

The 100 km rule is the mechanism. AWS keeps zones close together so synchronous replication stays fast. That same proximity means one drone campaign against one small country reaches every zone. Bahrain is about 780 square kilometers. Three zones inside it are three buildings in the same target area.

Data residency made the single copy legal and mandatory

The fix everyone will name is a copy in another region. For many of these customers that copy was against the law. UAE health data must be stored inside the UAE under the 2019 health data law. Bahrain's data protection law restricts transfers abroad. Mohamed Radwan of T-Systems said it in the same InfoQ piece: data residency is the law. The law did not contemplate the region ceasing to exist.

So an operator in Manama did what compliance required. It used the local region, spread its data across three zones, and kept no copy outside the country. Then the region stopped existing and its regulator had no answer either. I expect Gulf regulators to rewrite these rules within the year, because the alternative is to admit that compliance and survival are in conflict.

Any residency rule that confines data to one cloud region inside a small country assumes that region is permanent. That assumption failed in six months. The workable versions of the rule allow a second copy in a second jurisdiction under contract, or a copy on media the operator controls inside the country and outside the cloud provider's buildings. Both are old ideas. Both are cheaper than the alternative these customers just paid.

What I'd change this week

I maintain a backup server project, so I have a bias, and I'll state it. A copy that lives on infrastructure you do not rent from the same vendor is the only copy that survives the vendor's design envelope. Another cloud works. So does a rack in a colo you control. The medium matters less than the fact that it does not share a failure with the primary.

Four things I would check before Friday:

  • List every dataset that exists in exactly one region. One region, not one zone. Most shops have never run that query.
  • For each one, find out whether a copy in another region is legal. If it isn't, the answer is a copy outside the cloud provider, inside the jurisdiction.
  • Read the force majeure clause in your cloud agreement. The AWS Customer Agreement excuses failures caused by war and acts of terrorism. Do not budget for compensation.
  • Restore from the second copy into a region you have never deployed to, and time it. The Bahrain customers who survived were the ones who moved in March. Moving is a skill.

AWS handled the communication about as well as I would expect. It told people to leave in March, waived usage charges in me-central-1 for that month, and its final notice used the phrase "hosted exclusively" so nobody could misread who lost what. It did not pretend the region would come back. The failure belongs to an assumption shared by customers and regulators alike, that a cloud region is a place where data lives forever.

A region is a set of buildings within 100 km of each other. Buildings can be hit. AWS runs three zones in Israel and has three more planned for Saudi Arabia. I'd bet at least one Gulf regulator publishes a rule requiring an offshore or offline copy before the early 2027 Bahrain update arrives.