AWS Abu Dhabi and Bahrain Data Centers Still Down 6 Months After Drone Strikes | AINave

Six months after Iranian drone strikes damaged AWS data centers in Abu Dhabi and Bahrain, the company has advised customers to move their data to other regions and warned that the MEC1-AZ2 Availability Zone is permanently lost. For AI builders who run training or inference close to users in the Middle East, the incident is a practical lesson in why single-region deployments carry real physical risk.

What happened to the AWS Abu Dhabi and Bahrain data centers

On March 15, 2026, Iranian drones struck AWS facilities in the UAE and Bahrain, causing fires and severe infrastructure damage. AWS initially reported partial disruptions and launched recovery efforts for its MEC1 region, which includes Availability Zones MEC1-AZ1, MEC1-AZ2, and MEC1-AZ3. By mid-September, AWS announced that MEC1-AZ2 resources and data could not be restored. The company is still working on recovering regional and zonal resources in the other two AZs, but it has not provided a timeline for full restoration.

AWS Support has been helping customers move applications to alternate regions. Most customers re-established operations by restoring backups or copying data that remained accessible. For those who had not configured cross-region replication, the outage meant permanent data loss.

The lower API price matters most for long agent runs

The direct lesson for anyone building AI products is that multi-region redundancy is not optional when your compute lives in a geopolitically sensitive area. Model training pipelines, inference endpoints, and data lakes that were concentrated in the MEC1 region became unreachable for months. Teams that had automated backup-to-S3 in a different region or used AWS Backup with cross-region copy were able to shift workloads within days. Those relying on single-AZ persistence or local instance stores lost data completely.

This is especially relevant for AI builders who chose the UAE region for low latency to users in the Middle East, South Asia, and parts of Africa. They now face the operational cost of redesigning their deployment topology mid-outage, with no guarantee that the repaired data centers will be available before 2027.

This makes browser-use agents easier to deploy

Beyond the immediate disruption, the attack is reshaping discussions about data-center physical security. UAE officials are already considering underground data centers and drone defense systems to protect future facilities. That could increase construction costs but may make the region more viable for AI workloads over the long term.

The incident also stalled OpenAI’s plans to lease capacity in a 5-gigawatt UAE data center complex. According to the Wall Street Journal, no lease has been signed more than a year after the initial announcement. For AI builders that rely on AWS for GPU compute, this is one signal that large-scale capacity in the UAE may not materialize on the earlier expected timeline.

Caveats and what remains unclear

The evidence for the outage comes from third-party reporting and AWS dashboard updates rather than a detailed post-mortem from Amazon. AWS has not published a firm recovery schedule. The status of individual facilities, especially in Bahrain, remains uncertain. Iranian state media claimed the Bahrain data center was destroyed, but AWS has not confirmed that directly.

Additionally, the long-term impact on the UAE’s AI hub ambitions depends on whether hardened or underground data centers are actually built, which remains a proposal at this stage.

Summary

The AWS Abu Dhabi and Bahrain data centers outage is a generative AI infrastructure risk case study. For any AI builder, the takeaway is straightforward: diversify across regions before an external event forces you to do it under pressure. Use infrastructure-as-code to replicate environments, test cross-region failover regularly, and assume that any single availability zone can become a total loss with zero advance notice.