
When War Hit the Cloud, Careem Moved Its Systems and Restored Service
Careem recovered after moving systems out of an affected cloud region. Months later, Amazon said it could not restore access to data held exclusively in Bahrain. The two accounts show how much digital resilience depends on the distance between copies of data.
Overnight, Careem’s teams moved systems between geographic cloud regions. That was how its chief executive, Mudassir Sheikha, described the response to disruption at Amazon Web Services in the United Arab Emirates. In his March 2026 update, he said services had been restored. Rides, Hala and onboarding for Careem Pay had been affected; food and delivery services had continued operating.
The company, whose app combines rides, deliveries and payments, had to restore some functions while others continued serving customers. Its chief executive reported a successful recovery. He did not report data loss.
The operation that brought those services back puts the location of data at the center of the story. Users do not see a change of cloud region on their screens. For the business running the service, however, that move can determine whether it must wait for local infrastructure to recover or can resume operations somewhere else. In this case, the chief executive described the migration as part of the recovery, rather than simply announcing that the disruption had ended.
Amazon subsequently published notices about prolonged damage in another Gulf region, Bahrain. On March 24, the company confirmed that the conflict had disrupted operations there. It said it was helping customers move to other regions and urged them to continue doing so. Many customers were already operating successfully elsewhere, the company said. The same statement described two different conditions: infrastructure that remained impaired and customers who no longer needed it to operate.
A September notice for Bahrain set out a sharper limit. AWS said it could not restore access to resources and data stored exclusively in that region. According to its published timeline, damage to an additional availability zone in April had left the entire region unavailable. The company described damage that exceeded the resilience of regional services and systems running across multiple availability zones.
The published scope of the damage differed between the two countries. In the UAE, AWS reported a recovery limitation affecting the availability zone identified as mec1-az2. In Bahrain, its notice concerned resources held exclusively within the geographic region. Careem reported restoring its UAE services; its account did not describe a loss in Bahrain.
Multiple Buildings Are Not Multiple Regions
The technical distinction does not require a map of servers. AWS divides its infrastructure into geographic regions and operates multiple availability zones within each. Each zone contains one or more data centers in separate facilities, with redundant power, networking and connectivity. Customers can distribute operations across those facilities so that a failure in one does not necessarily disable an entire application.
That spreads local risk. It reduces dependence on a single building, but it does not automatically move data to another country or continent. Several copies can remain inside the same geographic region. When widespread damage reaches more than one facility, the decisive question is whether an accessible copy and a working computing environment exist outside the affected area.
Having a backup is also different from restoring a service. An application needs data available to a functioning computing environment, and its systems must be able to use that data. Moving to another region is therefore an operational task: reconnecting the information with the capacity to serve customers. Careem’s statement describes the result it reported achieving. AWS’s notices describe the limits of recovery in different infrastructure.
That distinction also changes the meaning of a statement that data has not returned. AWS’s notice concerns its ability to restore access to affected resources. A customer with an independent copy elsewhere may be in a different position from one whose copies all remained in the affected region. A cloud provider’s notices are neither a customer-by-customer inventory nor an agreed accounting of each company’s business losses.
In September, AWS committed to another update in early 2027. It did not promise that every resource would be recovered by then. For a business trying to resume operations, the distinction matters: a timetable for receiving information from a supplier is not a timetable for restoring service to customers.
The Gulf disruptions showed that digital infrastructure remains connected to the physical world it can seem to transcend. Power, buildings and connectivity still occupy particular places. A service can continue elsewhere only if there is a way to establish it there. At Careem, according to its chief executive, that route was put to work overnight. In Bahrain, resources remained that AWS said it could not make accessible again. The distance between those outcomes is measured not only in the number of backups, but also in where they are kept.
Compiled November 16, 2026. Records and findings keep the dates stated in their sources; each link in the text opens the document it relies on.