AWS won't restore two Gulf regions; multi-AZ design has a limit

Amazon Web Services will not restore service to the data centers it ran in the United Arab Emirates and Bahrain, and the reason it gave is a larger problem for the asset class than any single write-off: damage to the Bahrain region, the company said in a status update this week, spanned multiple availability zones and exceeded what its regional and multi-AZ services are designed to withstand; resources and data hosted exclusively there are unrecoverable. In the UAE, AWS has written off one zone, mec1-az2, while recovery work continues on mec1-az1 and mec1-az3.

The sequence matters for anyone underwriting Gulf capacity. AWS opened its Bahrain region in 2019 and its UAE region in 2022, and both were damaged in early March at the outbreak of the US-Israel bombing of Iran, when Iranian drones began striking targets in neighboring countries — Amazon's data centers among them. Iran's Islamic Revolutionary Guard Corps claimed a second attack on the Bahrain region in July, and several data centers inside Iran and an SES satellite ground station in Israel were also struck during the conflict, so the strikes reached commercial digital infrastructure across the region rather than one operator's footprint.

Multi-zone architecture is sold as a separation product: the buyer pays for zones that fail independently, so a fire, a substation fault, or a bad software push takes part of a footprint offline and leaves the rest serving. AWS's language makes plain that the design envelope in Bahrain did not cover this event. Independence is a physical claim about distance, power feeds, and terrain, and a campaign that can reach several zones inside one country defeats it. That argues no hardening inside a zone addresses a threat that arrives at the country level.

Hours are the unit of pain this industry sells against; AWS is now working in quarters, replacing the affected infrastructure and promising an update on restoration "in the coming months," with a plan for Bahrain due in early 2027. It says it has notified the relevant authorities and is working with them, and that it remains committed to customers in both countries. Read the dates together and a hyperscaler region can sit dark from early March through the end of the year, with the customer-visible recovery date set by a rebuild schedule rather than an engineering fix.

None of that has changed AWS's Gulf plans: it has already put a delivery date on the Saudi AI Zone, with the first 50MW by 2028 and the cloud region slated for a December 2026 launch, and in August it locked in two million more Nvidia GPUs through 2028. The next tranche of capacity goes into Saudi Arabia, next to a theater where AWS has just documented damage beyond its own design threshold.

Multi-AZ in the Gulf is not the same product as multi-AZ in Virginia, Dublin, or Oregon, and it should not carry the same price. Enterprises with data-residency rules or latency-sensitive customers will stay, because sovereignty and proximity do not negotiate. What changes is the continuity checkbox procurement uses to clear a cloud deployment: it now reads weaker in one geography than another, and that gap will steer where regulated workloads land. Hardening zones sit inside a hyperscaler's control, but making two zones in one country independent of a missile range does not. Gulf capacity is therefore likely to get built for in-region workloads at an accepted availability discount, rather than as a redundant tier of a global footprint.

As this publication has argued, consent is the priced commodity in data-center development — permitting, water, and local politics decide what gets built more than capex does. Sovereign risk belongs on the same list, and it is the part that cannot be permitted, paid off, or litigated. The house view that hyperscaler-anchored assets earn infrastructure pricing takes a knock here: two AWS regions sitting on one corporate balance sheet, in one conflict theater, delivered none of the diversification that pricing assumes. The market appears to have been assigning a single risk weight to hyperscaler geography it had not tested.

AWS will publish its Bahrain plan in early 2027 and update on the UAE before then, and the useful detail will not be the megawatts of replacement capacity. It will be whether the rebuild is architected and priced as a distinct availability tier, or whether the previous design returns and customers are asked to keep trusting zone independence within drone range. Rebuilding to the old specification would tell customers AWS has decided the past six months were an outlier.