AWS would prefer to forget March ever happened in its UAE region
Cloud giant waives an entire month of charges, then erases the billing data. There is literally nothing to see here.
I received an email / billing notification from AWS this week that may be the most diplomatically crafted communication in the history of cloud computing. Here it is, stripped of the usual boilerplate around it:
"AWS is waiving all usage-related charges in the ME-CENTRAL-1 Region for March 2026. This waiver applies automatically to your account(s), and no action is required from you."
No explanation. No mention of the Iranian drone strikes that physically destroyed two of three availability zones in the region on March 1st. No reference to the 109 services that went down, nor the customers who spent weeks unable to terminate EC2 instances via the console because the control plane was as dead as the hardware underneath it. No acknowledgment that an entire month of cloud infrastructure effectively ceased to exist. Not even a link to their remarkably short (presumably because it wasn't insulting the Financial Times' reporting) corporate blog post explaining that you probably shouldn't expect that region to be working reliably again any time soon.
Just: we're waiving the charges. You're welcome. Move along.
I want to be clear: I have no problem with this. It's a tough situation, and it's not AWS' fault, given that there is not yet an Amazon standing military force.
But here's the part that caught my attention. The email continues: "You will not see any March 2026 usage for the ME-CENTRAL-1 Region in your Cost and Usage Report or Cost Explorer once processing is complete."
They're not just waiving customer charges for a month; they're erasing the billing and inventory data!
The bill is the only inventory you've got
For most organizations, the AWS bill isn't just an invoice. It's the canonical record of what infrastructure exists, where it's running, and how long it's been there. The Cost and Usage Report (CUR) is the closest thing many companies have to a single source of truth that accurately describes their cloud footprint. Even the relatively-recently-launched AWS Resource Explorer, which purports to be an inventory service, is missing some key resource types. The one thing that lists everything remains the AWS bill; a single source of truth around what the hell's living in your cloud environment.
Compliance teams rely on it. Auditors request it. FinOps teams build their entire practice on it. When your security team wants to know if you had resources running in the Middle East in March beyond what CloudTrail says, they'll check the CUR. And as of March 31st, when they ask Amazon’s answer will be: No. You didn't. Nothing was there. A clean, empty month.
The infrastructure existed, kinda. It ran, or at least it sure tried to. Customers were billed for resources they couldn't turn off because the API was unreachable, because nothing works properly when you have two of three availability zones non-responsive or flat out destroyed. And now the record of all of that is being zeroed out.
In fairness, nobody plans for this
Before anyone starts composing their "AWS should have been prepared for this" hot take: No. Stop.
AWS engineers availability zones for certain classes of failure. Hardware dies, power goes out, a fiber gets cut by its natural predator "The Mighty Backhoe," a cooling system fails, or the like? AZs handle these. They're designed so that a problem in one zone doesn't cascade to another. That model works. It's been tested repeatedly. It's the backbone of every well-architected deployment on AWS. Know it or not, virtually every AWS customer's data recovery strategy depends upon this being true.
AZs are not designed for a state actor deciding the data center shouldn't exist anymore.
Nobody's fault-tree analysis includes "building hit by military drone." No disaster recovery runbook in the world has a section titled "Regional Armed Conflict."
And honestly? It shouldn't. Try putting that into your company data recovery plan, and you'll get uninvited from business continuity planning faster than I was at investment firm BlackRock when I pointed out that if both San Francisco and us-east-1 were unavailable, absolutely nobody was going to care about our roboadvisor for the foreseeable future, if ever. (I can't imagine why I was let go from that job.) The entire premise of cloud infrastructure is that you're renting capacity from someone who handles the physical concerns so you can focus on your software. That contract didn't account for the data center becoming a military target, and it's hard to blame anyone this side of the Pentagon for not having that particular item on the risk register.
So why erase the data?
I can think of two explanations, and they're not mutually exclusive.
First, the data might not exist to provide. When multiple AZs go offline because the physical hardware is melted and TLS now terminates "on the floor," the metering and billing infrastructure goes with it. You can't generate a usage report for resources running on servers that no longer have power, or physical cohesion. AWS may genuinely not have reliable telemetry for what was running, what stopped, and when. You can't charge for what you can't measure – and you can't report what you didn't record.
Second, pragmatism. Customers couldn't terminate resources with the AWS console because termination commands require acknowledgement; they're not hurled over the fence like a dead dog for your neighbor to worry about. EC2 instances were stuck in stopped states for weeks. Elastic IPs were trapped on dead hardware. If AWS had tried to bill normally and then process refunds, it would have meant thousands upon thousands of individual support tickets, each one a paper trail documenting precisely what failed and when. Life would have become absolute hell for the hardworking folks at AWS Support. A blanket waiver with a clean data slate is operationally simpler and diplomatically tidier in a single move.
I suspect it's both.
The precedent is the point
Whatever the underlying reason, AWS made the right call. Customers shouldn't pay for resources they couldn't control in a region that was physically destroyed by military action. The waiver is the correct outcome.
But.
The mechanism now exists, and it's been exercised. AWS can zero out an entire month of usage data across a region, and the result is indistinguishable from that month never happening. The bill, the one canonical source of truth, simply won't show it.
Today the reason is a war, and it's a good reason. But precedents don't come with permanent context attached. The next time a month of data goes missing from CUR, will there be an email? Will there be an explanation? Will there be a *reason* you'd accept?
I don't know. AWS isn't saying. They probably can't say much. And after March 31st, the data won't be saying anything either.