The Cloud Went Down. Suddenly Everyone Misses Their Data Center.
This week's extended Microsoft 365 outage has generated a lot of frustration, particularly around Exchange Online. Understandably so. When core services are unavailable for an extended period, organizations can find themselves struggling to carry out their missions.
In the days since, I've seen a renewed conversation about whether we made a mistake moving so many of our core IT capabilities to the cloud. After all, an on-premises Exchange environment might have continued working through this particular incident.
I find that argument interesting because I remember being on the other side of it.
In the early 2010s, I routinely found myself arguing against the very popular push to move everything to the cloud.
I remember someone making the argument that organizations don't generate their own electricity. They buy power as a service. Why should IT be any different?
It seemed like such a silly comparison. If an organization's mission was important enough, it ABSOLUTELY had the ability to generate its own power when necessary. And many did.
Besides, IT services weren't, and still aren't, nearly as interchangeable as electricity. You couldn't simply flip a switch and move your Exchange environment back and forth between your own data center and Microsoft's.
For me, the concern was always about giving up control. When your infrastructure lived in your data center, you controlled your own destiny. When something went wrong, you had the people, equipment, and access necessary to do something about it.
Moving to the cloud meant surrendering a considerable amount of that control to someone else.
What I didn't anticipate was just how reliable those cloud services would become.
Looking back, I could never have predicted that Microsoft 365 would prove as reliable as it has.
Consider what Microsoft is delivering today. An enormous, globally distributed collection of interconnected services supporting organizations of virtually every size, with an incredible range of capabilities that continue to evolve.
And somehow, the entire thing operates with a degree of reliability that we've come to expect from municipal utilities.
That's a pretty remarkable accomplishment.
The numbers are worth considering
Microsoft has published worldwide Office 365 and Microsoft 365 availability statistics for years. Looking at the decade from 2016 through 2025, the average of its reported quarterly availability figures was approximately 99.98%.
That's an impressive track record for a service operating at this scale.
It hasn't been perfect, of course. The first quarter of 2026 was particularly rough, with Microsoft reporting just 99.526% availability. And this week's outage certainly isn't going to help its reputation.
But focusing exclusively on the worst incidents without considering the broader record doesn't give us a particularly useful picture of reliability.
Things break.
Have we forgotten what running our own infrastructure was like?
If you were responsible for core enterprise infrastructure before the cloud became the default, my bet is that you've spent more than a few overnighters in a cold data center trying to make sure that, when morning came, the organization could actually carry out its mission.
I certainly have.
Failed storage systems. Misbehaving servers. Network problems. Updates that went sideways. The occasional outage where the apparent cause turned out to be something completely different.
Those were high-stress, high-stakes, low-sleep scenarios, and they were hardly unique to any particular organization.
And while having direct control over the infrastructure meant we could do something about those problems, it also meant we were entirely responsible for fixing them.
There's a reason organizations invested so heavily in redundancy, backup power, disaster recovery facilities, and continuity planning... and it was all very expensive.
You're fooling yourself if you think it's all rainbows and uptime on-premises.
Control and reliability aren't the same thing
I still believe that control matters.
There are absolutely circumstances where organizations need to maintain independently operated capabilities. Mission requirements, disconnected operations, regulatory obligations, and continuity needs can make that a perfectly reasonable architectural decision.
And yes, an appropriately designed on-premises environment might have continued functioning during this week's outage, at least for users and systems that didn't depend on the affected cloud services.
But surviving one particular failure doesn't automatically make an architecture more reliable.
An on-premises Exchange environment introduces its own dependencies, including power, networking, storage, security, maintenance, personnel, and potentially external services needed for mail delivery.
Those risks don't disappear simply because the servers are sitting in a room you control.
The better question isn't whether we should move everything back to our own data centers.
It's whether we've properly accounted for the operational dependencies we've created, and whether our continuity plans are adequate for the missions we're supporting.
Someone is always going to get the blame
None of this means Microsoft should get a free pass.
Customers should absolutely hold Microsoft and other cloud providers accountable for their service commitments. Extended outages deserve thorough explanations, meaningful corrective actions, and appropriate scrutiny.
Organizations should also be investing in effective continuity-of-operations plans that acknowledge the possibility of extended SaaS outages.
But the idea that this week's calamity somehow proves we'd all be better off running our own Exchange servers again is ridiculous.
The reality is that Microsoft has demonstrated a level of reliability, at an extraordinary scale, that I suspect most organizations would struggle to reproduce independently.
I certainly wouldn't have expected that 15 years ago.
The cloud hasn't eliminated outages, and it hasn't eliminated our responsibility to plan for them. It has, however, changed who's sitting in the server room at 4am trying to restore service when things go sideways.
And if your alternative is running everything yourself, just remember:
People don't stop throwing rocks when things go down if your services are on prem... they just start flying at you instead.