Leasing Time Again: How Azure Made Me Think of Mainframes

The first time I logged into the Azure Management Portal in early 2011 and spun up a SQL Azure instance, I had a strange feeling of déjà vu — despite being about two decades too young to have lived through the original. I was leasing time on someone else's computer. Again.

It took me a few days to figure out why it felt familiar. Then it clicked: this is exactly what computing looked like before the personal computer. IBM mainframes, time-sharing systems, punch cards submitted to an operator and results collected hours later. Companies didn't own their compute — they rented access to it, metered by the minute. The revolution of the 1980s was that you could finally own the hardware. The revolution of 2010 is that you can stop wanting to.

The Closet Server Era

Walk into almost any small business office in 2008 and you'll find the same thing: a repurposed closet, or a corner of the accounting department, with a tower server — maybe two if they were serious — a UPS, a switch, and a broadband line that cost $400/month because it was "business class." That was the data center. That was the disaster recovery plan. That was also where the heat went in summer.

The uptime story was generous. The "backup" story was usually a USB drive someone took home on Fridays. High availability meant hoping nothing broke on a weekend. A second server for failover wasn't even in the conversation — that was enterprise money, not small-business money.

I spent a fair amount of time in those closets. And every time I came back to it, the question was the same: is this really the right answer?

Azure Goes Generally Available

Microsoft announced Azure's general availability in February 2010. SQL Azure followed shortly after. By early 2011, the pricing had settled enough to do real math on it, and that math was more interesting than I expected.

A small business paying $400/month for a business-class broadband line — plus the $3,000 server they bought three years ago, plus the time someone spent patching it — was already spending real money. Azure SQL Database started at a few dollars a month for a Basic tier that handled what most small business line-of-business databases actually needed. Scale up only when you need to. No hardware refresh cycle. No cable replacement when the cat chews through something in the closet.

But the cost comparison almost misses the point. The real story is what you got in exchange.

High Availability: No Longer Enterprise-Only

Azure SQL Database came with built-in redundancy. Three replicas. Automatic failover. The kind of setup that would have required a secondary data center, a SAN, SQL Server Enterprise licensing, and a consultant (hi) to configure AlwaysOn — all for a small business that needed their invoicing software to stay up.

That math was genuinely different. Microsoft had already spent billions building out the data centers, the networking, the redundant power. They could amortize that cost across millions of customers. A small business was suddenly buying access to infrastructure at a scale they could never have justified owning outright.

This is exactly the mainframe model. IBM built the machine. You rented the cycles. The economics work because the fixed costs get spread across the entire customer base, and the marginal cost of adding another tenant is low. What changed in 2010 is that the "mainframe" is now distributed across data centers on three continents, and the "tenant" is a 12-person accounting firm in Tulsa.

What We Actually Moved

The first Azure migration I ran for a small client in early 2011 moved their SQL Server instance out of that closet and into SQL Azure. The broadband bill dropped from $400/month to $60/month — they no longer needed the business-class uplink to keep the server connected. The Azure costs were about $80/month for the database tier they needed. Net result: they were saving money and running on infrastructure that wouldn't go offline if someone tripped over the power strip.

The kicker: their "backup strategy" before this was the USB drive. After the migration, they had point-in-time restore going back 7 days, built in, no configuration required. That's not a feature they asked for. That's a baseline they got because it's what Azure includes.

The Thing About Leased Compute

The philosophical shift matters beyond the economics. When you own the hardware, you think about capacity in terms of what's in the rack. When you lease compute, you think about capacity in terms of what the problem needs. Those are very different mental models, and the second one is more honest.

Most on-prem servers were sized for peak load — which meant they were significantly underutilized most of the time. You bought the hardware for Black Friday and ran it idle the other 364 days. Leased compute lets you pay for what you use and scale when you need it. The mainframe billing departments figured this out in 1967. We're just rediscovering it with better networking.

There are tradeoffs, of course. You don't control the hardware. You're dependent on the vendor's uptime, their pricing decisions, their API stability. SQL Azure in 2011 had real limitations — no cross-database queries, no SQL Agent, a connection model that punished you for keeping connections open too long. Those gaps mattered and I'll get into them. But the fundamental economics were sound, and the trajectory was clear.

We were leasing compute again. The wheel had turned. And for the clients I was working with, it was exactly the right time to start paying attention to what that meant.

If you made this move early — or if you watched from the sidelines while trying to justify it — I'd love to hear how it landed. As always, I'm here to help.