Cloud didn’t kill the data center. It changed who owns it.
During a side conversation with colleagues at a recent AWS conference in San Francisco, we landed on a question:
What actually happened to the traditional data center?
Why buy racks of servers, size for peak demand, wait through procurement cycles, and carry hardware through years of depreciation when you can provision in minutes, scale with demand, and access newer technology without owning the refresh cycle?
But when we got into the economics, we agreed on something that sometimes gets lost in cloud discussions:
Not every workload becomes better simply because it moved to the cloud.
For some stable, predictable workloads, Total Cost of Ownership (TCO) can look very different after several years. Utilization, licensing, data movement, architecture, and existing investment all matter.
But cloud also buys something harder to capture in a spreadsheet:
Agility.
The ability to scale quickly, experiment without buying infrastructure, adopt managed services, and access newer compute without waiting for the next hardware refresh.
Recently, another version of the data-center conversation appeared in my social feeds.
One viral Instagram reel claimed a Meta data center in New Mexico was consuming roughly 75 million gallons of water annually and had “completely drained” the Rio Grande. Reel: https://lnkd.in/ej4M49Jv
The water concern is real. A July 2026 Congressional Research Service report notes that U.S. data centers directly used about 17 billion gallons of water in 2023, up from 5.6 billion in 2014.
But attributing the Rio Grande’s condition to one facility is a much bigger leap.
Still, the reel raised a different question:
When we talk about the environmental impact of data centers, where exactly do we think the cloud lives?
Every virtual machine, managed database, serverless function, AI workload, and API call depends on physical infrastructure somewhere.
Real buildings in real communities.
We moved infrastructure out of our buildings. We did not move it out of the physical world.
Cloud did not eliminate the data center. It abstracted it from most of us.
Large cloud and hyperscale facilities can operate more efficiently than many smaller or aging enterprise data centers. Uptime Institute’s 2026 data found newer and larger facilities routinely reporting PUEs of 1.3 or lower, versus an industry average of 1.52.
But concentration creates its own questions: power, water, community impact, geographic risk, and infrastructure required to support AI.
So perhaps the question was never really:
Cloud or data center?
A better question is:
Which infrastructure should we own, which should we consume as a service, and what trade-offs make sense over the life of the workload?
The physical infrastructure behind our digital world is not going anywhere.
#CloudComputing #DataCenters #CloudStrategy #Infrastructure #AIInfrastructure