AI Data Centers Are Going Nuclear. That's Directly Connected to the Cost of Storing Your Photos.

Aalo Atomics hit criticality on a test reactor at Idaho National Laboratory on July 4, Valar Atomics and Nvidia are exploring a nuclear-powered data center in Utah, and total committed nuclear capacity for AI data centers has now passed 9.8 gigawatts across the major hyperscalers. This is being covered almost entirely as an AI-compute story — which it is — but a meaningful and less-discussed slice of that demand isn't training runs at all. It's storage. Specifically, the enormous and growing volume of full-resolution photos and videos sitting untouched in cloud accounts, backed up in duplicate, and never compressed.

What Actually Happened This Month

Three separate nuclear startups — Aalo Atomics, Valar Atomics, and Antares Nuclear — hit criticality milestones ahead of a Department of Energy deadline this July, each under different pilot and demonstration programs. Valar's Ward250 reactor went critical on June 18 and is now generating initial power for a proposed Utah facility that would pair it with Nvidia's water-efficient cooling systems. This isn't speculative anymore; these are functioning reactors tied to specific, named data center proposals, not white papers about a hypothetical future grid.

Why Storage Is the Boring Part of This Story That Actually Matters to You

AI training runs are a spike — enormous energy demand for a finite period, then done. Storage doesn't work that way. A photo uploaded to cloud backup in 2020 that's still sitting there, full resolution, un-deleted, un-compressed, is drawing power and cooling continuously, year after year, on servers that never get to power down for that file. Multiply that by however many billions of duplicate, oversized, rarely-viewed photos sit in personal and business cloud accounts worldwide, and storage becomes a genuinely large, steady, and largely invisible piece of data center electricity demand — not a dramatic AI training headline, just a constant background draw that never really goes away.

The Uncomfortable Overlap With the AI Boom

Some of the same hyperscalers building nuclear-powered data centers for AI compute are also the ones storing consumer photo libraries at massive scale. The infrastructure is genuinely shared in many cases — the same facilities and power contracts covering both AI workloads and the ever-growing pile of full-resolution camera backups nobody ever goes back and compresses. That's not a reason to feel guilty about backing up your photos. It is a reason to notice that keeping a 12MB phone photo at full resolution forever, when a properly compressed version would look identical on every screen you'll ever view it on, has a real and compounding infrastructure cost behind it — one that's now literally being met with new nuclear reactors.

What Actually Helps, at Any Scale

You're not going to change hyperscaler energy policy by compressing your camera roll. But at the scale of a business managing thousands of product photos, client galleries, or archival images, the difference between storing everything at full camera-original size versus a properly compressed version is a real, multiplying difference in storage footprint — and by extension, in the electricity and cooling load behind it.

Optimage's compressor cuts a photo's file size dramatically with no visible quality loss, free, in the browser — the simplest lever available for reducing your own contribution to that footprint.

The Bottom Line

Nuclear-powered data centers are being built because AI compute demand is real and growing fast. Storage — unglamorous, unphotographed, easy to ignore — is quietly part of the same demand curve. Compressing what you keep isn't going to solve the grid problem, but it's one of the only parts of this story an individual or a small business actually has direct control over.

Related reading:

• Why Your LCP Is Failing and How to Fix It — the performance case for compression, separate from the energy case
• Serverless Image Optimization: Edge Functions 2026 — how modern infrastructure handles image processing at scale
• The Future of AV1 and Beyond — smaller files as a continuing industry direction, for storage as much as bandwidth