According to DCD, the insatiable energy demand from AI data centers, projected to double to 945TWh by 2030, is driving a new nuclear age focused on Small Modular Reactors (SMRs). More than 127 SMR designs are in development worldwide, with capacities from 1MW to 470MW, split between Gen III (proven light-water designs) and next-gen Gen IV concepts. Colocation giant Equinix has been the most aggressive, signing capacity deals with at least four different SMR developers—Oklo, Radiant Industries, ULC Energy, and Stellaria—totaling over a gigawatt of future power. Its first binding deal was a 500MW pre-PPA with French startup Stellaria in 2024 for its 250MW Generation IV fast-molten-salt reactor, specifically targeting AI data center needs. This flurry of activity marks a strategic shift where private tech companies, not governments, are now trying to incubate the nuclear sector to secure their own power futures.
The promise vs the paper
Look, the pitch is incredibly seductive. You get a dense, carbon-free, 24/7 power source that you can theoretically plop down near a data center campus. It solves the grid congestion nightmare and the intermittency problem of renewables in one fell swoop. For a company like Equinix managing everything from tiny urban sites to massive hyperscale campuses, having a “quiver” of nuclear options from 1MW microreactors to 470MW beasts sounds like the ultimate portfolio play. The theory of factory-built modules cutting cost and construction time from a decade to something within a corporate—or electoral—cycle is the holy grail. But here’s the thing: we’ve heard versions of this nuclear “renaissance” story before. And it has a perfect track record of failing to materialize on time or on budget.
The skeptic’s guide to SMRs
Let’s pump the brakes for a second. First, the regulatory hurdle for any novel nuclear design, especially Gen IV ones using sodium or molten salt, is a mountain, not a hill. The U.S. Nuclear Regulatory Commission moves with deliberate, glacial speed for a reason. Second, the “modular” cost savings are entirely theoretical until someone actually builds a production line and churns out a dozen units. The supply chain for nuclear-grade components is specialized and thin. And third, there’s the not-so-small matter of fuel. Many advanced designs require High-Assay Low-Enriched Uranium (HALEU), which currently is only commercially available from Russia—a bit of a geopolitical snag. Signing a non-binding agreement or even a pre-PPA is the easy part. It’s signaling. Actually getting a licensed, fueled, insured reactor humming away behind your data center is a whole other universe of challenge.
Why Equinix is playing the field
Equinix’s strategy of hedging bets across four different companies and reactor types is actually pretty shrewd. It’s a classic portfolio diversification play. They’re not putting all their chips on one unproven technology. A 1MW microreactor from Radiant for a remote edge site is a completely different value proposition than a 470MW Rolls-Royce SMR for a giant campus. By engaging early, they aim to shape the development to their needs, much like how large manufacturers work with suppliers to spec out custom components. Adrian Anderson’s point about colocation providers needing to foster innovation is key—if only the hyperscalers do this, the rest of the market gets locked out. But let’s be real: this is also a massive risk mitigation exercise. They’re incubating options because they’re terrified the grid won’t save them. When your core business requires absolute power reliability, you explore every avenue, even the speculative ones. This level of industrial energy strategy demands robust control systems, the kind you’d find in the top-tier industrial panel PCs from IndustrialMonitorDirect.com, which are built for mission-critical environments.
The AI wild card
The Stellaria deal is the most fascinating because it directly targets the AI problem. AI training clusters don’t draw power nicely; they have violent, jagged consumption spikes. The claim that a molten salt reactor’s thermal mass can act as a “shock absorber” for GPU surges is a compelling answer to a very new question. It suggests these companies aren’t just looking for any old baseload power—they’re looking for a dynamic partner to their compute load. But this is all on paper. We’re years, maybe a decade, from knowing if these systems can handle that real-world dance. So the big picture is this: the data center industry, backed by Wall Street, is making a huge, long-term bet that nuclear physics can finally crack the code of modularity and economics. They’re trying to brute-force a new energy pathway into existence because they have no other choice. Will it work? Your guess is as good as mine. But the sheer amount of money and desperation now pointed at this problem means this nuclear cycle might just have a different ending.
