As the race to power data centers intensifies, Ralph Alexander thinks we aren’t focusing on the right problem. In his metaphor, it’s like we need more light in a world with only incandescent lightbulbs. But instead of focusing on inventing the halogen, we’re building more and bigger incandescent ones.
“This rush to build data centers has been about doing what we do now…and that’s not the right answer here,” he said. “It doesn’t solve the basic inefficiency of the current data center, where so little of the actual power gets to the GPUs that are doing the processing.”
Alexander — who previously worked as CEO of Talen Energy and CEO of BP’s global gas, power, and renewables unit — is the co-founder of GridFree AI, which was incubated my venture studio and fund Montauk Climate and emerged from stealth earlier this week with $5 million in funding led by Giant Ventures. His co-founder Patrick Yantz was most recently global leader of edge site operations at Microsoft.
The company is engineering a modular data center infrastructure solution that integrates grid-independent gas power generation, battery storage, and cooling. Their plan is to improve the energy efficiency of data centers while simplifying construction and maintenance, via what they describe as a “Lego set” data center.
A ‘Lego set’ data center
The idea is to select the most efficient generation, storage, and cooling technologies in circulation and build a centralized manufacturing facility for easily replicable “Lego sets.” These are then transported via truck, barge, or rail, and assembled in a location with abundant fossil gas resources and fiber connection. A customer would only have to sign the lease and show up with its racks of GPUs, “literally like leasing a flat,” Alexander said.
In contrast, the current standard approach involves data center operators handling complicated logistics, like building cooling systems for instance, themselves.
GridFree AI estimates that this integrated approach would increase the amount of power available for the GPUs by about 50% — while reducing costs by about one-third through the elimination of legacy infrastructure.
“We could deliver so much more power to the GPUs, that in the same box, [the customers] can get bigger chips,” Alexander said. “They can focus on the things inside that box and we’ll focus on everything else. We’ll deal with landowners, we’ll deal with gas acquisition, we’ll deal with everything else.”
Additionally, GridFree AI’s infrastructure would eliminate the need for grid interconnection, a process that is only getting longer and more complicated. The interconnection issue is pushing data center operators to look for similar off-grid solutions like microgrids, as wait times for a connection stretch as long as ten years in some regions.
Even if interconnection isn’t a necessity for GridFree AI’s systems, Alexander said they would “probably apply to the grid anyway,” if only to keep options open.
“It doesn’t hurt, right?” Alexander said. “Even if it takes three, four, or five years. Who knows what the world will look like in five years, especially in this industry.”
Space for change
The speed of change is central to GridFree AI’s approach. The modular system is technology-agnostic, so it could also be used with nuclear SMRs if and when they’re ready to be deployed at scale. Delivering more power to the GPUs could also allow hyperscalers to upgrade their chips without having to upgrade the whole infrastructure. And modularity means that the systems can be progressively scaled to “make it as big as anybody wants,” Alexander said.
GridFree AI has selected some initial sites with friendly regulations, gas resources, and fiber. They’re clustered mostly in the Southeast, in areas around Houston, Dallas, and New Orleans, among others. The company is currently in conversation with potential customers but doesn’t yet have a target date for its first deployments.
The biggest challenge moving forward, according to Alexander, is going to be convincing customers that GridFree AI’s system can work efficiently. The company has prioritized making sure the system “was pretty buttoned up so that we had all the answers,” and it has only recently begun talking to potential customers. “There’s this question: ‘Can you really do that?” Alexander said. “And believe me, it crosses my mind too. Let’s make darn sure.”
