Key Takeaways

  • Mavria raised a $30 million pre-seed from Nat Friedman and a $100 million subsequent round to mass-produce modular data centers called Terrain in Houston.
  • Each Terrain unit integrates power, closed-loop cooling, and structural housing into a standardized 5-megawatt block built entirely off-site.
  • Using closed loops eliminates municipal water permits: operators truck in water once to fill the system, after which ongoing water consumption is negligible.
  • Connecting directly to local electrical distribution lines with batteries and curtailment terms lets operators bypass multi-year transmission substation buildouts.
  • Adam Blake estimates over 100 gigawatts of curtailed power sit idle across thousands of existing US distribution sites.

The AI Bottleneck Moved From Code to Concrete

The consensus in Silicon Valley treats compute shortages as a silicon supply issue or an algorithmic hurdle. Blake views it as a basic civil engineering trap. Hyperscalers and neoclouds design massive custom campuses that require years of zoning reviews, bespoke concrete pours, and regional transmission upgrades.

Blake founded Mavria to turn that construction process into an assembly line. As Blake put it: “We started Mavria because we believe the next phase of the AI race is going to be won or lost in the physical world and America's leading today in AI because of our technology and our people but our physical infrastructure isn't keeping up. So that's the gap that we're planning to close.” (play from 1:36:43)

Instead of pouring concrete foundations for two years before racking a single server, Mavria builds standardized, self-contained units in a dedicated Houston factory. The product, named Terrain, packs structural enclosures, power conditioning, environmental controls, and cooling systems into a modular five-megawatt footprint. As Blake explained, “Terrain is a manufactured data center that integrates cooling, power, controls, and a structure into a repeatable five megawatt block.” (play from 1:37:05)

Skip the Substation by Tapping the Distribution Grid

The standard path to 100 megawatts requires an interconnection study, a brand-new high-voltage substation, and years of waiting in utility queues. Blake sidesteps transmission queues by sizing units down to the distribution level. Local distribution grids cannot handle a 500-megawatt mega-campus, but they frequently have spare capacity in 5-megawatt increments spread across suburban and rural nodes.

To capture that stranded power, Mavria pairs its modular units with on-site batteries and curtailment agreements. When local peak demand surges on the local grid, the data center throttles power or runs on battery reserves rather than crashing the neighborhood circuit. Blake estimates this unlocked reserve is massive: “There are thousands of sites across the US that exist on existing distribution infrastructure. I mean, there's probably 100 plus gigawatts that if you could use batteries and agree to curtailment that you could deploy our product on existing infrastructure.” (play from 1:42:20)

The other hurdle to rapid site deployment is water access. Traditional cooling towers evaporate millions of gallons of water daily, triggering local environmental pushback and lengthy municipal utility reviews. Terrain circumvents this with closed-loop liquid cooling. As Blake noted: “When you use liquid cooling, you don't really need on-site water. You can truck it in to initially fill the closed loops and then anything you need on ongoing basis is pretty immaterial.” (play from 1:39:29) You haul water in by tanker once, fill the chillers, and seal them.

What to Do With This

Audit your infrastructure dependencies this week to see if you are waiting on a centralized utility permission slip that a decentralized setup bypasses. Map your current power, hosting, or real estate queues: identify where accepting a curtailment trade-off or bringing self-contained resources cuts your deployment delay from three years to three months.