Building off-grid AI data centers to bypass the grid wait
Whoever solves the grid, wins the century. - Chamath Palihapitiya.
This line reframes today's AI arms race. The race underneath the models and chips is energy. AI scales on compute, and compute scales on electrons. The US grid, 6 million miles long and powering 340 million people, is the largest machine in the world, and it is no longer keeping up. Transfer capacity between the US eastern and western grids is roughly 1.3 gigawatts, about what a single large AI data center now draws on its own.
Meanwhile, an estimated $3 trillion of AI data centers are slated for construction by 2030, and a meaningful share are unlikely to come online on schedule because the grid cannot power them. The national average wait for a 10MW grid connection is around 5 years. On the West Coast, closer to 7.
The constraint isn't only physical. Policy and public sentiment are tightening in parallel. At the state level, Maine recently passed legislation restricting new data center construction, and similar resistance is surfacing at the state and municipal level elsewhere. At the federal level, the administration's Ratepayer Protection Pledge asks data center operators to bring their own power or remain net-neutral to the grid, reflecting voter concern that AI-driven demand could raise residential electricity prices. The direction is consistent: new compute capacity is increasingly expected to source its own generation rather than draw from shared infrastructure.
Voxel Energy is exactly that model. Bring-your-own-power infrastructure for AI data centers.
Voxel is building off-grid AI data centers, powered by solar and large-scale battery storage, designed to go live behind the meter on a timeline of months rather than years. The target customers are neoclouds, colos, and enterprises that need compute capacity sooner than a grid connection allows. Voxel owns the energy infrastructure and sells power by consumption, with excess solar sold back to the grid as a secondary revenue stream.
Voxel's founders have spent most of their careers inside adjacent problems. Casey Spencer was a product manager on Tesla Autopilot and helped oversee the design and build of Hardware 3. Max Pfeiffer worked on Tesla Semi before starting an EV hardware company that he bootstrapped into a meaningful business. Evan Schmidt came out of McDonnell & Miller, where he ran data center builds from site selection through handoff. The three first met through a Tesla hacking community in Berkeley and have been working together for five years.
The grid is the bottleneck
US energy policy over the past two decades has been framed primarily around emissions. Within a decade of that shift, high-voltage transmission construction had fallen by roughly 97%. New solar projects face litigation in close to two-thirds of cases. About 70% of transformers and transmission lines are more than 25 years old, and the modernization gap is estimated at $578 billion.
Demand is moving in the opposite direction. The US will need roughly ten times the power over the next 15 years that was installed in the last 15. In 2025, both Microsoft and Google paused US data center expansions after community opposition. Supply is constrained, permitting is slow, and the delta is widening.
Hyperscalers have responded by bypassing the grid. xAI's Colossus projects in Memphis are expected to exceed a gigawatt of demand behind the meter, on dedicated off-grid generation. Others are following the same pattern. Collocation eliminates transmission losses, compresses deployment from years to months, and sidesteps the permitting queue.
Voxel is building that pattern as infrastructure others can buy, rather than something each operator has to build from scratch.
Where the economics come from
Two factors shape the unit economics.
Battery supply chain. Voxel sources large-scale storage through a proprietary supply chain developed over several years, at a significant discount to new cells purchased off the market. The relationships, engineering work, and certifications involved are not straightforward to replicate. The cost basis carries through to every deployment.
Energy architecture. Most data center designs inherit inefficiencies from assumptions about grid compatibility. Voxel is vertically integrated around the physics of how solar is generated, how storage discharges, and how GPUs draw power, which removes a category of conversion loss that the rest of the industry accepts as standard.
Taken together: a cost structure below grid-tied alternatives, an all-domestic supply chain that qualifies for ITC credits, and a deployment model aimed at months rather than years. A working prototype is operational in LA and is planned to run a customer's GPU cluster during YC Demo Day. A pipeline conversation with a major neocloud is under NDA, with staged capacity discussed in the tens of megawatts.
Why we're investing
Voxel is a full-stack infrastructure company. The supply chain was built over years rather than bought, and the team has operated at each layer of the stack before.
The thesis is straightforward. Energy is becoming a gating factor for AI at scale. Hyperscalers will continue to build their own off-grid capacity. Everyone else, including neoclouds, colocation providers, and enterprises running on-prem compute, will need a provider that can deploy quickly and at a competitive cost structure. Voxel is positioned to be that provider.
We're backing Casey, Max, and Evan to support their vision.
