The GPUs are ready. The servers are racked. And somewhere, in some data center operator's spreadsheet, there's a line item that reads: "Utility interconnection: 5-6 years."
Which is how long it can take these days to get hooked up to the electrical grid—an eternity when you're racing to deploy AI infrastructure. Voxel Energy, a Y Combinator Winter 2026 graduate based in San Francisco, thinks it has a workaround. A blunt one, really: Don't wait for the grid at all.
The company's pitch, unveiled this winter, centers on modular data centers that run entirely off-grid—solar panels up top, repurposed electric vehicle batteries down below, zero utility tie-ins. "The best grid connection is no grid connection," goes the tagline. It's either audacious or obvious, depending on how long you've spent in an interconnection queue.
More intriguing, perhaps: Voxel says it's not just theorizing. According to the company's launch materials from January 29, a working prototype is already live and running compute off-grid. Whether that's a single rack or something closer to commercial scale, the startup hasn't specified.
Power Conversion, Rethought
What Voxel has detailed, in a white paper released earlier in the year, is an architecture that eliminates most of the electrical conversions that plague traditional data centers. The typical facility takes AC power from the grid and converts it to DC for servers—sometimes multiple times over, with energy losses stacking up at each step. Voxel's paper claims those inefficiencies can reach 30% in poorly designed setups.
The company's solution keeps everything DC from generation straight through to the server racks. Solar feeds directly into battery packs—second-life EV units, specifically—which then power the hardware without the usual conversion gymnastics. Losses, Voxel argues, drop to around 4%. Backup generators stand by to recharge batteries when the sun doesn't cooperate, aiming for what the company calls "365-day reliability."
It's a clever bit of engineering, assuming it scales. And the battery choice isn't arbitrary. By tapping retired EV packs that still have useful life left for stationary storage, Voxel threads a sustainability narrative while also addressing AI's notoriously spiky power demands. The white paper walks through discharge scenarios—a 175Ah, 350V pack delivering 184 kW during training surges—that go well beyond steady-state load.
There's a circular economy angle here, too, one that might appeal to hyperscalers looking to burnish their environmental credentials while also solving an immediate infrastructure problem.
The Founders
Casey Spencer, the CEO, spent time at Tesla managing Autopilot projects. Max Pfeiffer, the CTO, also logged years at Tesla and founded Maxwell Vehicles before that; Forbes put him on its 30 Under 30 list in Transportation & Mobility recently. Evan Schmidt, the COO, brings commercial architecture and data center construction experience—arguably the less glamorous but equally critical piece of the puzzle.
It's a founding team that seems purpose-built for this specific problem. Building modular, off-grid facilities at any kind of scale demands both battery domain expertise and the unglamorous ability to prefabricate, ship, and assemble physical infrastructure quickly. Whether this trio can execute on that remains the open question.
An Industry Scrambling for Alternatives
Voxel's emergence tracks with a broader reckoning in the data center world. Last fall, the U.S. Energy Secretary floated the idea that if the grid couldn't deliver power fast enough, facilities should "bring your own generation." By early this year, The Washington Post was reporting dozens of off-grid projects underway nationwide, many blending natural gas with renewables in closed-loop configurations.
Industry chatter has shifted noticeably. CoStar News noted in January that batteries and onsite generation were no longer fringe ideas but "legitimate paths" to sidestep interconnection delays. Separate analysis around the same time described batteries as "the only technology that can keep pace with AI data center power fluctuations," a nod to the sub-second transients that occur during model training.
Voxel isn't pioneering this space alone—though it may be the most aggressive on the DC-native front. Crusoe Energy, for instance, partnered with Redwood Materials (which jumped into energy storage for AI facilities last year) to deploy a 12 MW microgrid using second-life batteries for modular compute. ECL launched a "fuel-agnostic" platform around the same time, targeting hydrogen and multi-source configurations.
What distinguishes Voxel, at least on paper, is the commitment to DC end-to-end. In February, the startup announced a partnership with Sift for live operations monitoring and predictive maintenance at the cell level. Materials from Sift claim Voxel's design cuts total energy consumption by "up to 26%" simply by eliminating AC conversion losses.
That's a meaningful delta if it holds up in practice.
Moving Fast—But How Fast?
Voxel took the audience choice award at Y Combinator's Product Showcase in late January, just before going public with the launch. Since then, the company has said "hundreds (Update: thousands) of acres" are under contract and that it's accepting facility reservations.
The value proposition hinges on speed. Where grid-dependent projects stretch timelines into the multi-year range, Voxel promises deployment in "months, not years." For operators racing to capitalize on AI compute demand—where every idle GPU represents bleeding revenue—that compression could be worth significant trade-offs.
Pricing remains undisclosed. CB Insights lists what appears to be a $500,000 convertible note round from February, though this remains unconfirmed and Voxel hasn't disclosed any fundraising beyond Y Combinator. The company is still tiny—LinkedIn pegs it at somewhere between 2 and 10 employees—but the land commitments and operational prototype, however modest, suggest this isn't vaporware.
The white paper floats plans to deploy "hundreds of megawatts to power GPUs," an eye-popping ambition for a startup founded just last year. Whether Voxel can translate a working prototype into utility-scale infrastructure—without the utility—is another matter entirely.
But for now, the pitch is simple enough: Show up with your GPUs, bypass the interconnection queue, and let salvaged EV batteries and solar panels handle the rest. In an industry choking on infrastructure bottlenecks, simple might be exactly what sells.
