TAR's $27M Bet: Grid-Independent Power for the AI Era

Pat Becker watched data center developers wait five years for a grid connection — and decided there had to be a faster way.

The power bottleneck nobody planned for

Every data center developer in America faces the same wall: interconnection queues. Requesting grid power for a new facility today means waiting three to seven years in most regions. PJM, the largest US grid operator, reported 2,200 active interconnection requests in its queue as of early 2026 — most from data center and AI infrastructure projects. ERCOT's backlog is similarly stretched.

This is not a transmission problem. It is a timing problem. The grid was never designed to absorb gigawatt-scale loads on a quarterly basis.

AI training clusters and inference farms draw power densities that would have seemed absurd five years ago. A single Nvidia DGX SuperPod consumes roughly 500kW. A cluster of eight — not unusual for a mid-tier neocloud — burns through 4MW continuously. Multiply that by the hundreds of facilities under development, and the gap between demand and available interconnection capacity becomes a structural constraint on the industry.

The company's bet is that data centers will pay a premium for speed. Its systems are not cheaper than grid power. Co-founder Pat Becker is explicit about this. But a data center that can go live in 2027 instead of 2030 captures three additional years of AI workload revenue — and that math shifts the unit economics decisively.

From problem to product

The company was founded in 2024 with a simple thesis: the data center industry needs a parallel energy infrastructure, not a faster connection to the existing one. The company spent 18 months designing modular skids that integrate solar arrays, battery storage, wind turbines, and simple-cycle gas turbines into a single factory-built unit. Each skid is pre-wired, pre-assembled, pre-tested, and pre-commissioned before it ever reaches the site.

The approach mirrors the standardization that hyperscalers brought to server design. Just as a server rack arrives ready to plug into a network, a power skid from the startup arrives ready to plug into a data center's electrical bus. Site work shrinks from years to weeks.

We focus on shifting work to factories where we can pre-wire, pre-assemble, pre-test, and pre-commission energy generation methods significantly faster and cheaper. Each power generation method requires changes to the deployment sequence.— Pat Becker, co-founder, TAR

The 10MW pilot site, which the company plans to commission this year, will include the full stack: solar, behind-the-meter batteries, and gas turbines sized for backup duty during extended low-renewable periods. Becker told Forbes that the nameplate capacity is significantly higher than 10MW — the lower figure reflects renewable intermittency rather than hardware constraints.

The grid paradox

Behind-the-meter generation is not new. Hospitals, military bases, and remote industrial facilities have run on microgrids for decades. What changed is the scale. Its first customer — a large neocloud provider — will deploy roughly twice the capacity of the pilot. The company's pipeline for 2027 exceeds 200MW, with "several additional gigawatts" penciled in for 2028.

This trajectory suggests a structural shift. If behind-the-meter becomes the default deployment model for AI data centers, the implications extend beyond the industry itself. Utilities that expected data center load to anchor their grid expansion plans will find that load never arrives. Interconnection queues, already strained, may cool as the highest-value customers self-select out.

Regulators are watching. FERC has already proposed rules for behind-the-meter generation at data centers, and PJM requested approval for a 50MW threshold that would subject larger behind-the-meter facilities to transmission tariffs. The regulatory framework is being written in real time, and The company's model — combining renewables with gas backup — will test where the line between on-site generation and utility service falls.

Why seed-stage infrastructure?

Infrastructure startups rarely raise seed rounds. Hardware is capital-intensive, and VCs prefer software margins. The $27 million round — sizable for a seed stage — signals that investors see a window. AI demand is accelerating faster than the grid can adapt, and the gap creates an opening that traditional energy project developers are too slow to fill.

A project like a 200MW solar farm requires three to five years for permitting, financing, and construction. A behind-the-meter system serving the same load can be permitted as on-site generation, financed through equipment leases or power purchase agreements, and deployed in months. The asymmetry is the opportunity.

Several competitors are emerging. Hitachi and X Labs announced a partnership to deliver behind-the-meter energy parks for data centers. ElectriGen is building a 1.8GW gas platform in Texas. Meta is deploying 366MW of modular gas units for its El Paso campus. Its differentiation is the renewable mix: it is the only player combining solar, wind, and batteries with gas backup rather than relying on gas alone.

What success looks like

A single successful 10MW pilot would validate the model. Scale-up to 200MW by 2027 would confirm repeatability. At that point, The company becomes not a startup with a clever product but a new category of energy infrastructure company — one that competes with utilities on data center power delivery.

Sources