Skip to content
September 10, 2026 · Austin, TX · AI · Energy

An Anthropic Investor Just Bet $120 Million on Skipping the Power Grid Entirely.

Spark Capital — the venture firm that led Anthropic’s $450,000,000 Series C and holds a board seat there through partner Yasmin Razavi — is leading a $120,000,000 Series A for TAR (Transformative American Resources), an Austin startup that builds power systems for AI data centers designed never to touch the public grid at all.

The round, reported by Bloomberg on Sept. 10, 2026, values TAR at roughly $1,000,000,000 — up from about $500,000,000 three months ago. The pitch is simple: the grid interconnection queue for a new large power customer now routinely runs five to seven years. TAR says it can deploy its modular solar-battery-wind-turbine systems in three months.

  • $120,000,000Series Aled by Spark Capital, an Anthropic board-level investor — Bloomberg
  • $1,000,000,000TAR valuationup from ~$500M at its June 2026 seed round — Forbes, Dealroom
  • 3 monthsdeploymentTAR's claimed build time, vs. a 5-to-7-year grid interconnection queue
  • 165%demand growthprojected rise in global data-center power demand by 2030 vs. 2023 — Goldman Sachs
  • 55 → 84 GWglobal demandworldwide data-center power demand, current vs. 2027 forecast — Goldman Sachs Research
  • ~90 GWannounced, ~2 GW livebehind-the-meter gas capacity announced at U.S. data-center projects vs. actually operating, as of April 2026
§ 01 / The Bet

TAR was founded in 2026 by Leonhard Soenke and Pat Becker, who had previously built and exited a creator-economy startup called Throne. Becker describes the technology plainly.

We are building modular, scalable, behind-the-meter energy systems consisting of solar, batteries, wind energy and — for emergencies or long periods of unfavorable weather — simple cycle gas turbines. The main premise is being able to provide power 24/7 without relying on the grid at low cost, minimal environmental impact and extremely fast.

Pat Becker · co-founder, TAR

The company’s pilot targets 10 megawatts of continuous power, with a pipeline Soenke has told Business Insider exceeds 200 megawatts by 2027 — a figure this piece could not independently corroborate beyond that single outlet. TAR raised a $27,000,000 seed round in June 2026 at roughly $500,000,000 valuation before this week’s jump. Prior seed backers Buckley Ventures and Align Fund also participated in the new round.

§ 02 / Why the Grid Is the Bottleneck

The federal government has spent the past year treating data-center power demand as a first-order infrastructure problem, not a niche one. In October 2025, Energy Secretary Chris Wright (Trump administration) directed FERC to fast-track a rulemaking on large-load interconnection under the DOE Organization Act. FERC issued Section 206 orders to every U.S. grid operator on the subject in June 2026.

Without copious amounts of energy in a certain quality and a certain type of energy, there is no AI.

Chris Wright · U.S. Secretary of Energy (Trump administration)

FERC Chairman Laura Swett (Trump appointee) has framed the commission’s interconnection push in similar terms — historic action, in her words, to modernize how the grid connects large new customers.

Goldman Sachs projects a 165% rise in global data-center power demand by 2030 — the bet TAR's investors are making is that off-grid supply grows just as fast. — Civic Intelligence illustration

Goldman Sachs projects global data-center power demand will rise 165 percent by 2030 relative to 2023, from roughly 55 gigawatts today to 84 gigawatts by 2027. Separately, other industry analysts (not independently verified for this piece and cited here only as directional context, not sourced figures) have put cumulative AI-driven data-center capital spending in the trillions of dollars through 2030. Against that demand curve, roughly 90 gigawatts of behind-the-meter gas capacity has been announced at U.S. data-center sites as of April 2026 — but only about 2 gigawatts is actually operating, with xAI’s 1,498-megawatt Memphis site accounting for most of what exists today.

How the Electrical Grid Is Being Rebuilt for AI | Bloomberg Primer
§ 03 / The Comparables

TAR is one entrant in a fast-forming category, not a category of one. Microsoft and Constellation Energy are restarting the Three Mile Island nuclear plant’s Unit 1 — a $1,600,000,000 refurbishment backed by a $1,000,000,000 federal loan, aiming for 835 megawatts online in 2027. Emerald AI, which takes a software-based grid-flexibility approach rather than building generation hardware, raised $150,000,000 at a $1,050,000,000 valuation on Aug. 25, 2026, backed by Nvidia, Samsung Ventures, Siemens, and GE Vernova among others.

Officials On The Record

David LaCerte (FERC Commissioner, Trump appointee): “We have the technology. We should use it now to enable faster interconnection of large loads, lower costs.”

Chris Wright (DOE Secretary, Trump administration): “They're the answer to drive down electricity prices. They're not the problem.”

AI demand driving up energy costs for everyday Americans, Bloomberg News finds · CBS News
X
TradedVC
@TradedVC · June 2026· paraphrase

Green energy infrastructure startup TAR raised a $27 million seed round to build modular, behind-the-meter power systems for data centers.

X
bpaynews
@bpaynews · June 2026· paraphrase

JUST IN: TAR closes a $27M seed round to fund modular, plug-and-play power systems for AI-era data centers, targeting near-24/7 local power via solar, wind, batteries and gas backup.

§ 04 / The Stakes

An Anthropic-adjacent investor betting on off-grid power is not a coincidence of branding — it’s a bet that the AI industry’s single biggest physical constraint over the next five years is not chips, but electricity, and that whoever solves the deployment-speed problem captures a share of every AI lab’s capital budget regardless of which model wins the underlying AI race. If TAR’s 200-megawatt 2027 pipeline holds up, it would still represent a small fraction of the tens of gigawatts the industry says it needs. The company’s bet, and its investors’, is that being three months instead of five years is worth funding at a billion-dollar valuation before the technology has scaled past a 10-megawatt pilot.

AI Uses Less Power Than You Think. The Real Problem Is Worse · The Prof G Pod, Scott Galloway
Sources & Methodology · 14 Sources
Bloomberg's report is paywalled; deal terms here are cross-confirmed across six-plus independent secondary sources. No video or X post specific to today's Series A had surfaced as of publication — the embeds below cover the broader AI-power-demand story and TAR's earlier June 2026 seed round, and are captioned accordingly rather than presented as coverage of today's news.