The Biggest Funding Round of the Week Wasn't an AI Company. It Was a Way Around the Power Grid.

The single largest venture-style check written this week wasn't for a model, an agent platform, or a chip. It went to a company that generates electricity and pipes it directly to a data center fence line, never touching the public grid at all.

National Grid Ventures agreed on July 1 to invest $1.75 billion for a 35% stake in Joulent, a technology-driven energy company built to deliver multi-gigawatt power at the speed AI compute now demands. The money funds Project Kilby, a 2.67-gigawatt facility in West Texas developed 50/50 with Chevron, under a 20-year power purchase agreement with a single customer: a Microsoft-operated data center. First power is targeted for 2028.

Joulent calls its model "across-the-meter" — turbine to server rack, with the wider grid never in the loop. That phrase is the entire investment thesis. Standard grid interconnection queues in Texas and most of the US now run five to seven years before a new gigawatt-scale load can get a firm connection date. Joulent's structure sidesteps that queue entirely by generating and consuming power on the same private wire.

The obvious read is that AI needs more power, so someone built more power. The more interesting read is who now sits at the center of AI infrastructure deals: not Nvidia, not a hyperscaler's balance sheet, but a 143-year-old British utility holding company and an oil major turning a shared turbine order into a co-located power plant. GE Vernova's turbines and reserved EPC capacity matter more to this deal's timeline than any GPU allocation does.

Joulent's round was the largest of the past week by a wide margin — bigger than Together AI's $800 million, bigger than any single AI application round Crunchbase tracked. When the biggest check in a week full of AI headlines goes to megawatts rather than model weights, that's a fairly direct readout of where the industry believes the next real bottleneck sits.

Every prior computing wave eventually collided with a physical constraint nobody priced early enough — fiber in 1999, cooling and floor space in the 2000s cloud buildout. AI's version appears to be electrons, and the early money solving for it isn't going to companies promising smarter software. It's going to whoever can put a gigawatt behind a fence line faster than a grid operator can say no. Worth asking which other infrastructure bottlenecks are still mispriced simply because nobody's written the check yet.