Google is deploying early-stage capital into Elementl Power to develop three advanced nuclear sites targeting 600 megawatts each—1.8 gigawatts combined—in a direct hedge against the electricity squeeze choking its AI infrastructure plans.
The arrangement is explicitly a development partnership, not a power purchase agreement. Each site remains contingent on planning, regulatory approval and construction. This distinction matters: Google is betting on future capacity, not securing immediate supply.
The Elementl move sits within a broader nuclear acceleration. Alphabet is separately nearing a multiyear deal worth at least $1 billion to purchase power from Constellation Energy equivalent to a single new reactor's output. Constellation's stock has risen sharply on the reports.
The dual-track strategy exposes a quantitative reality: AI data centers operate on a 3-to-5-year capex cycle. Modern large language model training requires 100+ megawatts of continuous power. A single hyperscale cluster consumes as much electricity as a mid-size city. Google's own data center power demand is climbing 15-25 percent annually according to utility disclosures, driven almost entirely by inference workloads.
Advanced reactors offer what renewables cannot: 90-plus percent capacity factor, zero-carbon baseload at the scale required for dense compute. A conventional 1.2-gigawatt pressurized water reactor generates 10 terawatt-hours annually—enough to run roughly 50,000 GPUs continuously at full load. Wind and solar require 5-to-10 times the land footprint for equivalent annual output and cannot guarantee the grid stability large data centers demand.
But here is the stress test: No U.S. advanced reactor project has yet reached commercial operation. NuScale's demonstration unit, the closest to completion, has slipped from 2024 to at least 2027. Elementl Power's sites face the same regulatory gauntlet. Nuclear permitting now averages 10-15 years. Even an expedited pathway leaves a multi-year gap between when Google needs the power and when it arrives.
Consortium deals with existing operators like Constellation bridge this gap. Constellation owns operational reactors that can be upgraded or re-powered. A direct purchase from Constellation supplies electricity within 1-2 years, not a decade.
The counterargument: Early-stage nuclear investments signal long-term commitment and may unlock policy support. If Google funds construction risk, it potentially influences siting and regulatory timelines. Data centers themselves have proven capable of accelerating infrastructure; Google's fiber buildout forced competitive deployment by incumbent telecom. Nuclear is capital-intensive and politically fraught in ways fiber is not.
The structural imbalance remains. Elementl's 1.8 gigawatts, if realized, addresses perhaps 18-24 months of Google's incremental demand. Alphabet's capex on data centers is now running $30-plus billion annually, with no ceiling in sight. The company is not pursuing nuclear because it solves the energy problem—it is pursuing nuclear because the energy problem is now existential and traditional sources cannot scale fast enough.
Google's move is not a solved problem. It is a public admission that one is coming.
