The AI arms race just got a clean energy constraint, and Google is betting on drilling holes instead of burning gas to win it.
The Summary
- Google signed a deal with Fervo Energy for up to 396 megawatts of geothermal power for a planned Utah data center starting in 2028
- Fervo's stock surged on the announcement, signaling investor confidence in geothermal as data center infrastructure
- This is Google locking in baseload renewable power for AI infrastructure two years before it even comes online
The Signal
Google is doing something most hyperscalers won't admit they need to do: securing dedicated clean power capacity years before the data center exists. The 396 megawatts from Fervo's Utah geothermal project is not solar with batteries. It's not wind with natural gas backup. It's always-on power from heat in the ground, which matters when you're running GPU clusters that can't wait for the sun to come back up.
The 2028 timeline tells you everything about how constrained the power market has become for AI infrastructure. Google is contracting power delivery two years out because there isn't spare clean capacity sitting around. Every other hyperscaler is doing the same math: figure out where the electrons come from before you pour the concrete.
"This is Google locking in baseload renewable power for AI infrastructure two years before the data center even comes online."
Geothermal is having a quiet moment. It doesn't scale as fast as solar farms or get the press wind projects do, but it delivers constant output. No weather dependence. No time-of-day curves. For training runs that take weeks and inference serving that happens 24/7, that matters more than cost per kilowatt-hour. Fervo uses horizontal drilling techniques borrowed from oil and gas to access heat in rock formations that traditional geothermal couldn't reach. It's not new physics, just new application of existing drilling tech.
The stock market clearly thinks Fervo is onto something, with shares jumping on deal announcement. Investors are pricing in a future where geothermal developers with proven capacity become strategic assets, not just power vendors. When Google commits to 396 megawatts from a single project, it validates both the technology and the commercial model.
The Utah location is not random. Western states have accessible geothermal resources, land for data centers, and favorable regulatory environments. Google gets to check the renewable energy box while building in a region with lower land costs than coastal markets. The calculation is: drill for heat in Utah or fight for grid capacity in Virginia.
The Implication
Watch for more announcements like this across hyperscalers. The power question is now the bottleneck for AI infrastructure expansion. Companies that figure out dedicated renewable baseload power deals will build faster than those waiting for grid upgrades or betting on offsets.
If you're building agent infrastructure or planning to run serious compute, the power source question just moved from "nice to have clean energy story" to "can we actually get electrons when we need them." Geothermal won't power every data center, but it's now proven as a real option for regions with the right geology. The companies that move early on power deals will have deployment advantages that matter in 2028 and beyond.