Can the ocean be the next grid?

At a steel yard outside Portland, Oregon, an 85-meter solid-steel structure shaped like a lollipop waits to be towed to sea. Its builders, a startup called Panthalassa, describe it as a data center. Last week Peter Thiel led a $140 million round into the company, a Series B that values it at roughly $1 billion before it has shipped a single commercial node.

The structure is the latest, and strangest, exhibit in a thesis that has consumed Silicon Valley for two years: that the conventional grid will not deliver the electricity AI requires inside the capex window investors care about, and that whoever finds power outside the grid first will own a strategic asset. The thesis is no longer contested. What has changed is how far investors are willing to follow it.

Microsoft committed $16 billion to restart Three Mile Island in late 2024, a Pennsylvania plant decommissioned in 2019. The deal felt baroque at the time; today it reads as the conservative version of the trade. Since then Meta has signed nuclear-supply agreements with Vistra and Oklo totaling 6.6 gigawatts, contracted roughly 300 megawatts of next-generation geothermal from XGS Energy and Sage Geosystems, and committed to ten new gas plants for a single Louisiana campus. Google has bought another 265 megawatts of geothermal across two contracts, one with Ormat Technologies and one with Fervo Energy, which closed an oversubscribed $462 million Series E in December. Crusoe Energy, which co-locates GPUs with stranded gas in the Permian, drew a $750 million credit facility from Brookfield in mid-2025. In March, Starcloud — a Y Combinator graduate that runs Nvidia H100s in low-Earth orbit and cools them by radiating heat into the void — closed a $170 million Series A at a $1.1 billion valuation. Each successive bet has moved farther from the grid the country actually has.