Descriptions:
Bloomberg Technology interviews Mike Shrover — former CTO of Meta and founding partner of Gigascale Capital — for a rigorous cost analysis of where AI compute infrastructure should actually be built. The conversation is prompted by SpaceX’s IPO prospectus, which explicitly outlines plans for orbital data centers at a stated cost of one dollar per token.
Shrover, who oversaw the build-out of tens of millions of square feet of cloud capacity at Meta, applies straightforward engineering economics. Putting mass into orbit currently costs roughly 100 times more than deploying the equivalent in the ocean, making submarine data centers — like those being developed by companies such as Panthalassa — dramatically more cost-competitive for almost every operator. The Southern Ocean alone offers an estimated 10 terawatts of untapped power alongside near-unlimited free cooling from cold seawater, solving two of the hardest problems in high-density AI compute. Land-based solar plus batteries, meanwhile, can be deployed in 12-24 months — faster than gas plants and vastly faster than orbital infrastructure.
The critical exception, Shrover argues, is SpaceX itself. Its vertical integration, existing Starlink launch infrastructure, and wholesale cost structure change the math entirely, making orbital data centers a viable — and likely profitable — business specifically for SpaceX while remaining uncompetitive for any other operator. He is skeptical but not dismissive of pitches in this space, noting that once tokens are visibly being generated from an ocean buoy or a satellite feed, major cloud providers including Meta will pay serious attention. Until proof-of-concept exists, it remains theory.
📺 Source: Bloomberg Technology · Published June 16, 2026
🏷️ Format: Interview
