Google: Starship needs 1,800 launches before orbital data centers make sense
Original: Google thinks SpaceX’s Starship has to launch 1,800 times before space data centers get off the ground
Why This Matters
Orbital AI compute is years away, but Google's peer-reviewed math sets the baseline for the entire sector's cost targets.
Google launched its first orbital compute satellite on October 1, 2026, aboard a SpaceX rocket from California. Built by Planet Labs, the prototype carries a Tensor Processing Unit — Google's GPU rival — and is part of Project Suncatcher, the company's bid to build 81-satellite compute clusters in low Earth orbit.
The satellite, part of Google's Project Suncatcher, marks the first time the company has sent one of its TPU chips into space. Managed by exec Travis Beals, the project aims to eventually operate a network of 81 satellites flying in close formation to process AI workloads in parallel. On-orbit, the TPU will run in 15-minute bursts to manage power and heat constraints on the Planet Labs standard platform.
A more capable two-satellite demo — purpose-built for heavy compute and linked by laser communications — is expected to launch in 2026. Beals emphasized that Suncatcher is explicitly designed around future AI workloads: 'We're trying to look ahead to not just what workloads exist today, but where they will be in five years.'
Alongside the launch, Google published a peer-reviewed white paper in Joule, offering one of the most detailed public analyses of the economics of orbital compute. The paper's headline finding: SpaceX's Starship would need to complete roughly 1,800 launches before the cost of getting hardware to orbit drops enough for space data centers to be commercially viable. The researchers caution the paper is not a full economic feasibility study, but the figure illustrates just how far the industry has to go.