Kees van Berkel

Technical University Eindhoven, The Netherlands

AI in The Sky =? Pie in The Sky

Abstract

Could large‑scale AI data centers deployed in low‑Earth orbit (LEO) become a cost‑effective alternative to terrestrial facilities? The analysis compares orbital and ground‑based systems across launch cost, power generation, cooling, radiation exposure, and atmospheric reentry, as well as compute‑network performance.

A key distinction is the shift from terrestrial Clos networks to space‑based mesh networks using laser inter‑satellite links. Using bisection bandwidth, bisection intensity, and roofline‑style models, we show that while LEO‑based inference may be feasible, training frontier‑scale LLMs in orbit is unlikely to be competitive with terrestrial data centers.

Biography

Kees van Berkel received an MSc degree in EE from TU Delft and a PhD degree in CS from TU Eindhoven. He worked for over 40 years in the semiconductor industry (Philips Research, NXP, ST-Ericsson, Ericsson, and GrAI Matter Labs), and for 26 years as part-time full professor at the TU/e. He pioneered asynchronous VLSI from theory to mass production, and likewise for vector processors for software-defined radio.

Today he is a consultant on neural computing for Snap Labs and an emeritus professor at the TU/e. His current research interest is models of (parallel) computation: dataflow, cellular automata, neural computing, and quantum computing. His favorite application domain is radio astronomy.

Download