Sun Microsystems

Contribution to AI

Symbolic AI spent the early 1980s convinced it needed purpose-built hardware: Lisp machines from Symbolics and LMI, tagged pointers and garbage collection in silicon, at prices that ran into six figures a seat. The machine Andy Bechtolsheim had designed as a Stanford graduate student took the opposite line — a stock Motorola 68000, off-the-shelf parts, Berkeley Unix, an Ethernet port — and by 1987 a Sun 3 running Lucid's compiler, sold under the name Sun Common Lisp, was competitive with dedicated Lisp hardware on Richard Gabriel's benchmarks for a fraction of the money. The argument for special-purpose AI machines evaporated with it, and the resulting collapse of the Lisp machine vendors is one of the events by which the AI winter is usually dated. What took its place was a laboratory built from identical general-purpose boxes. NFS, shipped in 1984 and put into the public record as RFC 1094 five years later, let a department export one set of home directories to every machine, so a researcher's data and code followed them to whichever node was free. That pattern — commodity compute pooled behind shared storage — is still how research clusters are organised, and Grid Engine, open-sourced in 2001, ran the batch queues on a great many of them. Java's effect came second-hand but lasted: Lucene, Hadoop and Mahout were written in it, and the big-data tooling that preceded the Python era ran on the JVM.

Drafted with AI and edited by hand (claude-opus-5, reviewed 2026-08).