The idea of specialized hardware for running database management systems has been around for a long time. For example, in the late 1970s, UK national champion computer hardware maker ICL offered a “Content-Addressable Data Store” (or something like that), based on Cullinane’s CODASYL database management system IDMS. EDIT: See corrections in the comment thread. (My PaineWebber colleague Steve Smith had actually sold – or at least attempted to sell – that product, and provided useful support when Cullinane complained to my management about my DBMS market conclusions.) But for all practical purposes, the first two significant “database machine” vendors were Britton-Lee and Teradata. And since Britton-Lee eventually sold out to Teradata (after a brief name change to ShareBase), Teradata is entitled to whatever historical glory accrues from having innovated the database management appliance category.
Britton-Lee, which I first visited in 1983, basically had a very early client/server system, based on a handful of Z80s (Zilog’s Z80 was a technically worthy competitor to Intel’s microprocessor family, back before it was obvious Intel would conquer the world). Bob Epstein, previously head of the Ingres project and later CTO of Sybase, was involved. Britton-Lee also owned an unrelated software vendor named Altergo, as some kind of financial play. (For a company that at various times had both Richard Currier and Vaughan Merlyn working there, Altergo never amounted to much. Of course, Richard was long gone by the Britton-Lee days.) Fine entrepreneurs though they no doubt were, neither Dave Britton nor Geoff Lee really seemed to quite fit the enterprise software or hardware CEO mode, and the whole thing never really achieved ignition.
Teradata may have come more out of the Tandem tradition, via Citibank. (Teradata’s official company history credits CalTech but not Tandem.) I first visited Teradata in 1984 (when they started shipping product), meeting Chief Scientist Phil Neches.and a CEO out of Amdahl in whose office I saw golf tournament trophies for the first time in my life. Cocky young stock analyst that I was, grilled them about their lack of support for standard tools, such as SQL (they were a QUEL shop) or fourth-generation languages. I distinctly remember going to a blackboard or white board (I forget the detail as to which), and holding forth about the features of a competitive 4GL, with Phil taking copious notes. The basic product architecture in those days was a tree of microprocessors (I think Intel 8086s), with each parent node talking to two children, until it got down to the lowest-level nodes that actually talked to disk. (This is what was called Ynet.) The architecture meant that almost exactly half the microprocessors talked to disk, and 50% wasn’t necessarily bad overhead at all.
And that’s most of what I recall about database machines or data warehouse appliances before the mid-1990s, so I’ll stop right there.