Datapoint Corporation

» see datapoint.com as it lived, on the wayback machine

~ the obituary ~

The story that gets told is a dinner. Phil Ray and Gus Roche took Robert Noyce of Intel and the president of Texas Instruments to a meal, handed each of them a postcard with a processor schematic drawn on it, and offered a bet: whichever company first put a whole computer on a single chip could keep the money Datapoint already owed it. Noyce pointed out that a microprocessor would cannibalize Intel's own shift register business, and then agreed anyway.

Whether the postcards were literally postcards is disputed on the Wikipedia article itself, and it does not matter much, because the underlying fact is not in dispute. Computer Terminal Corporation was incorporated in San Antonio in July 1968 by two engineers who had been working the Apollo contract for General Dynamics in Florida and were tired of Teletypes. Their first product, the Datapoint 3300, was a terminal with a television screen instead of a paper roll, and it sold well enough to be rebadged as the DEC VT06 and the HP 2600A. Their second product changed the industry and did not benefit them at all.

The Datapoint 2200 shipped in May 1970. Its instruction set architecture had been worked out by Victor Poor and Harry Pyle on a living room floor over the 1969 Thanksgiving holiday, while Poor was serving out his notice at another company. The 2200 implemented that architecture in ordinary TTL logic because the chip version was not ready in time. Intel's Ted Hoff and Stan Mazor delivered it as the 8008 in April 1972, and Datapoint tested it and walked away, because the chip was too slow and needed too many support chips around it.

I learned processor architecture in a university lecture hall in the mid-2000s, and the story started at the 8080, or sometimes at the 8086, depending on the course. Nobody said the words Datapoint 2200. This is a company whose distance from me is not geographic and not generational, it is archival: the thing it made is under my hands every working day and its name was edited out of the account somewhere between 1972 and the textbooks. Intel built the chip and Intel got the paragraph.

~ the case file ~

~ the chip that got away ~

The lineage is clean and well documented. The 8008 implemented the Datapoint 2200's instruction set. The 8080 of 1974 extended it. The 8086 of 1978 was designed for assembly-level source compatibility with the 8080, which is why traces of a 1969 terminal design survive in the register structure of a chip built fifty years later. Ken Shirriff has walked the whole path instruction by instruction.

Datapoint kept building its own processors out of discrete logic for another decade, through the 6600 around 1982, before eventually moving its hardware onto Intel 386 chips like everybody else. There is a version of this story where the company that specified the microprocessor also owns it. That version required Datapoint to accept a slow first attempt and wait, and Datapoint had a product launch date.

~ the network ~

The second invention deserves as much attention as the first. ARC, later ARCNET, was developed by John Murphy under Victor Poor in 1976 and announced in 1977, and it was the first commercially available local area network. Ten thousand installations went in worldwide. It used thin coaxial cable rather than Ethernet's thick yellow trunk, and it was token-passing, so behavior under load was predictable. You could also connect or disconnect a station without bringing the network down or reconfiguring anything.

It also enabled something Datapoint called Datashare and later RMS: terminals reaching each other's disks and printers without a mainframe in the middle. IBM and Novell both built variants of the idea afterward and ARCNET never got the market back. ARCNETplus pushed the standard to 20 Mbit/s and arrived at almost exactly the moment 100 Mbit/s Ethernet did, which settled the argument.

~ the slide ~

Under Harold O'Kelley the company grew revenues at forty percent a year from 1973, passing 100 million dollars in sales in 1977 and reaching 450 million by 1981. Then early 1982 happened. Datapoint had been booking sales that were not really sales, and when customers slowed their payments in the recession the whole arrangement inverted. Reversals and bad debts wiped out roughly 800 million dollars of market capitalization inside a few months, and the SEC ordered the practice stopped.

The Waco factory closed that year, Fort Worth closed in 1985, and San Antonio was left as the only plant. In December 1984 Asher Edelman disclosed an eight percent stake and bid 416 million dollars for the company. The board refused, fought him for three months, and lost. O'Kelley resigned on 16 March 1985 and Edelman took the chairmanship.

~ what got buried with it ~

The service division was spun out as Intelogic Trace, which went into Chapter 11 and had its assets sold off on 6 April 1995. The video conferencing patents, covering a picture-in-picture system called MINX that reportedly earned more from infringement suits than from sales, were sold to a reseller in December 1999. Datapoint itself filed Chapter 11 on 3 May 2000 and was broken up.

So the company died in the same year as the dot-com crash, thirty-two years after it started, having invented the instruction set that runs most of the world's software and the first LAN anyone could buy. It kept neither of them. What killed Datapoint in the end was not any competitor's better product. It was a set of books that said one thing while the customers were doing another.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »