Scientific Computer Systems, a San Diego-based manufacturer of minisupercomputers, has made the painful discovery that building a better mousetrap doesn’t always attract a world of customers to your doorstep.
Earlier this month, Scientific Computer discontinued its minisupercomputer product, the technical features of which had been lauded by industry observers. The announcement put an end to an effort that consumed five years and $50 million in venture capital and caused layoffs of 70 of the company’s 100 employees.
Stephen R. Campbell, vice president of marketing, maintained Monday that the company is still a going concern with a valuable asset: powerful networking software technology called VectorNet. Company officials insist that they intend to continue developing and marketing the software but concede that an infusion of new venture capital totaling “less than $5 million” is needed over the next several months if the effort is to stay alive.
Industry observers say the failure of Scientific Computers provides several object lessons, the most obvious being the high risk inherent in any high-technology venture. Others are the critical importance of introducing a product during the timely “market window,” and how industry pundits can be dead wrong in projecting market demand for a given product.
Moreover, the problems of Scientific Computer System illustrate how a company’s fortunes--and employees’ jobs--can hinge on events beyond the control of its executives, even those with an abundance of market and technical savvy.
Scientific Computer was founded in 1983 by Bob Schuhmann, formerly an executive with Floating Point Systems, another pioneering supercomputer manufacturer, and proceeded to attract venture capital from 2 dozen of the brightest lights in the investment community including J. H. Whitney & Co., TA Associates, Adler & Co., and TVI. Local venture firms Henry & Co. and CFB Ventures, a division of California First Bank, also chipped in.
In 1986, the company brought out its SCS-40 minisupercomputer product on time and just as powerful as promised. Between a Cray supercomputer and a Digital Equipment or International Business Machines mainframe in processing power, the SCS-40 could process data at 40 megaflops--40 million floating point operations or calculations--per second. That’s 10 times faster than a top-performing IBM mainframe and 1/20th the speed of a Cray.
Scientific Computer had some measure of success, selling 10 units its first year of sales and 40 all told at an average price of $550,000. Like other computers in its class, the SCS-40 was marketed to research institutions and companies that needed to process vast amounts of data, such as calculations used in aerodynamic modeling and oil field exploration, but which either could not afford or didn’t need Cray-level computing power.
What distinguished Scientific Computer from its dozen or so competitors was that it billed its machine as “Cray-compatible” or designed to operate off the same “compiler” or operating software called COS that drove the giant Cray machines, said Christopher Willard, senior industry analyst with Dataquest, a market research firm in San Jose. (Operating software translates instructions contained in applications programming into specific data-processing commands.)
The company believed early that it could convince Cray Research of Minneapolis to share its operating software. Cray would see the benefit because SCS-40 customers would eventually move up to more powerful Cray machines, suffering minimal pain in adapting their expensive applications software in the process, the reasoning went.
At first, Cray’s agreeing to “play ball” with SCS was not needed because the Minneapolis-based company had released an early version of COS to the public domain, meaning anyone could design a computer system around it. But, soon after the SCS-40 was introduced, Cray brought out an advanced version of COS that it kept proprietary.
Scientific Computer was thus stuck with an outdated Cray compiler that applications software publishers were reluctant to support, a fact that reduced the appeal of the SCS-40.
“Cray did not show any favoritism with SCS by cooperating or providing software,” said Jeffry Canin, an independent technology analyst based in San Francisco who formerly worked at Hambrecht & Quist investment bankers. “Cray did not perceive it to be in their interest to help start-up companies like SCS.”
SCS then retrenched by developing its own version of COS in cooperation with Boeing, a system based on another Cray operating software package also in the public domain called CTSS. At the same time, Boeing bought a number of SCS-40s and agreed to market the SCS machine. But the partnership fell short of generating the sales that SCS had hoped for, partly because the Cray-compatible marketplace was too small a niche and thus unattractive to software publishers.
Asked how the company could, in effect, bet its future on the cooperation and good will of Cray, former Scientific Computer President Jack Hugus said Monday that Boeing and Scientific Computer were laboring under the impression that Cray “would indulge us the functional equivalents” of current Cray software. Hugus stressed that the Cray compatibility question was decided by Schuhmann before he joined the company in 1986.
Hugus, who resigned from the company a year ago and who now is marketing general manager for General Electric’s government electronics systems division near Philadelphia, said the relationship with Cray was a “continually murky situation.”
Meanwhile, Scientific Computer’s archrival Convex Computer Corp. of Richardson, Tex. had beaten it to the market by 18 months, introducing a minisupercomputer based on UNIX operating software, a system also in the public domain but which is gaining increased popularity among workstation and mainframe computer users.
Convex’s decision to use UNIX was critical because it gave the company market access to a host of Digital Equipment mainframe owners using UNIX who would someday be ready to move up to a “minisuper,” Convex marketing vice president Frank Vince said last week. Even Cray has recently introduced UNICOS, an operating system based on UNIX that may eventually replace its COS software.
As critical months and Scientific Computer’s problems with Cray continued, Hugus said, it became clear that the minisupercomputer world was moving to UNIX-based operating software, Hugus said, not necessarily because UNIX was inherently a better language but because Convex hit the market window earlier and was able to establish itself over its competitors.
For whatever reason, the SCS-40 was left hanging in the market with minimal software support while Convex was able to market its machines as having the support of 200 or more application software programs.
“Yes, SCS has a better mousetrap in terms of its speed and technology. That wasn’t the problem with the company,” said Jack Dongarra, scientific director of the advanced computer research facility at Argonne National Laboratory in Argonne, Ill. “The failure was in concentrating on hardware and not devoting enough resources to software,” he said.
By the time Scientific Computer introduced a UNIX-based operating system of its own last year, it was too late for it to save the SCS-40 product.
Several companies including Scientific Computer got into the minisupercomputer business thinking that the market would grow much faster than it has. Projected by some analysts to reach $500 million in 1988 and $1 billion by 1990, only $330 million worth of minisupercomputers were sold last year, said Dataquest’s Willard, adding that the market could reach $1 billion in sales by 1992.
Severe price competition and a much smaller revenue pie to carve up has caused all but Convex, the minisupercomputer leader with more than 100 machines installed, to be money-losers so far. Because of these factors, Willard expects the number of minisupercomputer “players” to shrink from 15 in the mid-1980s to four or five over the next few years.