With regularity, Cray Research Inc. used to pooh-pooh the idea of stringing hundreds of relatively cheap, standard microprocessors together to create powerful supercomputers. This was the company, after all, that had practically invented the traditional supercomputer based on a few ultrafast custom processors. But now, years after competitors first began fielding micro-based machines, Cray Research is introducing one of its own, the T3D.
Better late than never, say stock analysts. The supercomputing market leader's sales growth has been flat for years, its shares mired in the low to mid 20s and trading below book value. Amid a weak economy, cutbacks in military spending, and the maturing of Cray's traditional product line, the newfangled T3D may be the company's best chance for growth. The market's focus is shifting rapidly to the machines, known as massively parallel processors (MPPs), which more than a dozen hungry competitors, including IBM, now sell. That is reducing Cray's traditional "vector" supercomputers to mainly a replacement market. Says Goldman, Sachs & Co. analyst Barry Willman: "Cray Research is at a crossroads."
There's no need to tell John F. Carlson, who last January succeeded longtime Chairman and Chief Executive John A. Rollwagen, who left in hopes of joining the Commerce Dept. Carlson, the company's former finance chief, is currently seeking board approval for a new strategic plan, which he declines to discuss. Robert H. Ewald, executive vice-president, meanwhile, has realigned the sales organization--in part, he wrote in a memo to employees, to "create more enthusiasm for ourselves."
The first order of business will be working through the issues raised by the T3D, slated for unveiling in Washington, D.C., on Sept. 27. Will it sell well enough to boost overall revenues and earnings? Or will it just cannibalize orders for traditional Cray hardware? The T3D must be linked with a traditional Cray supercomputer, which effectively limits sales to the company's installed base. "At its best, the [T3D] will preserve market share," says supercomputer market analyst Gary Smaby, head of the Smaby Group.
"SIDEWAYS" STEP. In fact, the market for Cray's traditional systems, selling for up to $30 million apiece, is all but saturated. Efforts to broaden beyond Cray's loyal following of scientists and engineers, especially by branching into commercial markets such as finance, have fizzled. An attempt to seed new customers with lower cost "minisupercomputers" also fell flat. Cray Research sold just 44 of its $250,000 and up EL minisupers in the first half of 1992 and this year, only 9. An even lower-priced system, the Superdragon, plugs into networks of engineering workstations. But it's not likely to cause more than a blip in sales and earnings soon.
So all eyes are on the T3D. Company executives believe their MPP entry will outperform rival offerings from MPP leaders Intel, Thinking Machines, and nCube. It uses Digital Equipment Corp.'s speedy new Alpha microprocessor. And by using liquid to cool them, Cray can cram as many as 2,048 of those chips tightly together and significantly speed communications among them. Rival designs cool their chips with air, which is much cheaper. Still, Carlson says because it is using cooling technology developed for earlier products, Cray is able to price the T3D quite competitively--analysts figure about $8 million for a 256-processor model, for example. And the company also has plans for a cheaper, smaller, air-cooled model.
Rivals scoff at this approach. David C. Douglas, head of hardware development at Thinking Machines Corp., calls it "a big step sideways." Henry Burkhardt III, CEO at MPP maker Kendall Square Research Corp., calls the T3D "disappointing" because it uses "20-year-old technologies." And Ed Masi, president of Intel Supercomputers Systems Div. and a former Cray Research executive, reckons the cost of liquid cooling will eat into profits: "I don't think they can make money on it matching my price."
In the recent past, Cray Research's big worry was competition from Japan. Today, though, as with so many established computer makers trying to cope with the microprocessor revolution, its main struggle seems to be with itself--its old self, the one that was so secure it could look down on upstart technology.