Bill Buzbee's goal last year was simple: keep the U.S. on the cutting edge of climate research. Scientists at the government-funded National Center for Atmospheric Research were developing more sophisticated ways of simulating the global climate--vital to understanding global warming--and needed bigger, faster supercomputers. So Buzbee, ncar's computing director, went shopping. But when he put competing machines through their paces, the clear winner was a system from Japan's NEC Corp., not from America's great supercomputer maker, Cray Research Inc.
That's when Congress threatened to cut off the funding for ncar's new system and Cray accused NEC of selling machines below cost. In late August, the Commerce Dept. agreed, charging both NEC and fellow Japanese supercomputer maker Fujitsu Ltd. with "dumping" machines on the worldwide market. Now the case is before the International Trade Commission, a U.S. agency that will decide if the alleged dumping injured Cray. A decision is expected in early October. If the itc says yes, then the U.S. will impose stiff duties, equivalent to 454% of price, on NEC machines sold in the U.S.
EXOTIC. With U.S.-Japan trade friction increasing again, this is no time to get entangled in a dispute that's more about protectionism than fairness. Indeed, the subtext here is a tangle of changing markets and national security concerns. Cray's so-called vector supercomputers, crammed with exotic technology, are still among the most powerful machines on the planet. But now these "hand-built Lamborghinis" are being replaced for most applications by cheaper systems created from groups of more mundane processors running in parallel, says supercomputer analyst Gary Smaby.
That's why to survive, Cray was forced last year to merge with Silicon Graphics (sgi), a leader of the new approach. But even the combined might of Cray and sgi may not be a match for NEC and Fujitsu.
This turn in the supercomputer market has set off alarms within the national security Establishment. Earlier this summer, an interagency committee led by the Defense Advanced Research Projects Agency (darpa) reported that U.S. supercomputer makers aren't healthy enough to develop the computers needed to crack enemy codes or design tomorrow's nuclear weapons. Without government help, "U.S. high-end technology will not be available to meet national security needs," the report warns. So it's not surprising that some supercomputer experts say the spooks were behind Commerce's ruling. After all, keeping out competition is one way to help U.S. industry.
But it's not the right way. The dumping case "has had a chilling effect on any company in the U.S. evaluating NEC or Fujitsu machines," says industry analyst Terry Bennett. The result: higher prices or, as in the case of ncar, reduced performance.
What makes this case galling to supercomputer customers is that the evidence isn't particularly strong. The central issue, of course, is whether the price of the NEC system was below its cost. Commerce, using numbers supplied by Cray, concludes that the cost of NEC's system is actually 454% higher than the $35 million price tag. But Commerce attributes two-thirds of the cost to research and engineering expenses, a figure analyst Bennett calls "ludicrous." "NEC made a very aggressive but not overly aggressive bid," he concludes. Still, this issue could be more fairly judged if NEC would release its own numbers, which it refuses to do. "It's frustrating when those who aren't playing by the rules are presumed to have the right numbers," says John L. Sullivan, general counsel for sgi.
For the case to go Cray's way, it must show that NEC's pricing around the world directly hurt sales. In fact, Cray's world market share in vector supercomputers has fallen from 73% in 1993 to 54% in 1996. But if the entire supercomputer market is considered--which most experts say is the right yardstick, given the shift to parallel machines--even Sullivan admits the case for injury would be tough to prove.
The big picture is that accusing the Japanese of dumping is the wrong way to cure the ills of the U.S. supercomputer industry. If we really want to ensure that the nation's military and intelligence communities stay on the cutting edge, Uncle Sam should simply--and directly--pay U.S. companies to develop the machines we need. Isn't that the way we do business in America?
Carey covers science and technology policy from Washington.