Chapter 18 - Squandered Competitive Advantage

Competitive advantage can be squandered by executive indecision in a surprisingly short time. A delayed decision can be far worse than a mediocre decision. In IBM’s case, the delays and indecision involved its mainstream product lines in mainframes and minicomputers. In both cases, Akers’s delays lost the company a significant share of the computing market that it never recovered. In mainframes, the delays involved implementing a new technology. In minicomputers, it was the five years required to consolidate the five different minicomputer families. Of course, the industry was buffeted by the standardization of operating systems and the commoditization of hardware prices. But a CEO like Tom Watson, Frank Cary, and Lou Gerstner would have weathered the blow, leaving the company in far better shape than did John Akers. In the end, his chairmanship demonstrates the destructive power of the weak or misguided CEO better than almost any other case. That IBM survived at all was more a function of size than management.

Akers became head of the midrange product group in 1981, when the need to consolidate was clear to everyone including himself. This was also the year IBM introduced its blockbuster 3081 model, with its epic technological breakthroughs in circuit density, ceramic packaging, engineering automation, and testing. What was needed for the future was a continuation of the technology track set by Black Jack Bertram. Instead, IBM introduced the lackluster 3090. Well-orchestrated price/ performance improvements and uncertainty over the expiration of the tax credit created a mainframe sellout. Nevertheless, we wrote worriedly: “The Armonkishers are midway through a ten-year base technology, not a dawning as was true [last time].”

As it turned out, IBM’s enormous investments of the late 1970s projected a ten-year payback. So over five years, the 3090’s price/ performance merely doubled. Seventy percent of the improvement came from clever circuit design rather than enhanced chip technology. Manufacturing processes were more automated but not significantly different from our previous visit to Fishkill five years earlier. All in all, turning out the new 3090 model “was a cakewalk in comparison to the previous 3081,” admitted the assistant plant manager. We thought that pace dangerously complacent in view of the acceleration in IT technology. Indeed, IBM’s previous advantage over the Japanese was sharply narrowed, if not totally lost. And its technology pace was substantially behind the new breed, such as Sun Microsystems.

An opportunity to regain some of the lost ground began to take shape when IBM invested $500 million in a new semiconductor facility to produce enough memory chips to avoid dependency on its Japanese rivals. “We realized [the chips] would soon be our life’s blood,” recalled Dr. Irving Wladawsky-Berger, the disarmingly pixyish genius who subsequently led IBM campaigns into e-commerce and open-source software (Linux). But memory-chip self-sufficiency proved unnecessary after the Japanese lost the business to the Koreans.

With the chip threat diminished by 1986, Dr. Wladawsky-Berger began suggesting that the new facility could begin the transition to a new technology that could radically reduce mainframe costs. Until then, the logic circuits used in mainframes had depended on expensive and exotic, high-powered, high-heat “bipolar” technology. But minicomputers and PCs used cheap, low-powered, low-heat technology—CMOS (pronounced “see moss”) in IBM lingo, for “complementary metal-oxide semiconductor.”

By 1988, the competitively sensitive idea of CMOS replacing bipolar chips in mainframes was discussed quite openly at a Research Board meeting. CMOS chips would eventually outperform bipolar technology since the extraordinary circuit-density trajectories projected by Moore’s Law meant a computer could be built from many fewer chips (or even just one). In a nutshell, fewer chips meant shorter distances for the electrical pulse to travel and faster compute speeds. After that, several CMOS processors would be strapped together in parallel-processing arrangements for another huge boost in total mainframe performance.