INTEL: THE NEXT REVOLUTION
Personal computers are now such fixtures that it's easy to forget the way things were before the PC era began. Incredible as it seems, that was only a dozen years ago. Back then, hardly anyone knew what a PC was, save for some electronics freaks who had soldered one together from a kit. In fact, the brain in all PCs -- a silicon chip called a microprocessor -- is less than two decades old.
The company that started it all was an aggressive Silicon Valley startup called Intel Corp. Founded in 1968 by the PhD engineers who still hold the reins, Intel invented the first microprocessor in 1971. That chip spawned the handheld calculator. Two years later came the computer-on-a-chip designs that ultimately ushered in the PC age. Since then, Intel has dominated microprocessors, launching successive models that have put more and more computing power at the fingertips of everyone from grade school students to top executives.
DESKTOP DREADNOUGHT. In just a few months, Intel plans to throw a coming-out party for its next-generation design, a superchip dubbed the 80486 and familiarly known as the 486. It will be a veritable mainframe-on-a-chip. Packing more than 1 million transistors, this thumbtack-size sliver of silicon will deliver the number-crunching wallop of a low-end IBM 3090 mainframe (chart).
The 486 will thus inaugurate a new age in desktop computing: personal mainframes. But Intel is after far bigger game as well. The Santa Clara (Calif.) company wants nothing less than to radically overhaul the whole computer industry and put its new chip at the core of everything from minicomputers to supercomputers. ''My hope and vision,'' says Intel's scrappy chief executive, Andrew S. Grove, ''is that our technology is going to be the heart, spine, and framework of the entire computer industry.''
Most experts agree that microprocessors will become the centerpiece of a new breed of mainframes, but not everyone is convinced Intel will be on top. The sheer scale of those unfolding markets has already attracted more competition than Intel has ever faced. Companies such as Sun Microsystems, MIPS Computer, and Hewlett-Packard have introduced sleek new microprocessors, and they contend that their chips will outpower the 486. Japan is determined to break Intel's grip, as well. Says Brian Halla, who left Intel in August to help rival LSI Logic Corp. get into microprocessors: ''This is a break in the action, where everybody gets a chance to play again.''
But Intel can't be easily dismissed. With its strong ties to IBM, Intel's technology commands the office computing market, and the mountains of software written for IBM PCs virtually guarantee that Intel will continue to be a major factor. The chips that Intel directly sells account for 20% of the total worldwide microprocessor business, according to market researcher Dataquest Inc. And since competitors produce most older Intel designs under ''second-source'' agreements, Intel estimates that fully 70% of all PCs rely on its technology. Sales of machines built with its 80286 chip, the brains of the IBM PC/AT generation, continue to expand nicely. At the same time, systems based on Intel's latest microprocessor, the very profitable 80386, are taking off. These 386 machines -- from Big Blue, Compaq, Tandy, and the clonemakers -- could account for up to half of all IBM-type PC sales next year.
That's why Intel is racking up revenue increases at a rate that even a startup a fraction of its size would envy. In the first six months of this year, earnings soared 211%, to $224 million, on revenues that rose 63%, to $1.4 billion. By yearend, revenues are expected to pass $3 billion. ''This is the glory year for Intel,'' says W. J. Sanders III, president of rival Advanced Micro Devices Inc.
CENTRALIZED POWER. With the 486 hitting the market next year, the first desktop computers incorporating it should show up around 1990. Their price tags will probably be close to $20,000, given the 486's expected initial cost of $1,500. ''They won't be on everyone's desk right away,'' says Paul A. Maritz, a vice-president at Microsoft Corp., ''because they'll be too expensive.'' Such systems will more likely start out as a departmental ''host'' computer, with the staffers in, say, customer service or marketing sharing its capabilities.
But the power a 486 can cram into a PC-size box will be nothing short of sensational. While PCs based on Intel's current microchamp, the 80386, can run a handful of programs simultaneously, it's expected that the 486 will be able to run dozens, each much larger and more sophisticated than anything available now.
That should finally give tomorrow's desktops enough juice to run software that makes PCs genuinely user-friendly. Because this new PC breed can do so many things simultaneously, it will function like a platoon of talented assistants and middle managers, handling routine tasks automatically and responding instantly to help out in an emergency. ''You will greatly concentrate power into the executive suite,'' predicts William F. Zachmann, senior vice-president of International Data Corp.
The bigger challenge for Intel will be to get its chip accepted as a standard in tomorrow's new class of mainframes as well. To date, microprocessors have barely scratched the surface of the large-computer business; they simply haven't been fast enough. So the silicon brain in minicomputers and mainframes -- called a central processing unit, or CPU -- is a sizable collection of high-speed logic chips chained together on a big printed-circuit board.
Not only are such CPUs expensive, they also are rapidly running out of steam. For several years it has been evident that the customary approach to CPUs will soon bump up against theoretical limits. So 21st century mainframes and supercomputers will require some new ''architecture'' to sustain continuing increases in performance. One answer is to use multiple processors, so many brains can gang up on tough problems. Just about every big-computer company either has developed or is developing such a parallel-processing system, including Cray Research, Digital Equipment, and IBM. But such machines require new software and often cannot run existing programs. Moreover, multiplying the CPUs can substantially boost the price of the finished system.
With the advent of the 486, asserts Intel's Grove, more and more computer makers will abandon CPUs and take to the trail blazed by a few newcomers, such as Sequent Computer Systems Inc. and Stratus Computer Inc., which are among the companies producing mainframe-class systems from clusters of microprocessors. While these machines also require special software, they sell for about one-tenth the cost of comparable CPU-based mainframes (page 80).
The force pushing the industry from CPUs to MPUs -- microprocessing units -- is the ever-shrinking size of semiconductor circuitry. The first 386 chip was designed with circuit lines that were 2 microns wide (a human hair is roughly 100 microns in diameter). The 486 will have 1-micron lines. That means the number of transistors crammed onto the chip will quadruple, thereby cutting the distances that signals must travel between transistors (page 77). Result: the new chip will be almost as fast as the high-speed devices now used in big CPUs.
DOWN MEMORY LANE. The 486 certainly has an illustrious heritage of innovation. Aside from developing the first microprocessor, Intel invented the DRAM, or dynamic random-access memory, a chip that is now indispensable in every computer, big or small. It also concocted two other types of memory chips that are widely used in consumer electronics, military systems, and industrial machinery: static RAMs and EPROMs -- erasable programmable read-only memories. ''They've done so many important things in this industry it's almost mind-boggling,'' says Gordon Campbell, president of rival Chips & Technologies Inc.
That string of successes has made legends of Chairman Gordon E. Moore, Vice-Chairman Robert N. Noyce, and Grove. Noyce paved the way for it all by inventing the integrated circuit in 1957, but it wasn't until 1968 that the team left Fairchild Semiconductor Corp., also founded by Moore and Noyce. While Moore and Noyce concentrated on research and long-range strategy, Grove became the hands-on manager who translated their ideas into commercial products.
Andras Grof fled Hungary after the 1956 Soviet invasion and arrived in the U. S. with only loose coins in his pocket. He changed his name to Andrew S. Grove, worked his way through City College of New York as a waiter, then headed west to earn a PhD in chemical engineering at the University of California at Berkeley.
At Intel, he earned a reputation as a tough taskmaster, though insiders say he has mellowed since his promotion to CEO early last year. And because he plans to retire as CEO in three years, when he turns 55, Grove recently delegated substantial authority to the senior vice-presidents who run Intel's various operations: David L. House, in charge of microcomponents; Laurence R. Hootnick, operations manager for microcomponents; Leslie L. Vadasz, head of a new thrust into systems, and Craig R. Barrett, Intel's manufacturing chief.
EXPRESS SERVICE. Intel will need all the teamwork it can muster. Hungry competitors now are nipping at its heels with a new design concept called RISC, for reduced instruction-set computing. The idea was developed by IBM more than a decade ago, but lately RISC has taken the electronics industry by storm. Basically, advocates say, it's a way to turbocharge MPUs, promising potential speedups of as much as 1,000%.
The idea is simple: Eliminate unnecessary instructions. A general-purpose microprocessor comes loaded with hundreds of instructions to cover a host of contingencies. As speedy as the chip is, it takes time to sort through all those instructions and act on them. But for most applications, maybe half of those instructions are never used -- or used very rarely. Leaving them out speeds things up. And not only is the chip faster, it is also easier -- hence cheaper -- to design and manufacture.
The trade-off is that a computer powered by a RISC chip won't run existing software, or at least not very efficiently. Competitors, however, are betting that the speedup in performance will be so dramatic that users will be willing to modify their programs for the new machines. Doing that is becoming relatively easy with new software-automation tools that adapt programs to the increasingly popular Unix operating-system program.
If RISC chips do deliver substantially more speed than a 486, neglecting the technology could hurt Intel in explosive growth markets such as engineering workstations, where power is king. But Murray A. Goldman, general manager of microprocessors for Motorola Inc., isn't complaining too loudly: ''This is the first time we've ever gone after a market where Intel is not a player.''
Two examples of hot new RISC competitors that have cropped up: Sun Microsystems Inc., a fast-growing supplier of workstations, and MIPS Computer Systems Inc., formed in 1984 by RISC researchers from Stanford University and IBM. Neither had previously been a factor in microprocessors. But Sun's chip, dubbed Sparc, has already been licensed by several computer makers, including AT&T, Unisys, and Britain's ICL -- and Tandem Computers is developing a new system with MIPS Computer's chip. A handful of chipmakers -- Texas Instruments, LSI Logic, Japan's Fujitsu, and Cypress Semiconductor, among others -- will produce and market one or both chips. And Motorola, Intel's chief competitor and the leader in workstations, recently unveiled its own RISC design.
BUILT-IN MARKET. To Grove, all of the pushing and shoving in an effort to horn in on office computing is reminiscent of what happened before Intel's 386 chip was introduced in 1985. Even before the rise of RISC, publicity mills have churned out claims that new-generation chips would outperform Intel's designs. And in the case of the 386, some of them did, in fact. Yet only Motorola's 68020 made any inroads, riding in on the popularity of Apple Computer Inc.'s Macintosh.
Grove concedes that all of the new options in superfast chips will probably trigger ''a segmentation of the market'' that would crack open some applications to the new contenders. But the mainstream office market remains secure, he insists, because the 486 has something no newcomer can match: ''compatibility with the piles of software already out there'' -- an estimated $12 billion worth of programs designed for IBM-type PCs. That software also is the leverage Grove is counting on to persuade the makers of tomorrow's mainframes to adopt Intel's processors.
But going to MPUs also raises a troubling question for mainframe makers: Would standardized chips foment more ''cloning'' of large IBM or DEC computers, with Japanese and other Asian producers grabbing a bigger slice of the world's computer pie? Moreover, MPUs seem certain to spur more startups into taking flings at the market for big computers, because they would be spared the up-front costs of developing from scratch a CPU and a core library of software. Already, PC vendors such as Zenith Data Systems and Compaq Computer are mapping out strategies to move up into the minicomputer and mainframe range with Intel's chips.
RISC might offer a way to fend off the clone makers. Using it, companies could develop their own proprietary microprocessors, trimming the investment for a super-MPU to 18 months and $20 million or less. Grove's reaction: If anyone really believes such low cost is possible, ''then one of us doesn't know what it takes to do this.'' To develop the 486 took Intel four years and $300 million.
But Intel isn't as dead set against RISC as Grove sometimes seems. The company recently unveiled its own RISC design, the 80960 microprocessor, which is aimed at ''embedded'' jobs -- controls buried inside robots, cars, planes, and electronic systems. Several more RISC chips are in the works, as well. Even the 486 itself will embody some RISC techniques, says Intel's House. ''There will be some surprises when we introduce the 486.''
And Intel thinks it has a better way to guard against clones: a microcircuit ''library'' in which the various segments of the 486 and other chips will be stored as modules. These can then be mixed and matched, and custom modules added, to fashion semicustom MPUs that can't be bought off the shelf. Over the past five years, Intel has invested $250 million in computer-aided design tools to speed development of such semicustom chips, although the project is more than a year behind schedule, and Intel recently scrubbed plans to market a broad family of semicustom chips.
Intel already has one waiting customer for the 486: itself. The company is nurturing a thriving systems business. It will package its chips to supply whatever a customer needs to build a PC, from printed-circuit boards all the way to finished PCs. It is also building big computers under its own name -- a parallel-processing system that can outrace a Cray supercomputer at certain scientific tasks. Nearly 150 of these so-called hypercubes, engineered around eight or more MPUs, have been sold so far.
Now Intel is shooting for a mainframe entry as well. In June it disclosed a 50-50 joint venture with Siemens to produce a line of ''fault-tolerant'' computers aimed at banking, military systems, and other applications where a computer crash would be catastrophic. The new company, called Biin (rhymes with wine), is the upshot of a secret project going back five years. Biin's first system could appear in October.
BETRAYAL. Intel's surprisingly successful upward integration pleases Wall Street. The company's systems business is actually growing faster than chips, climbing past 27% of total revenues this year and heading for 33% by 1990, predicts analyst Adam F. Cuhney at Kidder, Peabody & Co. But to some Intel customers, the new thrust smacks of betrayal. Says Michael S. Swavely, Compaq Computer's marketing vice-president: ''We don't like them making it easy for others to get into the PC business.''
If Intel's MPU gamble does falter, it won't be the first time the company has stumbled. After pioneering DRAMs, it withdrew in the face of Japan's brutal onslaught in the early 1980s. It backed out of static RAMs, too, and is slipping in EPROMs. ''One by one,'' says Campbell of Chips & Technologies, ''they've given away all those markets except the microprocessor.'' Chairman Moore says some business lines were dropped because Intel saw more promise in microprocessors and couldn't afford to invest in too many fast-growing markets at once.
Another reason was that the industry's premier design house didn't measure up in manufacturing. Grove and Barrett have attacked this deficiency rigorously, even ruthlessly. Since 1984 the company has closed eight outdated plants and weeded out more than 6,000 workers and managers -- nearly 25% of its workforce. It is spending a record $450 million this year on cutting-edge equipment and new plants. As a result, revenues per employee, which for a decade had been bouncing around at a mediocre $40,000 to $55,000, by last year had jumped to more than $100,000.
Intel is also still living down a reputation for arrogance, despite major improvements in customer service. ''They used to say, 'Here's what we've got -- why don't you buy some?' '' recalls John P. Frank, president of Zenith Data Systems. Then Intel's Hootnick pushed to remedy the situation -- with telling results over the past year, says Frank.
Some Intel-watchers worry more about the pending change in management than in processors. Both Moore and Noyce are about 60, and Noyce is now dividing his time between Intel and Sematech, an industry research consortium. After Grove relinquishes his CEO title, he might stay on as a vice-chairman, but he wants to devote more time to national issues, such as competitiveness, and try his hand at a novel. His two books on management -- High-Output Management and One-on-One with Andy Grove -- were widely praised.
MODEL T MAINFRAME. Grove has provided no clues to his choice of successor, though outsiders give the edge to House because of his bang-up job with the company's core microprocessor business. House, 45, has a varied background that would serve him well as helmsman. He is an engineer by training and just a thesis away from a PhD in electrical engineering, but he has now spent more than a decade in marketing and strategic planning. Before joining Intel in 1974, he was involved in developing computers at Raytheon, Microdata, and Honeywell -- a definite plus for a company coveting a bigger role in the computer industry. Unless House stumbles soon, he should remain the favorite -- because insiders say Grove has promised to name his successor next year.
So a lot is riding on how good the 486 turns out to be. Moore is confident the chip will be the foundation for continued growth. ''In the 19th century, if you wanted to travel far, you hopped on a train'' and rigidly followed its steel rails, he notes. It's the same with today's big, shared computers. But the coming of the 486 promises the free-wheeling joys of the automobile age.
Updated Aug. 25, 1997 by bwwebmaster
Copyright 1988, Bloomberg L.P.