The Revitalization of Everything: The Law of the Microcosm

From Ronald Reagan’s White House to the centers of French socialism, from the speeches of Democratic liberals to the pages of Britain’s Economist, one assumption about U.S. technology has long held firm: a key American asset is the startup culture of Silicon Valley and similar centers of entrepreneurship. In the late 1980s, however, the heralds of entrepreneurship as the key to American competitiveness have faced a serious embarrassment. A protest rang out from the leaders of Silicon Valley itself that far from an asset, the startup culture had become American industry’s gravest weakness in competing with Japan. Intel Chairman Gordon Moore denounced the venture capital financiers as the “vulture capitalists” of Sand Hill Road, the Palo Alto promenade of venture funds. Intel President Andrew Grove added that his company was threatened as much by Wall Street as by Japan. Not only did American vultures and headhunters prey on his best people, he said, but also the stock market appraised the fabs and fantasies of Intel defectors at many times the value of the real chips and achievements of Intel itself. In a memorable encounter at a Dataquest conference, Jerry Sanders, chairman of Advanced Micro Devices, confronted Pierre Lamond, once a leading figure at National Semiconductor and now a leading venture capitalist. “You took my best guys to Cypress,” Sanders charged, “and they made 3 million last year on 14 million in sales. Fast static RAMs. Big deal. Static RAMs are not exactly a major innovation. [Intel introduced them in 1972.] But so far, so good. The problem is what they’ll do next. They succeed by being focused. But all these small firms want to get big, and that means unfocused. After the first 50 million, believe me, it gets much harder. You can’t fill large fabs with niche products. So far these companies have just ripped off the big firms without contributing anything significant.”1 Lamond tried to laugh off the complaints of Sanders and others as merely the maundering of old men who want the world to stop turning as soon as they get on top. That’s an understandable response from today’s young entrepreneurs. But amid the late 1980s carnage in microchips, with the Japanese sweeping ahead in some measures of market share, the easy answers beg the question. The leaders of the flagship Silicon Valley companies were bringing a serious message: in the future, success would depend on cooperation and cheap capital, planning and political savvy, lifetime employment and government research. To the entrepreneurs, it was as if Julius Erving had accosted Michael Jordan and solemnly informed him that his basketball career was over unless he learned not to jump. Weakened by “chronic entrepreneurialism” (the coinage of Robert Reich of Harvard), the theory went, many semiconductor companies were headed for catastrophe. “The Japanese are systematically destroying the U.S. industry, piece by piece,” Grove claimed. “We are in a demolition derby and some of us are going to die.” For the Silicon Valley patriarchs, the remedy was intervention by the U.S. government to reshape and subsidize the industry. Their ideas bore fruit in Sematech, a chip manufacturing research and development consortium to be funded half by industry and half by government. Although Sematech could help U.S. manufacturers currently damaged by noncommercial Pentagon priorities and personnel demands, the idea of industrial policy seemed to vindicate the leading scholars of the Left. Such figures as Reich, Lester Thurow of MIT, and Chalmers Johnson of Berkeley have long maintained that whether mobilizing for war or for peace, government nearly always plays the leading role in evoking, developing, and financing new technology. Entrepreneurs can best serve as niche suppliers or subcontractors to major corporations. In the face of Silicon Valley’s success, surging forward with ever greater force as government business dwindled, their arguments seemed unconvincing. In the late 1980s, however, a new cadre of more sophisticated analysts has arisen to buttress the argument—most prominently a young scholar at MIT named Charles Ferguson, who emerged as a vocal Cassandra of American microelectronics.2 A former consultant at both IBM and LSI Logic, Ferguson shows intimate knowledge of the computer establishment and the semiconductor industry. His case against American entrepreneurs rings with conviction and authority. His conclusion is stark. Echoing Grove, Sanders, and Charles Sporck, president of National Semiconductor, he declares that without urgent and decisive government action, American semiconductor producers cannot compete in global markets. U.S. merchant semiconductor companies will perish, and U.S. production will be reduced to a few specialized plants at IBM, AT&T, and other large enterprises that supply chips solely for their own use. Since semiconductors form the heart of all advanced computer, telecom, and defense technologies, the failure of U.S. microelectronics will cripple the American economy and threaten the national defense. With an eye cynical but shrewd, Ferguson describes the life cycle of a typical startup in America’s information industry: It begins with a breakaway from a big company, often an entire team (including engineers, executives, and marketers) lured by prospects of wealth far beyond what their employer can provide. From Sand Hill Road they win an investment of up to $ 20 million. They lease equipment, office space, and sometimes even a plant. They hire additional engineers and workers from other companies. With frenzied effort and commitment, they launch a new device—a CAD system, a nonvolatile memory, an application-specific chip, a hard disk drive—mostly based on work at their former employer and competitive with its product line. Weakened by their defection, the large company cannot respond quickly. The new device outperforms all others in the marketplace and sparks massive orders, which the startup cannot fill. But “help” is on the way. A horde of imitators, also funded by venture capital (54 disk drive manufacturers, 47 gate-array companies, 25 workstation builders, and 11 nonvolatile-memory makers), rush into the market. The startup in question struggles to diversify and escape the stampede. Expenses soar. Mezzanine financing is needed from the venture funds. They are too deeply committed to deny it; the startup gets several millions more. The business teeters but clings to its lead. Finally, if all goes well, the company at last concocts a small quarterly profit and, with the hungry enthusiasm of venture capitalists, underwriters, lawyers, accountants, and consultants, proceeds to a public offering. Nonetheless, there are already key defections. In part fueled by them, the competition—whether other startups or major players—has become more powerful. The new company dauntlessly forges ahead, expanding its manufacturing capabilities and its marketing effort. The total investment rises toward $ 200 million. But the mood has decisively changed. The stock price is slumping and the initial frenzy of work has flagged. Disgruntled shareholders launch a class action suit, alleging they were misled by the prospectus and demanding several million in damages. (They, or at least their lawyers, will get it.) To add injury to insult, Hitachi and Toshiba are threatening to enter the market. The venture funds have found a new fashion; Wall Street is bored by semiconductors or CAD; the top engineers are restive as they gloomily watch their stock option values decline. Almost overnight, the company begins to age, and the paper millionaires who run it find themselves suddenly poorer. They are deep in debt and the IRS is attacking some of their tax shelters, used years earlier. They are slipping behind on their deadlines for the next product generation, the mortgages on their Los Altos houses, even the next payments on their platinum credit cards. Desperate for further funds, denied in the United States, they turn abroad. Another way of putting it is to say they decide to sell out. They resolve to put their company—chiefly the new technologies it owns—on the market. In the last several years many American startups have indeed sold out to Asia when the going got tough. The American system says, “So what.” But Ferguson says this is disastrous. In selling where the buyers are, Japan or Korea, the entrepreneurs are not selling just their own assets; they are selling the public assets of America. Embodied in the business are not only the owners’ work and ingenuity but also the value of their experience at a large company, the value of the damage inflicted on it after they broke away, and the value of subsidized research conducted at U.S. universities. In addition, there is the value of the licenses, patents, standards, and architectures embodied in the product and a vantage point in the U.S. marketplace. Finally, there is the value of tax code subsidies favoring small and growing enterprises that can show rapidly rising capital gains, R&D expenditures, tax-loss carryforwards, and investment tax credits. And although large corporations exploit tax benefits more massively and resourcefully than small ones do, most of Ferguson’s claims are plausible, if overwrought. The capital gains “subsidy,” for example, is now a penalty in proportion to inflation. In any case, the tax advantages of venture capital do somewhat inflate the costs of established companies of all sizes that have to bid for land, labor, equipment, and capital against the startups. Trained people become less valuable (if they learn anything special they may leave) and also more expensive (they can threaten to leave). While the entire industry must share the costs of the startup sector, according to Ferguson, the U.S. economy wins few compensating benefits. Ferguson is not content merely to criticize U.S. entrepreneurs; he has a far larger argument to make. It is not the fault of the entrepreneurs that they betray their country and their industry. It is the fault of the crippling limits of laissez-faire economics. Personal optimization is suboptimal for the country and the industry. Entrepreneurial self-interest leads, as by an invisible hand, both to the wealth of persons and to the poverty of nations. This was not always true, Ferguson acknowledges. There was a time when the players could jump and slam-dunk on Wall Street without hurting the industry. But regardless of historical successes, American entrepreneurialism now conflicts with the most profound and powerful trends in information technology. In the past, the semiconductor, computer, telecommunications, and defense industries could proceed separately, spearheaded by small enterprises. But with the rise of chip densities to millions of transistors, integrated circuits now embrace entire systems. Bringing together the once separate worlds of computers, factory automation, and telecommunications, these systems require technical, financial, and even political resources that small companies cannot muster. According to Ferguson, these changes spring from a basic law of information technology. In design, assembly, training, testing, networking, or marketing, costs rise with complexity and complexity rises with the number of objects to be managed. For example, the number of connections between nodes of a network increases by the square of the number of nodes. This rule applies remorselessly, whether to the proliferation of transistors on high-density chips, of components in complex computer systems, of lines in software code, of terminals in a data processing network, or of human interaction on a factory floor. Complexity swamps all systems made of large numbers of simple components, such as the American semiconductor industry produced in its heyday. Moreover, as Ferguson might add, the same law applies to semiconductor companies, which themselves comprise an industrial system of uncoordinated and narrow-based components. This system will suddenly be swamped by a tsunamic wave of complexity. The only answer, Ferguson maintains, is hierarchy: competitive rules, modular interfaces, and industry standards imposed from the top. This theme is familiar to followers of design pioneers Carver Mead and Lynn Conway, as well as industry observers who have also urged top-down approaches as the answer to the problem of electronics design complexity. Ferguson, though, is not speaking merely of designing a chip or a computer but of designing an industry or even an economy. Modularity and hierarchy must be forced from the top. They require rules and standards to guide the producers of future modules and connectable systems. They require control by large corporations and governments. The heavy cost of developing standards inflicts big penalties on the industry leader. It must make the technical effort to determine the standards and the political effort to impose them. These efforts expand markets and lower prices for the entire industry. But in a fragmented marketplace, the costs cannot be recovered. Unless government or oligopoly power disciplines the industry, the companies that pay for these standards and architectures become sitting ducks for rivals that can avoid these costs. In information technology, Uncle Sam has been the sitting duck. Nearly all standards and architectures in hardware, software, and network interfaces have originated in the United States, largely with IBM and AT&T (or under their sponsorship). These standards collectively cost scores of billions of dollars to develop and propagate. But because of anti-trust laws, lax enforcement of intellectual property rights, and MITI’s prohibition of auctions for licenses in Japan, this huge endowment has become available at far below its cost. Between 1956 and 1978, when most of the key innovations arose, Japan is estimated to have paid some $ 9 billion for U.S. technologies that required—depending on assignment of costs—between $ 500 billion and $ 1 trillion to develop. The real cost may have been even higher, however, because by setting critical design targets, standards shift the focus of competition from the invention of products to their manufacture on a global scale. Because manufacturing is itself a complex system, power again flows upward. A globally competitive plant for high-volume commodity production costs some $ 250 million. Requiring integration of an array of technologies—sophisticated computer controls and design systems, advanced magnetics and vacuum equipment, materials science and photonics, automation and robotics—manufacturing imposes huge economies of technical scope. At the same time, whether in integrated circuits, integrated CPUs, networked terminals, computers on-line thousands of miles apart, or telecom networks around the globe, communications, rather than components, increasingly define the systems. Interconnections dictate the structure of the solution. These constraints necessarily push the power out of small companies into companies large enough to define the entire system. As Ferguson concludes, the technologically optimal form for the global information industry might well consist of IBM, AT&T, and a few companies in Japan. Oligopoly is optimal, not only for the competitive advantage of particular nations but also for the welfare of the world. Yet the United States remains afflicted with an extraordinarily fragmented information arena: a chaos of companies, architectures, operating systems, network protocols, semiconductor processes, design systems, and software languages, all controlled by entrepreneurs avid for personal wealth or by sophisticated employees poised to flee to the highest bidder, whether a venture capitalist or a foreign conglomerate. Except at IBM and possibly AT&T, American electronics defines or defends no larger interests. And as Ferguson and the Silicon Valley patriarchs insist, the law of complexity pushes decisions ever higher up the hierarchy and dictates ever larger organizations. The system is sick, and chronic entrepreneurialism is the disease. The remedy, Ferguson, Reich, and others agree, is an onslaught of government subsidies and guidance favoring established businesses and counteracting the bias toward new companies. In essence, these industrial policy scholars want the United States to copy Japan as it is popularly viewed—boasting a powerful combination of MITI and favored industries. But they fail to see that in every one of the manufacturing industries that the Japanese have dominated, they have generated far more companies and more intense domestic rivalry than the United States. The Japanese created, for example, at least four times as many companies in steel, autos, motorcycles, and consumer electronics, and six times as many companies in robotics as the United States. It was such established American oligopolies as GE and RCA, General Motors and Ford that succumbed to the more fragmented and entrepreneurial Japanese. Indeed, the only key industries in which the United States launched appreciably more companies than Japan were computers and semiconductors. Despite Japan’s early lead in transistor radios and in mass production of consumer chips, Uncle Sam still prevails in computer and semiconductor hardware and software. What needs to be explained is not Japan’s success in microelectronics but America’s success. For the critics, the answer is the Pentagon, the initial big customer for computers and integrated circuits. But in fact, during the post-World War II decade when defense dominated technology and confiscatory U.S. tax rates stifled entrepreneurship, American manufacturers lost the lead to Japan in transistor mass production and to the Soviet Union in missiles. The semiconductor industry did not burst into its upward spiral until the 1960s, when tax rates were slashed and the military share of industry sales dropped some 80 % . The secret of the U.S. success was the very venture system that the critics condemn. It counteracted high capital costs by efficiently targeting funds, released energies that were often stagnant in large companies, attracted a crucial flow of inventive immigrants, and fostered a wildfire diffusion of technology that compensated for the lack of national coordination. The American strength derives not from the law of complexity but from the law of the microcosm. This law, discovered by Carver Mead of Cal Tech in 1968, says that complexity grows exponentially only beyond the chip. In the microcosm, on particular slivers of silicon, efficacy grows far faster than complexity. Therefore, power must move down, not up. In volume, anything on a chip is cheap. But as you move out of the microcosm, prices rise exponentially. A connection on a chip costs a few millionths of a cent, while the cost of a lead from a plastic package is about a cent, a wire on a printed circuit board is ten cents, backplane links between boards are on the order of a dollar apiece, and links between computers, whether on copper cables or through fiber-optic threads or off satellites, cost between a few thousand and millions of dollars to make. As a result, efforts to centralize the overall system drastically reduce the efficiency of computing. When Ferguson says that complexity increases by the square of the number of nodes, he actually makes a case not for centralized authority but for its impossibility. Beyond a certain point comes the combinatorial explosion: large software programs tend to break down faster than they can be repaired. For communications, connections may multiply. But coordination and command impose impossible burdens on centralized systems. These principles remain obscure to many observers because they conflict with a powerful tendency in the economics of data storage and transfer. When John von Neumann invented the architecture still dominant in computers today, the industry operated in the macrocosm of electromechanics. In the macrocosm, active elements (switches) tended to be more expensive than storage devices. The data processors or switching elements were fragile, complex, and expensive vacuum tubes with tungsten filaments. The storage devices were either throwaway paper punch cards or magnetized plastic. Most important of all, interconnections were made of virtually cost-less copper wire. Von Neumann’s computer architecture reflected these relative costs of paper, plastic, wire, and complex tubes. He economized on vacuum tube switches for processing and was lavish with wires and storage. Made of different materials, moreover, these components had to be kept separate. Although communication by wire between them was easy, the interfaces among the separate functions—converting data from the processor to the punch card, tape, or magnetic device, and back again—required complex and expensive peripheral driver, sensor, controller, and converter systems. As a fixed cost of a computer, such electromechanical peripherals enhanced the economies of scale for centralized processors, with large storage facilities attached. For example, the vacuum tube drivers, sense amplifiers, and other support for early magnetic core memories at IBM weighed several thousand pounds and were hundreds of times heavier than the memory core planes themselves. Since the cost of the peripherals was quite independent of the size of the memory, small amounts of storage cost hugely more per bit than large amounts. All these conditions of a still essentially electromechanical computer industry favored the mystique of IBM and its von Neumann mainframes. A famous market survey estimated a total world demand for about 50 computers. The machine sat on a pedestal in the central processing room, guarded by a credentialed guild of data processing gurus. These conditions led experts to see the computer as a leviathan instrument of Big Brother and to prophesy the emergence of huge organizations to manage its powers. Predictions of a totalitarian 1984 gained credibility from the image of Big Blue in suits of gray. Gravitating to the hands of the state as well as other large bureaucracies, the computer would give the state new powers to manage the economy and manipulate individuals. Students in the late 1960s wore badges parodying the familiar punch card warning: “I am a human being. Do not spindle, fold, or mutilate.” At Berkeley and other university campuses, Luddite rebellions broke out against this soulless new machine. South in Pasadena, however, Carver Mead’s exploration of the physics of solid-state computation had led him to completely contrary conclusions. He contended that there were no limits to the miniaturization of switches except spontaneous quantum tunneling. Features on a chip could be reduced from the 20-micron dimensions of the day all the way down to a quarter of a micron. In other words, Mead was predicting millions of transistors on one chip. When that happened, he reasoned, switches would be nearly free—merely an infinitesimal transistor on a silicon substrate. Working memory could be made of the same switches on the same material. Wire would become expensive, clogging the silicon surface with alien metals hard to lay down and prone to electromigration from heat or high currents. Leads to the outside world from the tiny chip would be a serious problem. These developments capsized every key assumption of the von Neumann paradigm. It was prodigal with wire and remote memory, and focused all processing in one supremely fast central unit. It favored giant computers. In concept totally obsolete, it survives today, only slightly modified, as the dominant computer architecture. In the new regime, however, the von Neumann components—storage, interconnect, and processor—all are made of the same materials by the same process. The entire computer has become increasingly solid state. Working memory is imprinted on silicon. Even the controllers to mass storage disk media are solid-state integrated circuits and monolithic analog-to-digital and digital-to-analog converters. Unlike the electromechanical peripherals of magnetic cores, these silicon peripherals cost radically less the greater the volume of production. In other words, the more computers built and sold, the cheaper the memory access. Controllers have plunged down the learning curve and now cost just a few dollars, an amount suitable for personal computing. The premium for separation of functions and for scale of operation and storage has given way to a still greater premium for integration and miniaturization in a world of widely distributed processing power. Provided that complexity is concentrated on single chips rather than spread across massive networks, efficacy in the microcosm does indeed grow far faster than complexity. The power of the chip grows much faster than the power of a host processor running a vast system of many computer terminals. The power of the individual commanding a single workstation—or small network of specialized terminals—grows far faster than the power of an overall bureaucratic system. The power of entrepreneurs using distributed information technology grows far faster than the power of vast institutions attempting to bring information technology to heel. Rather than pushing decisions up through the hierarchy, microelectronics pulls them remorselessly down to the individual. This is the law of the microcosm. This is the secret of the new American challenge in the global economy. The new law of the microcosm has emerged, leaving Orwell, von Neumann, and even Charles Ferguson in its wake. With the microprocessor and related chip technologies, the computing industry has replaced its previous economies of scale with new economies of micro-scale. Nonetheless, the new rules are not well understood. In the last five years, IBM, Digital Equipment, Hewlett-Packard, and most of the remnants of the once awesome BUNCH (Burroughs, Univac, NCR, Control Data, and Honeywell), as well as the entire Japanese computer industry, time and time again have foundered unwittingly on these new microelectronic facts of life. While multibillion dollar industries emerged in networked workstations, led by Sun and Apollo, and in minisupercomputers, led by Convex, Alliant, N-Cube, Multiflow, and many others, the leading companies still upheld the von Neumann principle that computing should be centralized and serial rather than parallel and distributed. Even DEC allowed a $ 3 billion workstation industry to erupt in the midst of its market rather than give up its concept of the time-shared VAX-run network. But the force of change has been inexorable. By 1987, measured in millions of instructions per second per dollar, the new Intel 386 microprocessor-based desktop computers were 90 times more cost effective than mainframes. Mainframes remained valuable for rapid and repeated disk access and transaction processing for banks and airlines. But small computers were barging in everywhere else. Distributed peer networks of specialized machines and servers were becoming ever more efficient than hierarchical networks run by mainframe or minicomputer hosts. In many special applications, like graphics and signal processing, embedded microprocessors outperform large computer systems. The computer on a chip costs only a few dollars and far excels the computer on a pedestal. The law of the microcosm even invaded the inner sanctum of the von Neumann legacy: the organization of individual computers themselves. The need for change was manifest in the famous Cray supercomputer, still functioning powerfully in laboratories around the globe. Judging from its sleek surfaces and stunning gigaflop specifications (billions of floating point operations per second), the Cray appears to be high technology. But it hides the scandal of the “mat”: remove the back panel and you will see a madman’s pasta of tangled wires. The capacity of these wires—electrons can travel just nine inches a nanosecond—is the basic limit of the technology. The Mead mandate to economize on wire while proliferating switches led inexorably to parallel architectures, in which computing jobs were distributed among increasingly large numbers of processors interconnected on single chips or boards and closely coupled to fast solid-state memories. Such machines sometimes outperformed von Neumann supercomputers in specialized applications. A symbolic victory came when HiTech, a parallel machine made of $ 100,000 worth of application-specific chips, challenged and generally outperformed the $ 15 million Cray Blitz supercomputer in the world chess championships for computers. As circuits become denser, costs decline exponentially. The very physics of computing dictates that complexity and interconnections—and thus computational power—be pushed down from the system into single chips where they cost a few dollars and are available to entrepreneurs. This rule constrains the future of even IBM. Ralph Gomory, IBM’s chief scientist, predicted in 1987 that within a decade the central processing units of supercomputers will have to be concentrated within a space of three cubic inches. The supercomputer core of the 1990s will be suitable for a laptop. Following microcosmic principles in pursuing this goal, IBM last December joined the venture capitalists and sponsored a new company founded by Steve Chen, a computer designer who had split from Cray. Chen plans to use massive parallelism to achieve a hundredfold rise in supercomputer power. Projecting conservatively from current trends, James Meindl, the magisterial former head of Stanford Electronics Laboratories, has predicted that by the year 2000 a billion components will be packed on a single chip. That will mean a drop in feature sizes to the one-quarter of a micron prophesied by Carver Mead in 1968. A billion transistors will offer the computing power of 20 Cray 2 CPUs on one chip, or 20 VAX minicomputers together with all their memory. This kind of projection, however, assumes little change in the von Neumann architecture. That architecture will be abandoned for many purposes, yielding a far greater enhancement of computer power than even Meindl’s estimates suggest. But there will be one key limit: stay on one chip or a small set of chips. As soon as you leave the microcosm, you run into dire problems. The physics dictate that when feature sizes drop, the voltages needed drop proportionately, the speed rises proportionately, the area needed for a computing element drops geometrically, and the switching energy and heat dissipation drop exponentially. Everything improves but the wires. Even for on-chip connections, the resistance and capacitance and delay rise in accordance with the resistance-capacitance product. For off-chip communications, the delay rises by a catastrophic exponential of the reduction in feature size. Restriction of the computing power to one or a very few chips will enable the promise of the new technology to be fulfilled. With fiberoptic networks and even with superconducting interconnects (recently demonstrated at IBM and several other laboratories to outperform fiber-optic threads by a factor of 100), we can confidently expect mastery of such challenges as continuous speech recognition, interactive three-dimensional graphics, complex expert systems, and real-time modeling of many natural processes. What we cannot expect is centralized command and control. Networks will function chiefly for communication of results; only at the cost of huge inefficiencies will they function for centralization and coordination of computing power. This constraint probably prohibits a centralized strategic defense system, for example, but permits very cheap and effective local networks of interceptors. Above all, the law of the microcosm means that the computer will remain chiefly a personal appliance, not a governmental or bureaucratic apparatus. Integration will be downward onto the chip, not upward from the chip. Small companies, entrepreneurs, inventors, and creators will benefit greatly. But large, centralized organizations will tend to lose relative efficiency and power. Predicting a continuing convergence of the computer world with the telecommunications industry, Ferguson implies that the oligopolistic tendencies of telecom will prevail in computers. But the flow of influence is in the other direction: the same law of the microcosm that is transforming the computer industry is already shaking the telephone network. Though afflicted by obtuse regulators fearful of monopoly, the telecom establishment is actually under serious attack from microelectronic entrepreneurs. Once a pyramid controlled from the top, where the switching power was concentrated, the telephone system is becoming what Peter Huber of the Manhattan Institute calls a “geodesic network.” Under the same pressure Carver Mead describes in the computer industry, the telecommunications world has undertaken a massive effort to use cheap switches to economize on wire. Packet switching systems, private branch exchanges, local area networks, smart buildings, and a variety of intelligent switching nodes all are ways of funneling increasing communications traffic onto ever fewer wires. As Huber explains, the pyramidal structure of the Bell system was optimal only as long as switches were expensive and wires were cheap. When switches became far cheaper than wires, a horizontal ring emerged as the optimal network. By the end of 1987, the effects of this trend were increasingly evident. The public telephone network commanded some 115 million lines, while private branch exchanges held 30 million lines. But the public system was essentially stagnant, while PBXs were multiplying at a pace of nearly 20 % a year. Moreover, increasingly equipped with circuit cards offering modem, facsimile, and even PBX powers, the 33 million personal computers in use in the United States can potentially function as telecom switches. Personal computers are not terminals, in the usual sense, but seminals, sprouting an ever-expanding network of microcosmic devices. AT&T and the regional Bell companies will continue to play a central role in telecommunications (if they are released from archaic national and state webs of regulation). But the fastest growth will continue to occur on the entrepreneurial frontiers at the fringes of the network. And inverted by the powers of the microcosm, the fringes will increasingly become central as time passes. One of the leaders of the assault on the Bell pyramid is Jerry Sanders of Advanced Micro Devices, who is focusing his company on providing chips for the telecom markets. Sanders once famously declared that semiconductors would be the “oil of the eighties.” Ferguson, Reich, and others now fear or favor giant corporations conspiring to monopolize chip production as OPEC once cartelized oil production. They predict that by dominating advanced manufacturing technology and supplies, a few companies will gain the keys to the dominant technologies of the information age. Unlike oil, however, which is a material extracted from sand, semiconductor circuitry is written on sand and its substance is ideas. The chip is a medium for ideas, much like a floppy disk, a CD-ROM (compact disk-read only memory), a 35–millimeter film, a phonograph record, or a videocassette. All these devices cost a few dollars to make in volume and all ultimately sell for the value of their contents—the images and ideas they bear. A memory chip sells for a couple of dollars; a microprocessor that costs about the same amount to make may sell for as much as an ounce of gold. A blank CD-ROM also costs a couple of dollars, but if it contains the operating system for a new IBM mainframe, Fujitsu may have to pay $ 700 million for it. In the microcosm, the message remains far more important than the medium. To say that huge conglomerates will take over the world information industry because they have the most efficient chip factories or the purest silicon is like saying Canadians will dominate world literature because they have the tallest trees. In appraising the predictions of the prophets of oligopoly, we have a powerful test. Most of the recent spate of predictions originated in 1985. Since then, we have seen three years of change and turmoil in the industry. Between 1982 and 1987, a record 97 semiconductor companies emerged. If the prophets were correct, these companies would be in trouble today. But the vast majority are already profitable, their total revenues now approach $ 2 billion, and their success ratio far exceeds the performance of the startups in the supposed heyday of the industry between 1966 and 1976. By most standards, the average productivity of the newcomers is far higher than the output of established companies. If the doomsayers were correct, few of these companies could afford to build semiconductor factories. But in 1986 and 1987, some 200 minifabs were constructed for making leading-edge, custom semiconductors. Some of these facilities offer the most efficient, advanced submicron processes in the industry. Contrary to widespread contentions that the industry is growing ever more capital intensive, the efficiency of the new production gear far exceeds its incremental costs. The appropriate measure of production in information technologies is not output of “chips” or even “computers.” Chips double in capacity every two or three years and computers gain power at a similar pace, retarded only slightly by the slower advance of software. Nor is the best measure the total output of transistors; although this index is an improvement over unit sales of chips, it misses the advances in design. The appropriate standard is not output in dollars either, for the cost per unit is plummeting while the real value of the output soars. The correct measure is functionality or utility to the customer. In delivering functions cheaply to the customer, the new semiconductor production gear every year advances decisively over its predecessors. For example, Chips & Technologies, founded in 1985, is the fastest growing company in the history of the industry. Its sales per employee are some $ 650,000, or four times the level of most established companies. Its product is a series of chip-sets for cloning IBM AT and 386-based computers with a small fraction of the number of chips used by IBM. This innovation allowed Tandy, PCs Limited, and other U.S. companies to regain dominant market share with clones of IBM standard computers. Together with another startup called Weitek (providing fast math coprocessors), Chips & Technologies also helped Compaq keep the lead against both IBM and the Japanese. Compaq is now a $ 1.2 billion business. Chips & Technologies and Weitek have avoided direct fabrication of chips, farming out their production to Japanese and American companies that have excess fab space. But other startups, like Cypress, Performance Semiconductor, and Integrated Device Technology, have created some of the most advanced fabs in the industry. Lamond’s Cypress—attacked by Sanders—has decisively swept by the $ 50 million barrier. Xicor has mastered a nonvolatile memory production process that has so far defied the best efforts of Intel and the Japanese. And Micron Technology, a 1980s startup nearly destroyed in the DRAM glut of 1985 and 1986, has made several key process innovations (reducing the size and number of layers on a DRAM to some 40 % below most of its rivals) and maintains production costs below its Japanese competitors. The first successful billion-component chip will be designed, simulated, and tested on a massively parallel desktop supercomputer that will yield functionality far beyond its cost. Those 20 Crays of computing power will make the chip incomparably more potent than current microprocessors that take scores of designers to create and are built in a $ 150 million factory. It will be manufactured on a laser direct-write system or an X-ray stepper that, dollar for dollar, will far outproduce its predecessors. Doubtless all the business magazines will still be speaking of the incredible expense in building and equipping a semiconductor plant—the X-ray stepper, for instance, will cost three or four times what today’s lithographic stepper costs. But in functional output per dollar of investment, the fab of the year 2000 will be immensely more efficient than the Goliath fab making DRAMs today. By the law of the microcosm, the industry will continue to become more productive in its use of capital, more intensive in its mastery of the promise of information technology. Finally, if the doomsayers were correct, the United States would still be losing global market share in electronics. But since 1985, U.S. companies have been steadily gaining market share in most parts of the computer industry and more than holding their own in semiconductors. In mid-1987, the U.S. semiconductor open market (excluding IBM and other captive producers that consume a third of U.S. output) again exceeded the total Japanese market in size (the Japanese have no companies categorized as captive). A year of Japanese leadership in that index had led to many prophecies of doom for the U.S. merchant companies. The drive to predict the demise of chronic entrepreneurialism continues strong in intellectual circles. Many critics of capitalism resent its defiance of academic or social standards of meritocracy. William Gates, the founder of Microsoft, left Harvard for good during his sophomore year. The system often gives economic dominance to people who came to our shores as immigrants with little knowledge of English, to all the nerds and wonks disdained at the senior prom or the Ivied cotillion. Some academics maintain a mind-set still haunted by the ghost of Marxism. But the entrepreneur remains the driving force of economic growth in all vibrant economies, including the U.S. economy, the most vibrant of all. 1 Sanders’ complaint is echoed in Alex d’Arbeloff and Frederick Van Veen’s HBR article, “Stop Taxing Away Big Companies’ Talent,” May–June 1986, p. 38. 2 His MIT thesis, International Competition, Strategic Behavior, and Government Policy in Information Technology Industries, encompasses his ideas.

The Revitalization of Everything: The Law of the Microcosm

Partner Center