GOLIATH IN THE VALE of Elah roared his contempt at the weapons and zeal of David: “Do you think me a dog that you contest me with sticks and stones?”
Bill Gates, the Goliath of software, sees himself similarly beset by zealous rivals with risible weapons. Entering his modest second-floor office on the edge of the Microsoft campus on a twilight evening in late November, I find him irked and addled by what he sees as a siege of slingshots from irrational media and capital markets in his industry.
Microsoft entered November on an autumnal blaze of upside news-earnings up 53%, sales up 63%, a double gi-gabuck quarter. But now, amazingly, people were speaking of a “crisis” at Microsoft. “That’s an emotional word,” says Gates, twisting uncomfortably in his chair.
IT WAS BAD ENOUGH when Netscape came on with 9,000 lines of code for a World Wide Web browser that took six weeks to build and that was given away free, and people began talking of his company’s downfall. Gates had pulled that trick himself in 1976, with a few thousand lines of code for the Basic programming language for the PC and then again in 1980 with MS-DOS. But to do it to Goliath seemed lèse-majesté.
Then comes Sun Microsystems with a new programming language called Java, and people like me, who by Microsoft standards don’t know anything about programming, who have never written a single line of code, presumed to tell him about its virtues. It’s safe, secure, interpreted, plat-form independent; it collects your garbage (automatic garbage collection). It compiles as if by incantation. It builds market cap as by magic. Give poor Bill a break.
“Yeah, right,” says Gates. “I have to wonder who screwed your head around….” Oh, he knows, he knows. “The Internet” makes everything different. You can make any claim you want, however bizarre or ludicrous, that would ordinarily be laughed off the stage, and if you add the mantra “on the Internet” at the end you can morph yourself from a typical media clown into a visionary, a prophet, a guru. “Nobody pauses to say, ‘Huh?’ “
Even the analysts will nod and the market will bow. It’s “on the Internet.”
One day in November, three days before my visit, Rick Sherlund, the Goldman Sachs analyst who helped bring Microsoft public in 1986 and had touted the stock for a decade, downgraded Microsoft’s shares from buy to “moderately outperform” on the basis of Internet incantations from Sun and Netscape. Sherlund also replaced Microsoft with Netscape on his recommended list. Indeed, Goliath’s net worth was shrinking by the minute as Microsoft’s market cap sank by $9 billion, and Sun’s and Netscape’s surged by $9 billion in a matter of weeks.
Does Bill Gates know Rick Sherlund, by any chance? Sure, Gates answers, “extremely well.” “He’s your man, he’s great,” says Gates, “if you want to run a spreadsheet.”
In this world of manias and emotions, “I have to make rational decisions,” Gates says, glaring at me. “Somebody who thinks that because of a browser that anyone can clone, because of a language that is magic, they [Netscape and Sun, the unmentionables] can overthrow the world-that person can’t even think two chess moves ahead. You’re not even in the game I’m playing.”
Okay, thinking forward a couple of moves, what is the big thing bearing down on Microsoft on that road ahead?
It’s middle age.
No, in Redmond, they have another name for it. Throughout the company, wherever you go-from hum-mus on pita bread at the cafeteria with a glib former Hollywood agent hired to handle publicity, to a Starbucks latte at the Microsoft model “home” with software sage Rick Rashid incandescent on a couch, or off to Rover’s Restaurant where Nathan Myhrvold spiels refulgently at a corner table through 12 courses of rococo “fatware” and my two missed flights-the word is middleband. From the top down, Microsoft is becoming a middleband company.
Gates, Myhrvold, James Allchin, Craig Mundie- nearly anyone in Redmond will step you through it. The future is not broadband, not narrow-band; it is some middle way. It takes Excel and Word and PowerPoint presentation graphics and multimedia CD-ROMs and the new Video ROMs-indeed all the front and back Office suites that are the core of the company, all the teeming towers of legacy code-and attaches that to the Internet. It’s Encarta and Baseball and digital TV all tied to the Web over middle-band circuits.
Gates explains: “We will translate Encarta into many languages and make it a front end to the Internet, so that when-ever you look up a topic in the encyclopedia, we can link you into what there is in the Internet on that topic.
“Now the Internet is not fast enough when you just want to go pure Internet, so every year you can buy the CD that holds the bulk of the material.” It will be a middleband world. I see the millennium at hand: People are all queuing up at Egghead for their Encarta update CDs.
From Gates on down, however, Microsoft leaders do grasp the essence of the change. Myhrvold trenchantly points out that communications standards no longer rise from the center out, from the LAN to the WAN. They be-gin on the Net and move inward to the LAN and transform it. “The LAN is dead,” as Myhrvold says.
But these same Microsoft leaders seem to believe that the transforming power of the Net stops short of the PC and operating system. Sun will set, but Windows will open wide on the World Wide Web (up to ISDN speeds). After all, as Allchin puts it, “We see the Internet as an extension of the operating system.”
I ask Gates how we can have a middleband world in the face of a rising tide of bandwidth. Gates in the past has spoken of virtually “infinite bandwidth.” But he does not see it today.
Twenty years from now we will have broadband in homes, he explains, but until then middleband is the best you can expect. Indeed, in his presentation to the press the week after my visit, on December 7, he strangely declared “broad-band is the holy grail, [but] it’s much further away than ever before.”
I ask about cable modems. But to Gates, cable modems are mere middleband: “Don’t get me wrong-you can do a lot in middleband. But cable is a shared medium. Cable modems are middleband. You get 10 megabits per second and share it with 500 homes and you are back to ISDN speeds.”
But, I protest, Netscape has a different view. It has joined with @Home (the Kleiner Perkins-TCI joint venture) to supply browsers and servers for a new scalably broadband Internet based on cable.
Gates’s voice reaches a new pitch. “I assure you Netscape has no relationship with @Home that Microsoft does not have. I didn’t spend three years talking with John Malone for nothing three years with Bruce Ravenel [Tele-Communications Technology Ventures’ senior vice-president and chief operating officer]. In the first place, browsers are trivial. We will have cable browsers. And we will have cable servers. We will do anything with @Home that Netscape does.” After all, TCI Technology Ventures invested $125 million in MSN (Microsoft Network).
So it went in the Redmond gloaming. I deeply admire Gates. The guts to leave Harvard at the end of his junior year and launch a new industry; the tenacity to build it into a planetary utility; the audacity and ingenuity of the original deal with IBM; the entrepreneurial confidence to cut loose from OS/2. The vision to be the only major software company to embrace Macintosh and save Apple by endowing the Mac with the leading GUI spreadsheet and word processor, Excel and Word, while at the same time gaining the graphics skills to create Windows. The bold challenge to Unix through Windows NT.
THERE IS NO DOUBT that Gates has been the exemplary business leader of our era. Compared with the leaders of IBM, heavy with “NIH” (not-invented-here) and degrees in business administration and finance, who could not even grasp the concept of sunk costs or the rule of self-cannibalization well enough to burke OS/2, who could not even see the huge opportunity to embrace the Mac OS, Gates is indeed a giant.
A week later, in an announcement emblazoned in Computerworld as “Capitulation,” Gates showed his superiority to the NIH syndrome at IBM. He declared that he was licensing Sun’s Java Internet animation language, in which he seemed suddenly to have discovered new virtues, and was essentially abandoning Microsoft Network as a proprietary paid service. It would serve as an attractive Inter-net entry point, open to all, with content and advertising prepared with anyone’s available software. Beyond that, Microsoft announced an array of impressive-sounding new Internet products: Blackbird publication tools for the Net; Gibraltar Internet server products four times as fast as Netscape’s; Visual Basic as a scripting language for the Net already far easier and more familiar than Java; upgrades of current Word versions that allow direct creation of HTML documents; and an array of other announcements.
But all the brave talk, the bestselling book, the stilted TV appearances, the announcement of a news channel with NBC, the stream of new products, the bold embrace of an Internet strategy, the spread of Windows 95, could not disguise the rising confusion in Redmond.
All of a sudden, Gates seemed to have lost his bearings. The man who elbowed aside an on-air Connie Chung as if she were a bothersome gnat and shunned NBC as a nuisance, now was clutching the old network, of all things, as a source of news and investing in it, as if it had a future. It was as if old NBC with its Max Headrooms of smiling anchor faces and two-minute splashes of “news” could morph into an in-formation resource simply by invoking the mantra “on the Internet.”
As in his long romance with Warren Buffett, Gates seemed to be reaching out to old money, power and prestige to bolster his company as it whirled in the vortex that he had described as the “Internet tidal wave.” It was as if he no longer trusted the PC to sustain his growth as an $8 billion revenue company, as if he needed sustenance from mass media.
With Netscape, Sun, @Home and other firms, Silicon Valley is in the ascendant again. But the software colossus is still losing ground on the road ahead, so Gates pivots on his peerless pinnacle simultaneously at the summit of the New York Times bestseller list and the Forbes 400 and looks back with a Macaulay Culkin smile from the cover of a book that is mostly news of yesterday.
I N T E L O U T S I D E ?
Two weeks later, back in Silicon Valley from the Vale of Elah, I visit Goliath’s prime mover, Andrew Grove of Intel. With revenues more than twice as large as Microsoft’s and a price/earnings ratio less than half as high, and commanding the world’s most awesome manufacturing facilities for the world’s most complex and portentous product, Intel seems to stand on firmer foundations.
Grove opens the meeting with jokes about the “tunnel of death” that perpetually menaces his industry in the pages of the media. Ensconced in a small open cubicle on the fifth floor of the Robert Noyce headquarters in Santa Clara at the heart of Silicon Valley, with “Intel Inside” inscribed on the roof to enlighten the planes from nearby San Jose International Airport, Grove effervesces wit and irony and bonhomie where Gates seethed sarcasm and defensiveness.
With Grove, there is no longing for canonization by old money, no sell-sign craving for the sickly glamour of Hollywood and TV, no fashionable yearning for business in “content.” Grove grasps that the PC is the ascendant force in the global culture of capitalism and that the Internet consummates the PC. Nonetheless, asked about the possibility of the teleputer-the $500 Internet PC-freed from the coils of Wintel, he echoes Gates in a celebration of current PC culture that somehow misses the point.
Grove associates the teleputer with dumb or static appliances, from set-top boxes to PDAs, in the catalog of PC sub-species that emerged in hype as substitute PCs. Most of them sold a few hundred thousand units, and then expired. “The new device will be produced and it too will sell a few hundred thousand units. But not 10 million units,” he says.
As Forbes ASAP editor Rich Karlgaard and I prepare to leave his office, he asks us to guess the cycle rate of the up-coming ’96 basic home-based PC. I suggest 100 megahertz. Grove shakes his head. He confides that he is headed for a meeting to decide whether the 1996 PC will bear a 120- megahertz or a 133-megahertz Pentium. “Everyone underestimates the progress of PCs,” he says with satisfaction.
“You could fix on a special-purpose device today, to satisfy Larry Ellison’s mother [who wants a simpler PC], but by the time it came out, the PC will have moved on, powered by incomparably more potent microprocessors, leaving the new machine trivialized and obsolete in its wake.” In other words, evolutionary products will suffice in this revolutionary time of exponentially expanding Nets and peripheral CPUs. Nonetheless, moving a few miles north on California’s Route 101, out of the hypergravitational fields of Philistea, one can still feel a radical shifting in the spheres of possibility. “And David took his staff in his hand and chose him five smooth stones.” I Samuel 17:40
FOR THE NEW ORDER, the ultrawideband wireless Sand Hill slingshot-the capitalist conjurer of the forces causing new sleeplessness in Seattle-is John Doerr of Kleiner Perkins Caufield & Byers. Shunning Herb Allen’s summits of schmooze, where the entrepreneurial big-time is an “audacious” investment in Coke or NBC, Doerr epitomizes the venture capitalist as industrial demiurge.
As technology investor Roger McNamee puts it: “While other venture capitalists say, ‘Let’s start a company,’ John says, ‘Let’s start an industry.’ ” So far, beginning at Intel in the early years, he has played a key role in launching industries in electronic design automation, RISC workstations, personal computers, financial software, multimedia and wire-less pen appliances (well, let that last one pass). His current new industry will be the biggest yet. It is broadband Internet.
From the vertiginous launch of Sun, Lotus and Compaq in the early 1980s to a fund gushing Go at pens, his career has seen several peaks and valleys. Doerr has even dallied with middle age. He once weighed a mid-life retreat from the madding bustle of Silicon Val-ley to contemplate the Tantra or the Tao: “I sometimes think,” Doerr told the New York Times in 1987, “I would like to become a Buddhist monk.” Sure, John. But, tell me, what yoga discipline was he brewing two years ago in Palo Alto, at breakfast at Il Fornaio head-to-head with Jim Clark? What karmic rites was he conjuring with Clark and Bill Joy in the winter of 1994 on a three-hour conference call among wildernesses of Marriott on the road? What karass was he kenning in December 1994 among the tacky booths and bins, the barkers and indoor bikinis at the Western Cable Show in Anaheim? What tables were tipping in January 1995 in his Woodside home, among pizzas and pastas with his wife, Anne, and Marc Andreessen, Bill Joy, Andy Bechtolsheim, Jim Gosling and Rick Schell (Netscape’s VP of engineering)?
Why, in early 1995, was he lurking around the NASA Ames Research Center at Moffett Field in Mountain View, Calif.? Is he seeking evidence of alien IPOs, or just some hard-core Unix Christian libertarian netbender from outer space to levitate a new industry in Palo Alto?
AT&T venturer Thomas Judge told Forbes ASAP’s Nancy Rutter in 1993: “You have to be on the fringes to make money in [Doerr’s] business, and that’s where he is.” Follow Doerr, however, from day to day, call to call, from Sun to Oracle to America Online to Netscape to Macro-mind, and you will find yourself at the fringes just as they invert into the Zen center of the sphere.
A lean figure, with blondish hair, a cowlick and horn-rims, Doerr at 44 is as bashfully all-business and frenetically bitwise as Gates whom he resembles, but he is still flouting the gravity of middle age. On a crisp Sunday in mid-December, I catch up with him at Buck’s restaurant, near his home in Woodside in Silicon Valley. Wearing a dark suit from church, he checks for messages on his Skytel pager, greets fellow venturer Bill Davidow passing by and then opens a black briefcase full of technotoys. From across the restaurant, this venture colossus looks to be a frowsy salesman perhaps a little desperate to present his wares. He removes a Mac Power-Book 5300 and Sony speakers and lays them out on the table. Amid empty latte glasses and plates of ravioli pesto and his own half-eaten hamburger, Doerr is ready to give a demo of the new industry-a fore-cast for next year’s Netscape-style IPO.
It takes a minute or so to boot up the Mac, checking through the 32 megabytes of RAM (teleputers, Doerr says, will boot up instantly from flash ROM). But from there on out, it is all immediate gratification. Click to ignite a Java Web page with streaming stockmarket data, a c/net talking- head newscaster and a vocano video from Venezuela. “Wow, look at that new PowerBook go,” exults a jolly woman observing from the next table.
But Doerr is on a rush through a world of his own. Click again and you have the Sunnyvale Sun, efflorescent with vivid speech, sports clips and classified personals. Newspapers will be hot on the broadband Internet. Click on the classifieds and you can presumably meet their makers in living color.
This is a glimpse of @Home, a mere demo of Doerr’s new broadband Internet company. As the service develops you will soon be able to download movies and other pro-grams on demand. Over 28.8 modems or even 128-kilobit-per- second ISDN lines, all such dynamic fare would be agonizingly slow to access. By contrast, Doerr says, in @Home everything is instant, full motion and always on.
How can this be, you ask, on an essentially middleband Internet? “That’s the genius of Milo Medin, @Home’s net-work chief,” he explains, “linked to the genius of Marc An-dreessen of Netscape.” NASA Ames’s network king, Medin is now building a scalable, extensible architecture for a cable-based World Wide Web. With some help, I might add, from John Doerr and his five smooth stones.
B R E A K I N G W I N D O W S A N D B O T T L E N E C K S
Indeed, four of the five companies-Sun, Netscape, Macromind and Intuit-Doerr estimates, “have added more than $10 billion to their market cap in the last six months because of their Internet initiatives,” this new model of computing based on the Web, while Microsoft lost a similar total by briefly resisting it. Now @Home is on target to generate another multibillion market cap by exploding the current bandwidth bottlenecks of the Net.
At Intel, Doerr worked at a desk down a corridor from Bruce Ravenel. Then an architect of the 8086 and the 8087, Ravenel is now TCI Technology Ventures’ influential chief of technology. Igniting the fuse for @Home were words between Doerr and Ravenel as the two old grads from “Noyce-Moore U” wove their way through the Western Cable Show in the first week of December 1994. “I dare not call it an epiphany,” Doerr says, “but Bruce and I were at the Motorola booth where they were showing off a sleek little $300 box the size of a modem that would enable telephone calls over a cable line. ‘What would it cost,’ I asked, ‘to add an Ethernet port to the device, so you could link a computer to the Internet through it at up to 10 megabits per sec-ond? ‘ When the Motorola guy guessed, ‘Maybe $30,’ our eyes got as big as saucers.”
When Doerr sees a hole in the line, he hits it hard and fast. Two weeks later, four days before Christmas 1994, he was in John Malone’s office in Denver presenting a plan for a Silicon Valley startup to bring broadband Internet over cable. Likely IPO market value in two years? $3 billion. Malone also hits hard and fast. After a three-hour meeting, the TCI chief signed off on the venture with-out a qualm. Doerr’s chief job was to “line up some unique Silicon Valley technical genius to make it work.”
Tapping Internet experts at Sun and around the Valley, Doerr found only one name popping up on every list. It was that hard-core Unix Christian libertarian netbender from outer space, Milo Medin. A 32-year-old wunderkind manager of the multiple networks converging at NASA Ames Research Center in Mountain View, Medin had spent the last 12 years making increasingly crucial contributions to the Internet’s growth. But uh-oh. For a week or more, Medin refused to answer any calls from Kleiner Perkins. He says, “I thought they were a bunch of lawyers.”
DOERR WAS ATTACKING the key problem of the Internet. With the number of host computers doubling every year since 1970 and the power of the computers doubling every 18 months, the Internet had mastered every challenge of capacity by multiplying cheap local routers and servers. The Net has already overtaken the U.S. Postal Service as a carrier of mail (by one estimate, a trillion e-mail messages compared with 180 billion postal deliveries). And the Net similarly has pushed the number of digital data bits ahead of the total of voice bits on the phone system.
Over the last two years, however, the traffic has taken a turn toward GIFs and graphics, doubling the number of bits every few months. As the Internet careens toward its destined collisions with television and telephony as the prime sources of information, entertainment and communication for the public, the prime obstacle is bandwidth. Many in the industry have begun to blink and bluster in the bright light of optical media and other broadband pipes.
Led by Bill Gates, they believe in middleband and ISDN. They dabble defensively in TV. As a shared medium, even cable, so Gates contends, will dwindle to ISDN rates as the number of customers on the system rises. But at 128-kilobit-per-sec-ond or even at 1.54- megabit-per-second T1 rates, ISDN means picture quality inferior to NTSC television. A shared medium linked to slow routers means bottlenecks throughout the system if cable modems yield a thousandfold increase in bit traffic beyond existing modems (up from 28.8 kilobits to close to 30 megabits per second). On-line services will bog down in slow access, sticky searches, jerky movement, blurred faces.
Such a middleband net will not be able to maintain its current momentum of growth and power. It will not be able to challenge television and telephony. Yet the valuations of Internet companies depend on a continued exponential ascent. Thus many people believe the Net is overhyped, overvalued, starved for “content.” They believe Doerr’s broadband revolution will fail.
Milo Medin is Doerr’s weapon to break the bottleneck. He has spent most of his adult life overcoming crises on the Net. Medin’s fast rise began in 1987 when he led the creation of NASA’s Internet and almost came to an end in 1993, when the swarthy Serb with the spruce mustache and the piping voice and broadband gush had barged into Washington to persuade the entire government to em-brace the Internet protocols (TCP/IP). Fueled by this universal language, the Net had grown exponentially up a wall of worry.
The problems did not reach critical mass until 1994, however, when its traffic began doubling every couple of months. Andreessen’s Mosaic browser had spurred the World Wide Web into a multiterabyte cyclone of growth and expansion, concentrated in California.
Here the entire Internet converged in one room in one building at Ames. On one side of the room were the Cisco 7000 routers and Digital gigaswitches and Northern Telecom add-drop multiplexers of the FIX (the Federal Internet Exchange), the government peering-and-exchange point, managed mostly by Medin. On the other side was MAE West, the Internet access exchange point for most of the private Internet, including the pullulating college and university Nets, the regnant Internet service providers such as ANS and Sprint, and the budding local fiefdoms of Netcom and The Well.
Then in April 1994, the government ended its $25 million in annual Internet subsidies and arranged for a private takeover. Among the winning bidders were Pac-Bell and Ameritech, assigned to run two key network access points (NAPs).
Long expecting the withdrawal of government subsidies to throw the Net into chaos, many observers would welcome the ensuing crash. The private sector would flub the job. There would be a tragedy of the commons. As in feudal Britain, when the commonly owned lands were overgrazed and ruined, the commonly owned Internet would suffer a glut of graphics “GIFraff” and traffic jams. Sure enough it was happening.
PacBell’s ATM switches from Newbridge Networks, with skimpy buffers designed like a PBX for voice-traffic pat-terns, choked. California, the source of two-fifths of Internet traffic, was down. The Western NAP jammed, and Ameritech’s Chicago NAP was also largely out of commission. Bay Area e-mail from a student on Netcom to a small business on BBN’s BARNet had to steer clear of the Ames hub and pinball through routers all the way across the country to Sprint’s NAPs in Pennsauken, N.J., or Reston, Va., and then all the way back again. As a result, the Eastern Inter-net hubs also began to tilt. An obvious solution seemed to be to have FIX West, the government hub at Ames, take over the traf-fic that was fleeing the Pacific Bell NAP.
Savvy residents of Silicon Valley, the Ames management was sympathetic. Then Medin gained the blessing of NSF networking chief Steve Wolff and enlisted Jack Waters at MCI, a crucial Internet backbone supplier that no longer used its PacBell connection.
Within two weeks, Medin created a system comparable in capacity and reliability to the original FIX, with expanded Net management capabilities, power supplies, communications ports and routing facilities. The result was a broadband national peer-ing- and-exchange point, with a cumulative capacity of some 10 gigabits per second. It combined traffic from all the major commercial Internet suppliers with the bitstreams from government laboratories and agencies.
Meanwhile, throughout this period of crisis and turbulence, no ordinary Internet customer experienced any un-toward deterioration of service. Although Medin’s contribution was only part of a major national effort, he became the talk of the Net. Doerr had to sign him up.
AS DOERR SUMS IT UP, “Milo was running the largest IP net in the federal government. When they decided to set up a White House.gov Web site, they asked where to put it. They put it on Milo’s server. Milo helped run the fiber ring around Moscow. Internet connections for Australia and Antarctica and for deep space probes ran through Milo. He was supplying IP connectivity for the entire Scandinavian subcontinent. He had some 200 remote nodes. And he ran it all with some 99.98% uptime.”
@Home CEO Will Hearst of Kleiner Perkins likes to tell a story that gives some clues as to how this young Net nerd from NASA became a legend in his own time: “In 1988, a Finn-call him Lars-hacks his way into Milo’s computers. Ticks Milo off. He does a trace route and finds his way back to the administrator of the domain in Finland. It’s an academic site. Milo already knows Lars’s IP address. You can’t hide from Milo. He says to the administrator, ‘We have a problem. Please have a conversation with Lars.’ That upset the Finns, who say, ‘We are not going to do that! We respect civil liberties here! You can post a complaint if you like, but we can’t tell the guy what to do.’ So Milo goes into a slow boil. Says, ‘I’ll give you about 30 minutes to get that guy’s files off our machine.’
“Nothing happens. So Milo issues an order: ‘Take down Scandinavia.’ The switch is pulled. Three countries go dark. They don’t notice it immediately, but pretty soon e-mail messages are not getting returned. At last, three senior administrators go to Lars, so the story goes, and they say: ‘We don’t care if you hack into the CIA; we don’t care if you bring down NSA; and we don’t mind if you abscond with all the financial bits in the Federal Reserve. But don’t mess with Milo at NASA.’
“The Finns called back Milo, said the situation had been taken care of. Milo said fine and put the service back up.”
NOW DOERR and Medin are again confronting the perennial doomsday adventists who gather on mountaintops of slightly older money and disparage the future of the Net, talking crisis, over-load, overhype, overvaluation. Tragedy of the Commons. The experts are chiming in. From Howard Anderson of the Yankee Group to Andrew Seybold and Bob Metcalfe, lead-ing analysts are prophesying a crash in 1996.
Medin has been there before. The answer to traffic jams on a narrowband Net is creation of a broadband Net. Don’t tell him it is not technically possible. Who are you kidding? This is the age of the telecosm.
Bill Gates, though, thinks it is the age of middleband. It is obvious beyond cavil to Gates that his regime, ruling 80% of the world’s computers, is destined to prevail. He commands a market share so overwhelming that Washington’s antitrusters see it as a monopoly in need of government dissolution. For Gates, among the most ludicrous claims to be validated by the mantra “on the Internet,” is the idea that Windows machines are an inferior minority system difficult to digest in the prevailing habitat of Unix and TCP/IP.
To Medin, however, it is a matter of simple fact that Windows and NT are awkward systems, hard to incorporate in his domains except as mere terminals. To Medin, Unix is the heart of the Internet, the matrix of creativity in net-working, the bearer of thousands of programs and services and tools and scripts and languages that together comprise the pullulating fabric of the rampantly growing Web. So far, Medin has a strong case.
Sixty percent of the managers of Internet host computers use Macintoshes as their preferred personal machine. On the Internet, as a platform for servers, whether for the World Wide Web, e-mail, FTP, Telnet, Gopher or NeWS, Microsoft’s favored NT now ranks seventh, with a 4% share, behind Sun, which commands a 56% share, Apple, Silicon Graphics, IBM, Digital Equip-ment and Windows 3.1.
Around the time that Gates was assuring me of Microsoft’s impregnable position with @Home, Medin was reviewing the Seattle company’s software concepts for his new network. The @Home people wanted to adopt Microsoft’s Explorer browser if they could (TCI favored its interactive TV ally), but it was simply impossible. Explorer ran on neither Unix nor Macs, and could not handle multicasting.
Netscape’s browser already worked with all the existing systems, including the various Windows.
Under the influence of Marc Andreessen, who had learned net-working in the broadband 45- megabit-per-second environment of the National Center for Supercomputing Applications (NCSA), Netscape had long ago left behind all the comforts of middleband. Andreessen was eager for broadband connections. Gates was not even in the game that Medin and Andreessen were playing.
All right, suppose that “browsers are a trivial technology,” as Gates told me dismissively. It was servers that Microsoft really wanted to sell to @Home. Their Gibraltar system was running Microsoft’s somewhat balky internal Internet at a pace some four times faster than Netscape’s server might. Here, Microsoft benefited from its homogeneous cam-pus environment. Netscape had to employ the “union” code-using the lowest-common-denominator instructions to coordinate several varieties of Unix, Mac and Windows NT. Meanwhile, Microsoft could optimize Gibraltar for all the most powerful instructions in Windows NT, so it was much faster.
But the @Home people were perplexed. How could they use Windows NT, an alien system on the Internet, unfamiliar to their employees or to the Internet service-provider personnel who would work the @Home headends and other nodes? Microsoft was behaving in the Internet environment as if the company were still safe in the imperial realms of Windows desktops.
So @Home, which promises to be the most important force in the next phase of Internet evolution, and Medin, the intellectual firebrand at the heart of @Home, seem unlikely to embrace Microsoft’s offerings.
MILO’S DARK SHADOW OVER MICROSOFT
As for TCI, its enthusiasm was dented a bit by the collapse of the MSN project, in which Malone invested $125 mil-lion. But TCI has recouped its loss. Jim Clark was persuaded to offer TCI a small share of the Netscape IPO. After the initial public offering and the subsequent boom in Netscape shares, TCI’s holding was worth $125 million, leaving TCI quite comfortable with its new allies in Mountain View.
A deeper look at Medin’s plans casts darker shadows in the path of Microsoft. Asked about the notion of an Internet computer being free of Windows and other Microsoft levies, Gates stops rocking on his chair and gets to his feet. He turns and paces urgently back and forth across his office. He gesticulates, summons the history of past challenges, refers repeatedly to dumb terminals and other unappetizing machines, and hurls forth rhetorical questions: “Do you want to go onto the Internet when you are doing word processing, do you want to go on the Internet when you are using PowerPoint or Excel?” In other words, do you want to forgo all the wonderful new OLE interactions among Microsoft programs and all the new Microsoft hot links and other forthcoming tools when you go on the Net?
Under the pressure of Gates’s energy and conviction and hypotheticals, I answer, “Of course not.” But the real answer is, “Sure, if in exchange I can have a computer that outperforms a current Wintel machine on the Net and contains linking capabilities comparable to OLE for one-third the price.”
Gates himself sketched out the answer in his famous Internet Tidal Wave memo, issued in May to galvanize his company in the face of the new threat. He pointed out that not only could he access far more information on the Internet, he could also find, search and browse it more readily on the Net than on a LAN or, for that matter, he might have pointed out, Gates’s own hard drive or CD-ROM.
The error of all the critics of the $500 teleputer is their assumption that it will be inferior to current PCs. It will be, they claim, a PC minus a fast CPU, short a high-resolution monitor, without a fast memory or large drive.
This assumption misses the compounding impact of microcosm and telecosm. The advance of chip technology through Moore’s Law, together with the advance of net-work bandwidth, will endow a machine not inferior but hugely more powerful, than the most supercharged Pentium workstation on a local-area network linked to the Internet at ISDN speeds. The Law of the Microcosm ordains that one-chip systems will be better, not worse, than intersecting boards strewn with devices linked by wires and buses. As Wilf Corrigan, chief of LSI Logic, observes, “From calculators to cellular phones, every time a system has moved onto a single chip, it has wreaked havoc with the existing industry.”
IN PREPARING THE WAY for one-chip teleputers, Medin con-cedes that the current Inter-net will not support broadband services. “You link a broad-band modem to the existing Internet and what you get is an impedance mismatch”-a bunch of fire hoses attached to a network of garden hoses. In order to accommodate the fire hoses of @Home, Medin will have to enlarge the band-width of the Net, from the humblest service provider to the NAPs at the top of the net-work hierarchy, where the leading service providers join to “peer and exchange” data.
Leasing capacity from the telephone companies, @Home will create a new broadband network linking to the existing NAPs at MAE East in Tyson’s Corner, Va., at the Sprint NAP in Pennsauken, N.J., and at MAE West in Mountain View. This will expand the capacity of the so-called Internet back-bone (in fact, an ever-shifting array of virtual vertebrae), which currently works with maximum pipes running at 45 megabits per second. Over the next two years, Medin plans to upgrade his backbone to 622 megabits a second.
Most important and revolutionary, though, are Medin’s plans for the local loops and service providers of the Net. Contrary to the claims of many critics that the Internet PC implies a return to the now-discredited model of the main-frame and dumb terminals, @Home resolutely distributes intelligence and memory through the network.
At the heart of the @Home system is ingenious hierarchical memory management and caching to conceal the mazes of slow routers, sluggish switches and narrowband wires that lurk treacherously among the higher reaches of the Internet. Indeed, when Doerr finally got through to Medin and, with Will Hearst, first proposed cable modems to him, he said they wouldn’t work. There would be “impedance mismatches” with the hardware and software in the rest of the network.
“This kind of blew the air out of their tires,” says Medin. “But then I told them how the system could work.”
“You have to think of it as a distributed computer system. In such systems, every processor cannot access memory at once. You build caches and shared-memory protocols and you mirror and replicate a lot of the data so that it’s always available locally. That’s what you’re going to have to do on the Internet.”
In other words, the Internet is a computer on a planet. Like a computer on a chip, its raw bandwidth cannot handle the necessary throughput. Thus its communications depend on ingenious hierarchical memory management, with registers, buffers, latches, caches and direct-memory access controllers.
Studies of Internet use show that some 80% of the traffic is still local. If a particular Web page is popular in a particular locality, you have to have that page in the hard drive or even in RAM on a local server. You have to use the multicast capabilities of cable to broadcast popular information to all ad-dresses. Above all, you have to make the system scalable. You have to phase in bandwidth, moving fiber links and nodes deeper into neighborhoods as demand rises. All this is perfectly possible technically, Medin assured Doerr.
“After I was through, they decided they had to hire me.
“I still thought we would have to build all the software ourselves. I didn’t know Netscape was thinking the same way. But I went down to Netscape and got together with Marc Andreessen, who is a friend, and we had a real mind-meld on all this stuff. They were doing the software already. Netscape became our main software partner. It turned out that their browser is designed for multicast. And their proxy server is great for caching information and delivering it to users on demand.”
Medin thinks that a key to making the system work is to distribute lots of cache through all the local points of presence. With this kind of network, the teleputer might become not only far cheaper but also far superior to today’s PC. A now famous Gartner Group study shows that the average office PC costs $40,000 over five years when you factor in software and network maintenance.
Perhaps 75% of cumulative PC costs now come from staff support. @Home will supply tech support, maintenance and storage more efficiently, whether centrally by phone or at local headends.
For a glimpse of the future, visit Boston College, where cable modems supplied by Continental Cablevision are already becoming “addictive” to many students and professors. On the basis of this experience, Forrester Re-search is now predicting sales of some 7 million cable modems by the turn of the century. Medin thinks this es-timate is conservative.
With cable modems you will come to demand wireless connectivity throughout your home or small office, so that your teleputers can link to the Net wherever they are with-out plugging them in to a connector or dialing up a connection. Only cable can accomodate such demands. “Internet PCs fit with @Home like ice cream and hot fudge,” sums up Medin.
Now the big question: Is it possible to build such a machine? “Sure it is,” says Medin. “Just take a Sony PlayStation, essentially based on a one-chip ASIC, and re-place the CD-ROM connector with an Ethernet adapter. You’ll get 3D graphics, Dol-by III sound, a 30-megahertz CPU controller, a memory-access controller, and a 10-megabit-per-second 10BaseT link to your cable headend.”
C O N S T E R N A T I O N I N S I D E I N T E L
Here the new paradigm be-gins to threaten the cause and complacency not only of Bill Gates but also of the other master of Wintel, Andy Grove. When I ask him about such an ASIC solution to the problem of the $500 PC, consternation breaks briefly through the surface of his bonhomie.
He snaps: “I won’t comment on the fantasies of Brian Halla,” the former Intel manager now executive vice-president of product marketing at LSI Logic. Yet LSI Logic offers precisely Medin’s PlayStation solution to the problem of cheap teleputers, free of Wintel code.
LSI Logic is the supplier of the workhorse chip for the PlayStation. Using some 2 million transistors, this integrated chip combines a 30- megahertz Silicon Graphics MIPS processor, a 60-MIPS geometry transfer engine, a direct-memory access unit, and Sony’s proprietary MDEC device (for hybrid MPEG and JPEG decompression) for full-screen video playback.
For a Netstation, the MDEC would be replaced with the appropriate decompression engines, and added to those would be Reed-Solomon and Viterbi error correction together with a cable modem module that receives 64-QAM signals and sends QPSK. Based on its experience with the PlayStation chip, which LSI Logic will be producing in volume on the world’s first commercial 0.25-micron fabrication lines, LSI Logic estimates that it could sell a teleputer on a chip for around $50 in volume.
This machine is the consummation of a long LSI Logic strategy. In the mid-1980s, the company suffered a serious crisis as NEC, Fujitsu and Toshiba all opened fast-turnaround design centers in the U.S. to deliver high-speed, high-density gate arrays. At the same time, LSI launched a spinoff, Headland Technology, to make chipsets for PCs in competition with Chips & Technologies and VLSI Technology.
“Supporting Headland,” says Halla, “was like walking around with an open artery. Intel remorselessly sucks out all the margins in PC hardware.” Chastened by the Japanese in gate arrays, LSI learned from Grove not to take on Intel in PC markets.
To a company special-izing in gate arrays and chipsets, these lessons were not inspiring. Then Wilf Corrigan, LSI’s salty founder and CEO-a Silicon Valley legend from Liverpool who previously played key roles at Fairchild and Motorola-underwent a triple bypass.
People talked of retiring him, giving him a title with a new consortium, U.S. Memories, where he might have learned not to compete with the Koreans in DRAMs.
Two weeks after surgery, however, Corrigan returned to work at LSI and developed a new strategy that would transcend the strategies of both Intel and the Japanese. Under the new plan, LSI built state-of-the-art fabrication facilities and design tools that could enable creation of a software library of “Core-Ware.” CoreWare programs would generate a large variety of key functions, from CPU kernels to signal processors and graphics engines, that could be deployed in weeks on single special-purpose chips tailored to high-volume applications mostly outside the Wintel ambit.
By 1995, this strategy was bearing rich fruit. The company announced it had developed a fab process that could place some 49 million transistors on a single sliver of silicon some 200 millimeters square. LSI released a series of bellwether high-volume devices that moved the company beyond the path of the PC, out into the network and into the consumer appliance. The Sony PlayStation chip ran 1995’s most successful CD game machine. LSI’s MPEG-2 decoder will go in the next version of RCA-Thomson’s hugely successful direct-broadcast satellite receiver. LSI also sup-plied the first ATM segmentation and reassembly chip for several key equipment companies, and the first 100-megabit switched-Ethernet solution for a fast Ethernet pioneer.
Looking beyond Intel, this experience led Corrigan and Halla to conclude that their company commanded all the crucial CoreWare ready to deploy a teleputer on a single device manufacturable in volume for about $50. Attesting that Halla’s view is not fantasy is the success of the Sony PlayStation, now on sale for $299, leaving room to buy a monitor and still stay under $500. LSI also has several as-yet-unannounced design wins for cable modem chips aimed at the markets to be opened by @Home.
Intel’s response to such capabilities is its currently embattled program for native signal processing. NSP allows software implementation of real-time functions performed in special-purpose hardware under the “CoreWare” model. As Grove points out, DSPs and compression chips may be getting cheaper all the time, but from Indeo video to Proshare teleconferencing to Intel’s new Vertical Blanking Interval webware, NSP comes free of charge-if you are already buying a Pentium (and you prob-ably are).
But no sooner will the CPU suck in another real-time role than a new virtual temptation will glimmer on the horizon. For reaching the ever-receding real-time goals-from continuous speech recognition to 3D rendering-the Core-Ware approach will prevail, at least until the arrival of the new super broadband mediaprocessors from MicroUnity and other Silicon Valley firms.
REGARDLESS OF WHAT HAPPENS on these far frontiers of technology, John Doerr will launch his five smooth stones as the foundation for a new industry. With Intuit leading the move to Internet financial services, Netscape pioneering Internet software, Macromind supplying the authoring tools for multimedia, @Home providing the bandwidth and Sun offering Java and UltraSPARC, the entente is on its way.
But perhaps Doerr’s most important stone is Sun. Asked to name the key influences in TCI’s shift toward the Net, John Malone mentions Doerr first, hesitates and then stresses the role of Scott McNealy, chairman and CEO of Sun Microsystems. Relentlessly, year after year, McNealy would travel to Denver and give his pitch to Malone: Buy sets of Sun servers and link them to your headends in order to supply data services. At first Malone resisted. “Two-way data,” he used to say, “is not a business that I want to be in.” But as the Internet grew, McNealy’s argument gained new force. In late 1994, it triumphed.
Malone now believes that two-way broadband communications is the heart of his business. For a total investment of less than $188 million, Doerr claims @Home can launch a business yielding at least $500 million in cumulative revenues by the year 2000. Later this year, the Sunnyvale system will be up and running. Weeks later, depending on cooperation from other cable companies, the entire state of Connecticut will move onto broadband two-way cable.
At that point, all the other cable companies will accelerate their drive to upgrade their facilities to accommodate the gold rush. By the turn of the century, @Home hopes to extend service to all the major urban and suburban centers.
With Sprint, TCI leads a cable group that is paying $2 billion for wireless personal communications service (PCS) spectrum across the country. Through New York-based Teleport Communications and other bypass providers, TCI and other cable firms already command fiber-optic rings through most major metropolitan areas. With cable providing broadband backhaul for PCS, @Home’s founders think the company will emerge as the backbone for a full-service digital communications network, including high-resolution teleconferencing, on-demand films and other pay-per-view video, local news and school listings, classified advertisements, World Wide Web resources, and multimedia programming. TCI itself is furiously upgrading and streamlining all its billing systems to accommodate this rich transactional environment.
Z E R O M A R G I N A L M A G I C
Perhaps most important, as Nathan Myhrvold ex-plains, is the extension of the computer model of flat-rate pricing into the field of communications. When you buy a PC, you purchase its MIPS and bits essentially at a flat rate. The average cost per MIP or bit of memory you use is determined by how much you use the machine. The marginal cost is zero. As a result, people have a powerful incentive to use computers as intensively and creatively as possible.
This flat-rate pricing effect-where incremental costs are essentially zero-largely explains the huge success of the general-purpose PC and the companies supplying it with software and peripherals. Faced with a zero marginal cost of incremental use, PC owners channel as much of their information processing, education and entertainment as possible through the PC. Flat-rate pricing makes the PC a dire threat to all contiguous industries and related functions.
Similarly, on the Internet model, you will pay a flat rate for bandwidth. Again marginal costs will be zero. Average cost will respond to the extent of usage rather than to a Public Utilities Commission tariff or some per-minute charge. As Myhrvold points out, this approach will give you a tremendous incentive to exploit bandwidth as fully as you can, channeling as much communication as possible away from systems that charge incrementally and toward flat-rate systems. In the end, nearly all communications will gravitate toward the Internet model, and companies will prosper to the extent that they can ally themselves with this tremendous force of creativity and economy.
Myhrvold now says that bandwidth is growing at the same pace as Moore’s Law. Gates, too, though long alert to the effects of exponentials in semiconductors, is strangely blind to the faster trajectory of communications. He finds the Internet a big surprise: “Who predicted it?” he asks. “Let’s find the guy and make him king.”
But by any measure – nodes, total bandwidth, traffic – the Internet has been doubling every year since 1970, and many people have predicted that it would come to dominate communications. I prophesied in 1989 that it would usurp television.
Today I believe the band-width tidal wave will sweep away the notions of Gates and Myhrvold of a smooth middleband transition for Microsoft and its boob-tube collaborators.
Increasingly released from regulatory restrictions, band-width is now expanding far faster than MIPS and bits (see Forbes ASAP, “The Bandwidth Tidal Wave,” Dec. 5, 1994). Over the next five years, for example, @Home will increase the bandwidth to home- and small-business computers by a factor of thousands. While Moore’s Law doubles com-puter power every 18 months, the law of the telecosm, by the most conservative possible measure, doubles total band-width every 12 months. This adds up. Over the next decade, computers will improve a hundredfold while bandwidth will expand a thousandfold.
Until this year, the computer and software industries have drafted behind Moore’s Law, while hiding behind what Roger McNamee has dubbed “Moron’s Law”-the telecom regulations that stifle bandwidth expansion. The industry thus has thrived by employing MIPS and bits as a replacement for bandwidth by using compression, de-compression, switching speed, and logic circuitry to make up for the constrained bandwidth of public networks.
Grove capped off this tradition in early October, in a memorable keynote address, following South African president Nelson Mandela to the podium, at the Telecom 95 quadrennial exposition in Geneva. Grove wowed the large audience of telco potentates with an on-stage real-time demonstration of Intel’s Proshare teleconferencing technology. And yet, what chiefly struck the viewer was the mediocrity of the partial-screen facial images. They were far lower in resolution than ordinary television.
As long as the pictures are inferior to TV images, PC teleconferencing will remain chiefly a niche or a stunt. As bandwidth expands powerfully over the next decade, it will seem increasingly perverse to substitute processing for band-width, and more and more inviting to substitute band-width for processing.
The logic of MIPS and bandwidth works both ways. Not only can processing make up for bandwidth, but bandwidth, as Claude Shannon pointed out in 1948, can serve as a substitute for switching and other computer functions. With bandwidth now expanding faster than processing speeds, new architectures will prevail by substituting bandwidth for MIPS and bits. Today, the bulk of bandwidth to homes is coaxial cable laid over the last 25 years by the cable television industry. Exploiting that bandwidth for the Internet is the single greatest opportunity in the history of information technology.
HAVING ASCERTAINED that Doerr had been meeting regularly with all the pioneers of the new paradigm that Andreessen was “mind-melding” with Medin, and that Mal-one had been consulting with Doerr and McNealy, I ask McNealy whether he has talked to Medin.
“No,” he answers, “I have not talked with Medin at all … since lunch on Wednesday. But back then he and I had sore necks from basically agreeing with what the other was saying. We’re fighting like crazy to become the standard platform for Medin’s environment. Our companies are in-credibly well aligned. We have a list of about 12 engineering efforts that we are undertaking and driving at Medin’s request. The power of a network comes from the number of nodes times the bandwidth. By this measure, the @Home opportunity is as big as there is.” McNealy revealed that the day before in his office at Sun, he saw the demo of “a diskless, CD-less, floppyless, OS-less computer, and it was great. It had just about every bus, serial and parallel and S-bus, and every kind of interface you can imagine. With connectors on all four sides, it was a model for what I call a zero-administration client.
“Consider: If you give a user a disk drive, a CD, a floppy, an OS and 16 megabytes of memory, you have made him, whether he wants it or not, a system administrator. He has so many resources to manage. What they showed me yesterday was a virtual machine written in Java, and it booted up instantly off flash ROM and ran like crazy be-cause the virtual machine rides so close on the hardware.”
Echoing Medin and Corrigan, McNealy evoked the future of the teleputer: “Put a touch screen on it and make it a kiosk, put a large screen on it and make it a workstation, put in an infrared detector and make it a set-top box, put a joystick on it and make a game machine, put a cable modem or an ISDN port on it and make it a PC or a digital phone. You never run out of disk space; you never have to back it up; it’s mirrored so you never lose your files. You have an uninterruptible power supply. Your phone or cable line is much more reliable than your hard drive on your PC. You get used to the security of the system with no disks to corrupt and with Java programs that execute only in a virtual machine and cannot invade your system.”
McNealy might have added, in ecumenical concern for Larry Ellison’s mother, “Put in some Oracle code and you have a terrific, cheap database client in an emerging world of far-flung databases.”
All the participants in the new regime agree. Combined with a broadband network, the teleputer will be more flexible and powerful than existing PCs. Rolling out both the network and the teleputer will be the central activity in the industry over the next two years. Responding to it will be the principal challenge to Gates and Grove, and possibly a route of redemption for their companies. All the leading figures in this Silicon Valley renaissance have endured recent periods of trial and failure. Malone suf-fered the collapse of his Bell Atlantic merger in 1994 and the long stagnation of his stock. Corrigan suffered physical collapse and the slump of his company. McNealy endured a long tunnel of shrinking market share as his SuperSPARC processor proved too complex to keep pace with rivals. Clark of Netscape lost influence at Silicon Graphics, the company he had founded, and finally had to leave in order to retain his self-respect. Even Doerr lost his touch for several years.
For Bill Gates, however, business life has been an al-most unrelenting ascent toward riches untold for one so young. In the end, his success has made him seem a bionic business leader leached of his humanity.
His company has appeared to government and to competitors alike as a monopoly threat, targeting existing rivals and systematically suppressing them, rather than creating new products and industries. Much the same is said of Intel.
Perhaps this is the time for the Wintel team to face a domestic challenge. From it they may well emerge stronger. Without them, it is clear, the Internet will be weaker. Only from the crucible of competition between paradigms can emerge a robust and redemptive new economy of information.