The Economist

SURVEY: THE IT INDUSTRY May 8th 2003
From The Economist print edition
**NO GRAPHICS**

Paradise lost

So far, information technology has thrived on exponentials. Now it has to get back to earth, says Ludwig Siegele

CLOSE your eyes and think of information technology. You might picture your PC crashing yet again, and recall that your teenager was supposed to fix it. That leads you to the 12-year-old hacker who broke into a bank's computer system the other day, which brings to mind the whizz-kids in a garage inventing the next big thing that will turn them into the youngest billionaires ever.

In IT, youth seems to spring eternal. But think again: the real star of the high-tech industry is in fact a grey-haired septuagenarian. Back in 1965, Gordon Moore, co-founder of Intel, the world's biggest chipmaker, came up with probably the most famous prediction in IT: that the number of transistors which could be put on a single computer chip would double every 18 months. (What Mr Moore actually predicted was that the figure would double every year, later correcting his forecast to every two years, the average of which has come to be stated as his “law”.)

This forecast, which implies a similar increase in processing power and reduction in price, has proved broadly accurate: between 1971 and 2001, transistor density has doubled every 1.96 years (see chart 1). Yet this pace of development is not dictated by any law of physics. Instead, it has turned out to be the industry's natural rhythm, and has become a self-fulfilling prophecy of sorts. IT firms and their customers wanted the prediction to come true and were willing to put up the money to make it happen.

Even more importantly, Moore's law provided the IT industry with a solid foundation for its optimism. In high-tech, the mantra goes, everything grows exponentially. This sort of thinking reached its peak during the internet boom of the late 1990s. Suddenly, everything seemed to be doubling in ever-shorter time periods: eyeballs, share prices, venture capital, bandwidth, network connections. The internet mania began to look like a global religious movement. Ubiquitous cyber-gurus, framed by colourful PowerPoint presentations reminiscent of stained glass, prophesied a digital land in which growth would be limitless, commerce frictionless and democracy direct. Sceptics were derided as bozos “who just don't get it”.

Today, everybody is older and wiser. Given the current recession in IT, the idea of a parallel digital universe where the laws of economic gravity do not apply has been quietly abandoned. What has yet to sink in is that the current downturn is something more than the bottom of another cycle in the technology industry. Rather, as this survey will argue, the sector is going through deep structural changes which suggest that it is growing up or even, horrors, maturing. Silicon Valley, in particular, has not yet come to grips with the realities, argues Larry Ellison, the chief executive of Oracle, a database giant (who at 58 still sports a youthful hairdo). “There's a bizarre belief that we'll be young forever,” he says.

It is not that Moore's law has suddenly ceased to apply. In fact, Mr Moore makes a good case that Intel can continue to double transistor density every 18 months for another decade. The real issue is whether this still matters. “The industry has entered its post-technological period, in which it is no longer technology itself that is central, but the value it provides to business and consumers,” says Irving Wladawsky-Berger, a senior manager at IBM and another grey-haired industry elder.

Scholars of economic history are not surprised. Whether steam or railways, electricity or steel, mass production or cars—all technological revolutions have gone through similar long-term cycles and have eventually come of age, argues Carlota Perez, a researcher at Britain's University of Sussex, in her book “Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages” (Edward Elgar, 2002).

In her model (see chart 2), technological revolutions have two consecutive lives. The first, which she calls the “installation period”, is one of exploration and exuberance. Engineers, entrepreneurs and investors all try to find the best opportunities created by a technological big bang, such as Ford's Model T in 1908 and Intel's first microprocessor in 1971. Spectacular financial successes attract more and more capital, which leads to a bubble. This is the “gilded age” of any given technology, “a great surge of development”, as Ms Perez calls technological revolutions.

The second, or “deployment”, period is a much more boring affair. All the quick bucks have been made, so investors prefer to put their money into the real economy. The leading firms of the new economy become bigger and slower. The emphasis is no longer on raw technology, but on how to make it easy to use, reliable and secure. Yet this period is also the “golden age” of a technology, which now penetrates all parts of society.

These two periods of a technological revolution are separated by what Ms Perez calls a “turning point”—a crucial time for making the choices that determine whether a technological revolution will deliver on its promises. In her book, she concentrates mainly on the social and regulatory decisions needed to allow widespread deployment of new technology. But the same argument applies to technology vendors and customers. To enter their “golden age”, they have to leave their youthful excesses behind and grow up.

A duller shade of gold

This survey will examine how much grey the IT industry (and their leaders' hair) has already acquired. The first three chapters are about technological shifts, and how value is moving from the technology itself to how it is applied. Many of the wares that made the IT industry's fortunes in the installation period are becoming a commodity. To overcome this problem, hardware vendors are developing new software that allows networks of machines to act as one, in effect turning computing into a utility. But the IT industry's most profitable layer will be services of all kinds, such as software delivered as an online service, or even business consulting.

The second half of this survey looks at institutional learning, which has caused the value created by the IT industry to be increasingly captured by its customers. For the first time in its history, the IT industry is widely adopting open standards. Equally important, buyers are starting to spend their IT budgets more wisely. Meanwhile, the industry's relationship with government is becoming closer.

All this suggests that the technology industry has already gone greyish at the temples since the bubble popped, and is likely to turn greyer still. Sooner or later the sector will enter its “golden age”, just as the railways did. When Britain's railway mania collapsed in 1847, railroad shares plunged by 85%, and hundreds of businesses went belly-up. But train traffic in Britain levelled off only briefly, and in the following two decades grew by 400%.

So are the IT industry's best days yet to come? There are still plenty of opportunities, but if the example of the railways is anything to go by, most IT firms will have to make do with a smaller piece of the pie. As this newspaper (then called The Economist, Weekly Commercial Times, Bankers' Gazette, and Railway Monitor) observed in 1857: “It is a very sad thing unquestionably that railways, which mechanically have succeeded beyond anticipation and are quite wonderful for their general utility and convenience, should have failed commercially.”

Brad DeLong, an economics professor at the University of California at Berkeley, puts it somewhat more succinctly: “I am optimistic about technology, but not about profits.”

Modifying Moore's law

Many of the innovations that made the IT industry's fortunes are rapidly becoming commodities—including the mighty transistor

IF GOOGLE were to close down its popular web-search service tomorrow, it would be much missed. Chinese citizens would have a harder time getting around the Great Firewall. Potential lovers could no longer do a quick background check on their next date. And college professors would need a new tool to find out whether a student had quietly lifted a paper from the internet.

Yet many IT firms would not be too unhappy if Google were to disappear. They certainly dislike the company's message to the world: you do not need the latest and greatest in technology to offer outstanding services. In the words of Marc Andreessen of Netscape fame, now chief executive of Opsware, a software start-up: “Except applications and services, everything and anything in computing will soon become a commodity.”

Exactly what is meant by “commoditisation”, though, depends on whom you talk to. It is most commonly applied to the PC industry. Although desktops and laptops are not a truly interchangeable commodity such as crude oil, the logo on a machine has not really mattered for years now. The sector's most successful company, Dell, is not known for its technological innovations, but for the efficiency of its supply chain.

As the term implies, “commoditisation” is not a state, but a dynamic. New hardware or software usually begins life at the top of the IT heap, or “stack” in geek speak, where it can generate good profits. As the technology becomes more widespread, better understood and standardised, its value falls. Eventually it joins the sector's “sediment”, the realm of bottom feeders with hyper-efficient metabolisms that compete mainly on cost.

Built-in obsolescence

Such sedimentation is not unique to information technology. Air conditioning and automatic transmission, once selling points for a luxury car, are now commodity features. But in IT the downward movement is much faster than elsewhere, and is accelerating—mainly thanks to Moore's law and currently to the lack of a new killer application. “The industry is simply too efficient,” says Eric Schmidt, Google's chief executive (who seems to have gone quite grey during his mixed performance at his previous job as boss of Novell, a software firm).

The IT industry also differs from other technology sectors in that its wares become less valuable as they get better, and go from “undershoot” to “overshoot,” to use the terms coined by Clayton Christensen, a professor at Harvard Business School. A technology is in “undershoot” when it is not good enough for most customers, so they are willing to pay a lot for something that is a bit better although not perfect. Conversely, “overshoot” means that a technology is more than sufficient for most uses, and margins sink lower.

PCs quickly became a commodity, mainly because IBM outsourced the components for its first venture into this market in the early 1980s, allowing others to clone the machines. Servers have proved more resistant, partly because these powerful data-serving computers are complicated beasts, partly because the internet boom created additional demand for high-end computers running the Unix operating system.

But although expensive Unix systems, the strength of Sun Microsystems, are—and will probably remain for some time—a must for “mission-critical” applications, servers are quickly commoditising. With IT budgets now tight, firms are increasingly buying computers based on PC technology. “Why pay $300,000 for a Unix server,” asks Mr Andreessen, “if you can get ten Dell machines for $3,000 each—and better performance?”

Google goes even further. A visit to one of the company's data centres in Silicon Valley is a trip back to the future. In the same way that members of the Valley's legendary Homebrew Computer Club put together the first PCs using off-the-shelf parts in the early 1970s, Google has built a huge computer system out of electronic commodity parts.

Modern Heath Robinsons

When the two Stanford drop-outs who founded Google, Sergey Brin and Larry Page, launched the company in 1998, they went to Fry's, an electronics outlet where the Valley's hardcore computer hobbyists have always bought their gear. Even today, some of the data centres' servers appear to be the work of tinkerers: circuit boards are sagging under the weight of processors and hard disks, and components are attached by Velcro straps. One reason for the unusual design is that parts can be easily swapped when they break. But it also allows Google's servers to be made more powerful without having to be replaced completely.

What makes it easier for Google to swap off-the-shelf components is that much of its software is also a commodity of sorts. Its servers run Linux, the increasingly popular open-source operating system developed by a global community of volunteer programmers, and Apache, another open-source program, which dishes up web pages.

Because Google has always used commodity hardware and software, it is not easy to calculate how much money it has saved. But other firms that have recently switched from proprietary gear say they have significantly reduced their IT bill. Amazon.com, the leading online shopping mall, for instance, managed to cut its quarterly technology spending by almost $20m (see chart 3).

The most interesting feature of Google's data centre, however, is that its servers are not powered by high-end chips, and probably will not have Itanium, Intel's most powerful processor, inside for some time yet. This sets Google apart among hot Silicon Valley start-ups, whose business plans are mostly based on taking full advantage of the exponential increase in computing power and similar growth in demand for technology.

“Forget Moore's law,” blared the headline of a recent article about Google in Red Herring, a now-defunct technology magazine. That is surely overblown, but Google's decision to give Itanium a miss for now suggests that microprocessors themselves are increasingly in “overshoot”, even for servers—and that the industry's 30-year race for ever more powerful chips with smaller and smaller transistors is coming to an end.

Instead, other “laws” of the semiconductor sector are becoming more important, and likely to change its underlying economics. One is the fact that the cost of shrinking transistors also follows an exponential upward curve. This was no problem as long as the IT industry gobbled up new chips, thus helping to spread the cost, says Nick Tredennick, editor of the Gilder Technology Report, a newsletter. But now, argues Mr Tredennick, much of the demand can be satisfied with “value transistors” that offer adequate performance for an application at the lowest possible cost, in the same way as Google's. “The industry has been focused on Moore's law because the transistor wasn't good enough,” he says. “In the future, what engineers do with transistors will be more important than how small they are.”

This is nothing new, counters Paul Otellini, Intel's president. As chips become good enough for certain applications, new applications pop up that demand more and more computing power, he says: once Google starts offering video searches, for instance, it will have to go for bigger machines. But in recent years, Intel itself has shifted its emphasis somewhat from making ever more powerful chips to adding new features, in effect turning its processors into platforms.

It recently launched Centrino, a group of chips that includes wireless technology. The Centrino chips are also trying to deal with another, lesser-known, limiting factor in chipmaking: the smaller the processors become, the more power-hungry and the hotter they get (see chart 4). This is because of a phenomenon called leakage, in which current escapes from the circuitry. The resulting heat may be a mere inconvenience for users of high-end laptops, who risk burning their hands or thighs, but it is a serious drawback for untethered devices, where it shortens battery life—and increasingly for data centres as well, as Google again shows.

Cool chips

The firm's servers are densely packed to save space and to allow them to communicate rapidly. The latest design is an eight-foot rack stuffed with 80 machines, four on each level. To keep this computing powerhouse from overheating, it is topped by a ventilation unit which sucks air through a shaft in its centre. In a way, Google is doing to servers what Intel has done to transistors: packing them ever more densely. It is not the machines' innards that count, but how they are put together.

Google has thus created a new computing platform, a feat that others are now replicating in a more generalised form. Geoffrey Moore (no relation), chairman of the Chasm Group, a consultancy, and a partner at Mohr, Davidow Ventures, a Silicon Valley venture-capital firm, explains it this way: computing is like a game of Tetris, the computer-game classic; once all the pieces have fallen into place and all the hard problems are solved, a new playing field emerges for others to build on.

Moving up the stack

The network is becoming the computer—and the IT industry's dominant platform

COMPUTING is supposed to be the ultimate form of automation, but today's data centres can be surprisingly busy with people. When an application has to be updated or a website gets more visitors than expected, system administrators often have to install new programs or set up new servers by hand. This can take weeks and often turns out to be more complicated than expected.

Google's data centres, however, look deserted most of the time, with only about 30 employees to look after a total of 54,000 servers, according to some estimates. This is in part because machines doing searches need less care than those running complex corporate applications; but more importantly, the firm's programmers have written code that automates much of what system administrators do. It can quickly change a computer that sifts through web pages into a server that dishes up search results. Without the program, Google would have to hire many more people.

It all goes to show that another law in computing, proclaimed by Gordon Bell, another greying industry legend, still holds true: in IT, the dominant platform shifts every ten years or so. Mainframes, minicomputers, PCs and servers are now likely to be followed by a grid of computers, either within a data centre or as a disparate collection of connected machines. The network will at last be the computer, to paraphrase a slogan coined by Sun Microsystems. Machines will no longer simply be attached to a network: instead, the network will allow them to act as one.

Yet this new platform, which computer scientists like to call “grid computing”, is less about replacing old technology and more about managing the existing gear—another sign that IT is maturing. Merrill Lynch's Steve Milunovich, one of the leading hardware analysts on Wall Street, says that IT has entered the era of “managed computing”. Forrester Research, a high-tech consultancy, has coined the term “organic IT”—a computing infrastructure that is not only built on cheap parts, but is also as adaptive as a living organism. Whatever label the industry settles for, the race to lead in the next round of computing is already on. The new platform gives those threatened by commoditisation a chance to differentiate themselves by moving up the technology stack to a potentially more lucrative layer.

There is every incentive for HP, IBM, Microsoft and Sun, as well as a raft of start-ups, to encourage this shift, but there is also a real need for a new platform. Computing has certainly got faster, smarter and cheaper, but it has also become much more complex. Ever since the orderly days of the mainframe, which allowed tight control of IT, computer systems have become ever more distributed, more heterogeneous and harder to manage.

Managing complexity

In the late 1980s, PCs and other new technologies such as local area networks (LANs) allowed business units to build their own systems, so centralised IT departments lost control. In the late 1990s, the internet and the emergence of e-commerce “broke IT's back”, according to Forrester. Integrating incompatible systems, in particular, has become a big headache.

A measure of this increasing complexity is the rapid growth in the IT services industry. According to some estimates, within a decade 200m IT workers will be needed to support a billion people and businesses connected via the internet. Managing a storage system already costs five times as much as buying the system itself, whereas less than 20 years ago the cost of managing the system amounted to only a third of the total (see chart 5).

What is more, many of today's IT systems are a patchwork that is inherently inefficient, so firms spend 70-90% of their IT budgets simply on keeping their systems running. And because those systems cannot adapt quickly to changes in demand, companies overprovide. They now spend almost $50 billion a year on servers, but the utilisation rate for these computers is often below 30%.

Besides, complexity is bound to increase, predicts Greg Papadopoulos, Sun's chief technology officer. Today, the electronics to hook up any device to the network cost about $1. In ten years' time, the price will be down to one cent. As a result, he says, the number of connected things will explode, and so will the possible applications. For example, it will become practical to track items such as razor blades (10% of which apparently disappear on their way from the factory to the shop).

When things get too complicated, engineers usually add a layer of code to conceal the chaos. In some ways, the current shift in computing is the equivalent of what happened when cars became easy to use and drivers only had to turn the key instead of having to hand-crank the engines. In geek speak, adding such a new layer is called “raising the level of abstraction”. This happened when PC operating systems first hid the nuts and bolts of these computers and gave them a simple user interface, and it is happening now with the new platform, which is already being compared to an operating system for data centres or computing grids.

Just like Google's management program, this grid computing software (only half-jokingly called “griddleware” by some) automates much of the work of system administrators. But it is also supposed to serve a higher purpose: “virtualisation”. Put simply, this means creating pools of processing power, storage capacity and network bandwidth. A data centre, or a collection of machines on a network, thus becomes a virtual computer whose resources can be allocated as needed. The ultimate goal is that managed computing will become rather like flying a modern jet plane: IT workers will tell the system what kind of applications it should run, and then deal only with exceptions.

Although the rivals in this new field are pretty much on the same technological track, their strategies are different. Some of the numerous start-ups already have working products—and no hidden agenda, says Mr Andreessen, of Opsware, the leading newcomer: “We don't need to push our customers also to buy other stuff from us.” The incumbents, on the other hand, want the new software layer to protect their old business models as well. HP's Utility Data Centre (UDC) initiative and Sun's N1 plan are supposed to help these firms sell their profitable hardware. IBM's “autonomic computing” effort goes hand-in-hand with Big Blue's IT services business. And Microsoft's Dynamic Services Initiative (DSI) is tightly linked with its Windows operating system.

Yet despite such arm-twisting, customers are unlikely to bet solely on newcomers. Only the biggest vendors will really be able to deliver managed computing, argues Shane Robinson, the chief technology officer of HP, which has much riding on the new platform. According to the Gartner Group, a consultancy, HP is leading in virtualisation, and views management software as its big opportunity.

Once thing is clear: once all the technical challenges of grid computing have been overcome, hardware will have become a true commodity. Machines, storage devices and networks will lose their identity and feed into pools of resources that can be tapped as needed. This liquefaction of hardware, in turn, will allow computing to become a utility, and software a service delivered online.

Techniques, not technology

IT firms hope to turn the dismal science into a profitable business
TO MOST people the world is round, but geeks often see it as a stack of layers. In corporate computing, it starts with the hardware, on top of which sits the operating system, then the database, the applications and finally IT services. When their layer is getting commoditised, technology companies tend to move up the stack, where more money can be made.

In their quest for greener pastures, IT firms have reached new heights by moving into cutting-edge economics. Both HP and IBM have opened labs to conduct research on the subject, in the hope that this will help them to offer their customers more sophisticated services.

To be sure, economics has had its place in the IT industry for some years now. HP, for instance, already uses software that simulates markets to optimise the air-conditioning systems in its utility data centres. And IBM's Institute for Advanced Commerce has studied the behaviour of bidding agents, in the hope of designing them in such a way that they do not engage in endless price wars. Now HP is reaching even higher, with experimental economics. As the name implies, researchers in this field set up controlled experiments with real people and real money to see whether economic theories actually work. Perhaps surprisingly, it seems that they do, as demonstrated by the work of Vernon Smith of George Mason University in Virginia. (Mr Smith is considered the founding father of this field and won last year's Nobel prize in economics.)

Secret agent

HP goes further. The firm's team of five researchers does not test economic theories, but tries to create “novel mechanisms to improve the fluidity of interactions in the information economy”, says Bernardo Huberman, head of the group. In everyday language, the researchers are working on clever tools that make it easier to negotiate online, establish reputations and make forecasts.

Mr Huberman's group already has something to show for its efforts. It has developed a methodology for predicting uncertain events using a small group of individuals. First, they find out about their subjects' attitudes towards risk and their ability to forecast a given outcome. They then use this information to weight and aggregate their predictions of events, resulting in fairly accurate forecasts. These tools will first be used inside the company. The top management of one of HP's divisions is already testing the forecasting methodology to predict its revenue. But ultimately the firm wants to find outside customers for its research findings. American intelligence agencies, such as the CIA, have already shown interest. They need better tools to weigh the opinions of those who analyse incoming information.

So at what point will firms such as HP and IBM have moved far enough up the stack to cease to be traditional IT vendors and become service providers or consultancies? Most analysts agree that this metamorphosis is still some way off. But it already seems certain that in future IT firms will increasingly be in the business of techniques rather than technology.

Top of page.

Home About Resources Investors Businesses Members Admin