This is an ongoing set of notes, updating my original essay “Lawrence’s Laws—A Personal View of the Computer Industry”, to address developments that have taken place in the industry since then.
ORSINO How dost thou, my good fellow?
FESTE Truly, sir, the better for my foes and the worse for my friends.
ORSINO Just the contrary—the better for thy friends.
FESTE No sir, the worse.
ORSINO How can that be?
FESTE Marry, sir, they praise me, and make an ass of me. Now my foes tell me plainly that I am an ass, so that by my foes, sir, I profit in the knowledge of myself, and by my friends I am abused; so that, conclusions to be as kisses, if your four negatives make your two affirmatives, why then the worse for my friends and the better for my foes.—William Shakespeare, Twelfth Night
Brand loyalty is usually considered an asset for a company. The fact that there are a large number of current customers who are predisposed to consider the company’s future products in a positive light can only be a good thing, right? This may be true in many industries, but perhaps not in computing.
The computing platform with perhaps the most loyal, even fanatical, user base had to be the Commodore Amiga. And yet the loyalty of the customers was never enough to prevent Commodore from lurching from one financial crisis to another, until it went out of business altogether in 1994. At the same time, the loyal users refused to allow the platform to just die; people kept trying to resurrect it, even as effort after effort continued to founder on the sheer lack of a way to make money from the technology.
The computing platform with the next most loyal user base must be the Apple Macintosh. This is a product which is still being made, and is still earning money, though one wonders for how much longer as the company moves into new product areas.
On the other hand, consider Microsoft. This company dominates the PC business like no other. Yet you would be hard pressed to find a customer who actually confesses to a liking for the company. Over the last 10 years, Microsoft has become notorious for heavy-handed marketing tactics, and products that are full of security holes and prone to virus infections. You can hardly pick up a current issue of any computer magazine without seeing more complaints about these problems. Yet when you look at the sales figures, the dominance of Microsoft in the marketplace seems just as strong as ever. People talk about moving away from Microsoft products, yet when it comes to actual buying decisions, they are yet to put their money where their mouth is.
Clearly, Microsoft is not exactly wallowing in an abundance of customer loyalty: if its users could find products that were at least as good from someone else, many of them would switch like a shot. The fact that they aren’t switching isn’t because of any brand loyalty, or because of any “backward-compatibility” excuse (consider: users didn’t initially switch to Microsoft products because it was “backward-compatible” with anyone): it must simply be because its products, with all their abundant faults, remain better than anything else out there.
The conclusion I draw from all this, that I hope is illustrated in the Shakespeare quote, is that, in the computer business, the loyalty of friends could actually be bad for a company. It lulls the company into a sense of complacency: because there is an existing user base that will continue to buy its products even if they aren’t exactly Earth-shattering, there is no need to strive harder, to make products that will appeal even to those who weren’t attracted by the company’s brand to begin with. When you don’t have this brand loyalty, when in fact there is an undercurrent of hostility towards your existing success, then you stand or fall on the quality of the products themselves.
Earlier I made some comments about how Moore’s Law will reach its limit as the slowing growth in various markets for semiconductor products comes up against the increasing capital costs of fabrication plants for making chips with higher circuit integration densities.
I’ve just had independent confirmation of this viewpoint: Morris Chang, CEO of Taiwan Semiconductor says:
Moore’s Law will have to slow down. Economically, we’re running into a wall here. We can’t afford to do these expensive designs anymore. We can’t afford a big fab anymore.
According to this item, the price of a chip fab can now exceed 3 billion US dollars.
Following up on my earlier comments about the inevitability of Wi-Fi taking over the Bluetooth market, this item reports on some devices which are starting to do just that. There seem to be some limitations with the products in the report, but no doubt competition will remedy that in future releases.
When Sun introduced the Java programming language in the early 1990s, it was touted as the answer to all the woes that programmers had with C++. Java was supposed to be a simpler, cleaner language, more truly object-oriented for greater programmer productivity and code quality, and with integrated garbage collection to make it much easier to manage memory (always a big headache for programmers writing modern applications). It also claimed to offer true “write once, run everywhere” portability whereby you didn’t have to change a single line of code to get the same program to run on multiple platforms.
Trouble is, it never quite lived up to any of its claims. Its efficiency never came anywhere close to challenging that of C++ code. The portability never lived up to the promises, either: in particular, there were just too many platform-specific quirks in the implementations of the Java user-interface libraries.
These days, Java programming is confined to a few minor niches, while the position of C++ is stronger than ever. Despite Sun’s best efforts, I think Java has reached its peak, and the only way that it can go from here is down. In this business, economics is important, technology is not. And of all the unimportant kinds of technology, one of the least important is the language that software is written in. Despite Sun’s branding campaign, users just aren’t going to care whether the software they’re using is written in Java or not; there simply is no compelling advantage that Java gives programmers, in terms of the capabilities or quality of the software it produces.
Moore’s Law, in its currently-accepted form, states the that number of transistors that can be crammed into a single chip doubles about every 18-24 months (though the precise interval has varied over the years). Note that it does not say anything about the performance of the chip, which will not scale linearly with the number of transistors.
Over the decades since the birth of the microprocessor, there have been many successive predictions of the demise of Moore’s Law, as various looming technological limitations threaten, one after the other, to put an end to the exponential growth in chip complexity that it predicts. Yet, with remarkable regularity, solutions to these limitations seem to appear just in time, and Moore’s Law just keeps marching on.
If it isn’t clear to my readers already, let me spell it out: I predict that the ultimate limitations on Moore’s Law will be, not technological, but economic. Chips will stop growing exponentially in complexity when it is no longer cost-effective for them to do so. I believe the exponential growth in chip complexity has been a direct consequence of the (heretofore) exponential growth in the semiconductor industry.
Of course, this exponential industry growth now seems to be coming to an end. The PC market has reached saturation, and most new PCs being sold are replacements for old PCs. Newer markets like game consoles, PDAs and cellphones are also no longer growing quite as wildly as they used to. At the same time, the overwhelming majority of the cost involved in manufacturing semiconductor chips is the capital cost of setting up the chip fabrication plants (“fabs”): once the fab is built (at a cost which can nowadays be billions of dollars), the unit cost of manufacturing each chip can be measured in cents. And as the chips become more complex, the cost of the new fabs correspondingly increases, which means that more chips must be sold to recoup the capital costs. Which comes up against the limits to growth of the market. And this, ultimately, will mean a stop to Moore’s Law.
So when is this barrier likely to be hit? I will be conservative and say it’s likely sometime in the next 5-10 years, allowing for a bit more life in the cellphone market in particular. That is, unless something like one of the following occurs (in order of decreasing likelihood):
Even as the PC market has resumed modest growth, Apple’s sales continue their precipitous drop. Just to rub it in, this report confirms that Apple’s market share is now down to a record low of 2%.
I think it’s time to admit that the introduction of OS X has been a failure. More than two years after its introduction, it shows no sign of winning Apple any new markets, even as existing Mac users continue to desert the platform. In short, it has done absolutely nothing to stem Apple’s ongoing decline in market share.
Nearly three years after I wrote my original essay, Linux still remains of minority interest as a desktop operating system. However, it does seem to be making gradual, slow inroads into this market. According to one report, it has already surpassed the Apple Mac in desktop market share.
On the downside, Linux companies are still having trouble coming up with a workable revenue model. Mandrakesoft has filed for bankruptcy, and Turbolinux has just given up, while others continue to struggle. Even Red Hat, the most successful of them all, is still only teetering on the edge of profitability.
On the other hand, even though Linux may not (yet) be successful in a free desktop markektplace, that hasn’t stopped governments like those of Perú, Germany and others from pushing for its adoption within their jurisdictions. It was particularly interesting to hear the German minister use the term “monoculture” in his rationale for encouraging diversity of computer platforms: most arguments that try to appeal to Darwinian evolutionary theory to justify the form of some human endeavour don’t go beyond the meaningless cliche “survival of the fittest”.
I have to admit it, Microsoft has yet again provided a counterexample to my Laws. Contrary to my original predictions, Windows XP has finally succeeded in pushing aside the old Windows 9x/ME line.
Conversely, my idea about a future, low-cost server system cobbled together from a Windows 9x base looks pretty likely never to occur.
Two years after its first public release (not counting the public beta), Apple’s next-generation operating system has, apparently, finally reached a user base of 5 million, or 20% of the total Mac user population. The company has, for the most part, discontinued Mac models that can boot up in older versions of MacOS, and committed its own development resources exclusively to supporting the new OS.
On the other hand, the company has started losing money again. Lots of Mac users continue to be unhappy over Apple’s intention to force them to switch to the new OS whether they like it or not.
I think the verdict on OS X is still open. It has not (yet) succeeded in winning significant numbers of new users to the Mac platform. Or if it has, then it is driving away equal numbers of existing Mac users, leaving its overall market share essentially unchanged.
My prediction that Apple’s introduction in 2000 of dual-processor machines at the same price as older single-processor models would usher in a new era of multiprocessing for the masses failed to materialize. Multiprocessor machines still remain a relatively niche-market product, though Intel’s recent introduction of “hyperthreading”—a lower-cost, correspondingly lower-performance version of multiprocessing—may yet widen this market. Whether that will happen will depend on the same old principle: offer X% of the benefits for Y% of the price, where X > Y. Current indications are that this is not (yet) the case.
This is a new example I would like to offer of my Laws in action. USB 2.0 is Intel’s enhancement of its original low-speed USB interface, which is intended to compete directly with FireWire. Because USB 2.0 has the marketing clout of Intel behind it, some analysts have automatically predicted that it will eventually push aside FireWire, even in those markets where the latter already has a foothold. However, I don’t think this is going to happen, for several reasons:
The only way USB 2.0 could have gained a foothold was if it offered a substantial fraction of the capabilities of FireWire, but at a smaller fraction of the price (see the section“The Inferior Product Wins” in my original essay). Since it has failed to offer this, no amount of marketing clout will be able to make it successful. There are, of course, many technical reasons why FireWire might be considered superior to USB 2.0, but as I made clear in my original essay, technical considerations are largely irrelevant to the success or failure of a product.
This issue has been the subject of controversy for some time. Bluetooth aficionados insist that the two wireless technologies do not compete, that they are targeted at entirely different markets—Bluetooth at connecting small numbers of (relatively) unintelligent devices together over distances of a few metres, Wi-Fi at full-fledged peer-to-peer networking over longer ranges. Because Bluetooth operates over shorter ranges and can be implemented with lower power consumption, that should give it a niche where Wi-Fi cannot compete, right?
However, remember my law about the mass market taking over the niche market? Wi-Fi is already firmly established in the wider marketplace, and continues to go from strength to strength, as people find new ways to use it, for example in setting up informal neighbourhood networks, or even in long-range wireless links—an application for which it certainly was never designed. This adaptation to new, completely unexpected application areas is a sure sign of a technology that has achieved “critical mass” as far as the marketplace is concerned. Bluetooth, by contrast, is just getting started.
Added to which, consider this interview with Stuart Cheshire, mastermind of the Zeroconf effort. In particular, he says:
I would much rather have a computer with a row of identical universal [TCP/IP] communications ports, where I can connect anything I want to any port, instead of today’s situation where the computer has a row of different sockets, each dedicated to its own specialized function.
My hope is that in the future — distant future perhaps — your computer will only need one wired communication technology.
The fundamental idea he is expressing here is economies of scale: once TCP/IP protocol stacks become cheap enough and common enough (the two factors driving each other, of course), it no longer sounds ridiculous to talk of running a full TCP/IP connection between, say, your mouse, your keyboard and your PC. In fact, it may be cheaper to connect them that way than to persist with existing, less complex communications protocols for this purpose—just as it is already cheaper to connect your keyboard, mouse and printer all via USB, than to continue adding parallel or keyboard/mouse ports specifically for these purposes, even though USB is a much more complex protocol than anything that worked through those older ports.
So what is the relevance of this to the Wi-Fi-versus-Bluetooth debate? It is that Wi-Fi already implements a full TCP/IP protocol stack, whereas Bluetooth does not. It is interesting that Stuart qualifies his above remarks by saying that a future computer will only need one wired communication technology; his comments apply equally well to the competition between different wireless communication technologies.
It may be true that Bluetooth can (currently) be implemented at much lower levels of power consumption than Wi-Fi. However, remember the remarks in my original essay about economic arguments winning over technological ones: since the Wi-Fi vendors are already making a profit, while the sellers of Bluetooth products are not, there is a strong economic incentive on the former to come up with solutions to problems like this. An obvious ad-hoc solution may be simply to bring out a lower-power implementation of Wi-Fi. There may be some valid technological objections why this won’t work as well as one might like, but I don’t think that will be enough to stop people from trying. Where there is a will, and the chance of a profit, a way will be found.