Moore’s Law and Communications

Martin E. Hellman
Professor Emeritus of Electrical Engineering
Stanford University
Last Revised: 11 June 2003

In 1965, Gordon Moore of Intel postulated what is now known as Moore’s Law, that the computing power of IC’s doubles approximately every year and a half, at no increase in cost per IC. Moore was extremely prescient as this law has continued on course for decades, up to the present time. [Note 1]

I prefer two modifications to Moore’s Law. First, the reductions in the cost of computation predicted by Moore’s Law predate the IC. A related decrease in computational costs had been going on for the preceding several decades, starting with relay computers in the 1930’s, moving to vacuum tube computers in the 1940’s, and transistorized computers in the 1950’s.

Second, since humans think more readily in base 10, I prefer an equivalent version of Moore’s law that says computing costs decrease by a factor of ten every five years. (Five years is three periods of a year and a half plus half a year. Three reductions by a factor of 2 is a factor of 8, and the extra half year takes us to almost exactly a factor of 10.)
Applying this law to the last fifty years – ten periods of five years – we find that a computation that costs $1 today would have cost approximately $10 billion fifty years ago. From a computational point of view, we are all billionaires!

This rapid and steady decrease in the cost of computation has been like a mountain spring that feeds a river. Those of us in the developed world live on the banks of this river and have derived great wealth and other benefits (e.g., non-invasive medical diagnosis through CAT scans and MRI).

To put Moore’s Law in perspective, imagine it applied to automobiles. The typical car that costs approximately $20,000 today would have cost $200,000 five years ago and been limited almost like corporate jets to high level management and very wealthy individuals. Twenty years ago, cars would have cost $200 million and been as rare as rocket launches into outer space. Conversely, looking just ten years into the future, cars would cost $200. Body shops and auto repair facilities would go out of business, with great economic dislocations, both positive and negative.

We have seen such changes in the computing landscape. IBM cleverly played the early stages of the computer revolution to become the main frame power house. Replacing manual systems for accounting and other tasks realized immense savings for its customers, allowing IBM to maintain high profit margins and make a fortune for its shareholders. But IBM faced real danger when minicomputers and then PC’s and servers replaced main frames at cost savings comparable to what IBM had wrought over manual systems. Only by reinventing itself as a services company, did Big Blue manage to stay alive and even prosper.

An analog of Moore’s Law applies to communications. Fifty years ago, the cost of a transcontinental phone call was on the order of $1 per minute, with approximately half the cost due to switching and half the cost due to long haul communications.
Since switching can be accomplished by computational means (e.g. packet switching), the $0.50 of switching cost can be accomplished today at an infinitesimal cost, approximately a billionth of a cent per minute, if it is done in the most cost-effective manner. The cost of the long haul communications has also fallen by at least a factor of a thousand and probably a million or more. [Note 2]

The net result is that all phone calls today should be virtually free. Instead of being pleased with the seeming bargain rate of $0.05 per minute I now pay, it should be seen as highway robbery compared to pricing in a truly competitive environment. Even using the conservative factor of a thousand for the reduction in communications cost, the free market pricing should be on the order of 0.05 cents (not dollars!) per minute. That’s twenty minutes for a penny. My five cent a minute plan is one hundred times as expensive.

We can think of the communications analog of Moore’s Law as another mountain spring, feeding another river emanating from the mountains on the other side of our valley. But there have been numerous dams that have stunted what should have been a strong, steady flow into an unsteady trickle. For a long time, the dams were the FCC and its PTT foreign counterparts which have kept communications costs artificially high out of a concern for subsidizing consumers at the expense of business.

Starting about ten years ago, the Internet poked a hole in those dams and allowed a larger flow in the “communications river.” With the advent of the Internet, I can send large amounts of data anywhere in the world, at no additional cost over my monthly ISP charges. While nuclear power once falsely promised electric power too cheap to meter, the Internet has provided just such a benefit to at least some communications.
The sudden breaching of the dams holding back the communications river had the potential to produce wealth and other benefits even more rapidly than did the computational revolution because there was a store of water (potential reduction in costs) behind those dams that had accumulated over fifty years. I believe that some of the “irrational exuberance” of the Internet bubble was due to such hopes.

In spite of the great advances of the Internet Revolution, the Communications Revolution has been largely unrealized. My cable modem service is no faster and slightly more expensive today than it was when I started using it several years ago. If Moore’s Law were operative, it would be four times faster at no increase in cost. The FCC’s dam has been replaced by one constructed by my cable company which cannot offer me low cost, higher speed service without cutting into its lucrative high-speed business market. While some of this blockage is due to short-sightedness, some is also due to the fundamental nature of communications.

When DEC introduced its VAX minicomputers in the 1970’s, they packed the power of the smaller IBM mainframes that cost many times more. And there was nothing IBM could do to stop customers from moving to DEC to save money. Similarly, when SUN workstations arrived on the scene ten years later with technology that was a generation ahead of the VAX, there was nothing DEC could do to prevent its customers from jumping ship to cut costs. The flow of the computational river could not be staunched by entrenched vendors.

In contrast, when a faster cable modem comes out, I can only use it if my cable company agrees and sets up a similar modem at its end. The same is true for faster DSL modems being controlled by the phone companies. While there was some hope that CLEC’s would add an extra competitive dimension, both the excesses of the Internet bubble and the nature of communications prevented that hope from being fully realized.

The Internet bubble hurt competition by breeding spendthrift companies that could not survive once the flow of capital was cut off. Initially VC’s told their companies to grow as fast as possible, without regard to cost. By and large, they would not fund ventures which planned for sustainable growth through internally generated cash flow. A few years later, when the VC’s cut off the flow of funds, they left unsustainable structures that crumbled.

But there is an even more fundamental problem. The nature of communications, where a few key players control the pipe to the end user, indicates that the flow of the communications river is likely to be fundamentally different from that of the computation river. Rather than a steady, strong flow, there will be long droughts, while a store of water develops behind the dam of a few entrenched vendors. Then, when the cost reductions that are possible become so great that they cannot be ignored, capital will flow into large ventures that breach the existing dams, but erect new ones of their own to protect their own revenue streams.

We may see innovation stifled for five, ten or even fifteen years, followed (as with the Internet Revolution) by a sudden reduction in costs by several orders of magnitude. Riding a raft down this river, which runs dry for long periods and then has a flash flood of immense magnitude, will be much less fun and much more dangerous than was river rafting on the computation river.

I see two hopes to improve this situation, one technological and one political. On the technological front, 802.11 or similar “guerilla warfare” technologies might obviate the last mile problem and create water courses that cannot be dammed. On the political front, initiatives such as Senator Burns’ “Tech 7” might bypass the last mile problem by buying off those who dam the river.

Or, if the barriers to free flow of communications cannot be removed with financial incentives, then more forceful legislation may be desirable. While in general I prefer to let free markets determine pricing and other variables, in some cases government interference is necessary. We are missing many great opportunities that may require forcing some entities to do what is not in their immediate self interests, but would greatly benefit the public good.

Footnotes:

[1] There have been several variations of Moore’s Law, such as a doubling every year or every two years. While these variations affect the quantitative statements made below, they do not change the qualitative conclusions. Whether the reduction over the last fifty years has been by a factor of ten billion or ten million or ten trillion makes little difference to the conclusions of this note, and the variations in those numbers, and related numbers, will not be mentioned again and should be understood as tacit.

[2] Since the communications revolution has been stymied by regulation and monopoly or duopoly power, we do not know for sure what this factor would be in an unregulated, free market. I estimate it to be on the order of 10^8 since fiber optics operate at frequencies of 10^14, while coax was limited to 10^7 Hz fifty years ago. That gives a gain of 10^7, with the extra factor of 10 coming from the much smaller diameter of fiber compared to coax. The cost of laying either material is due primarily to the labor involved, not the medium, so the cost per mile is proportional to the area of the medium. In the body of this paper I use the much more conservative factor of a thousand to avoid getting into arguments which are immaterial to the point of this paper.

This page can be viewed in Romanian courtesy of azoft.