Nielsen's Law of Internet bandwidth states that:

• a high-end user's connection speed grows by 50% per year
• you don't get to use this added bandwidth to make your Web pages larger until 2003

The dots in the diagram show the various speeds with which I have connected to the Net, from an early acoustic 300 bps modem in 1984 to an ISDN line when I first wrote this article (and updated to show the 31 Mbps cable modem I got in 2010). It is amazing how closely the empirical data fits the exponential growth curve for the 50% annualized growth stated by Nielsen's Law. (The y -axis has a logarithmic scale: thus, a straight line in the diagram represents exponential growth by a constant percentage every year).

Nielsen's Law is similar to the more established Moore's Law. Unfortunately, comparing the two Laws shows that bandwidth grows slower than computer power. Moore's Law says that computers double in capabilities every 18 months: this corresponds to about 60% annual growth. As shown in the table, bandwidth will remain the gating factor in the experienced quality of using the Internet medium.

Average bandwidth increases slowly for three reasons:

• Telecoms companies are conservative: they need to dig up streets and install equipment in hundreds of thousands of central offices so they think twice (or thrice) before investing the necessary billions of dollars. Even after they invest, it takes time to update their sprawling physical plant.
• Users are reluctant to spend much money on bandwidth. If you buy twice as fast a computer, your software runs twice as fast; if you buy twice as large a harddisk, you can store twice as many files. But if you buy twice as fast a modem, then you don't download Web pages twice as fast: the speed of the Internet is a function of both the individual user's connectivity and of the infrastructure. You don't get the full benefits of your own bandwidth upgrades immediately — only gradually as the Internet and the host servers improve.
• The user base is getting broader all the time as mainstream users get online. These new users are more likely to be low-end users than high-end users (all the geeks have been online for years), so the average shifts ever lower.

Of course, there are many technologies to deliver faster bandwidth, and Bell Labs has already demonstrated the ability to shoot a terabit per second down an optical fiber. Unfortunately, these technologies will not deliver huge bandwidth increases to the masses any time soon. ADSL and other technologies will give individual users the speed of a T-1 line and better, but this will not happen in large numbers until approximately 2003.

Design Implications of 50% Annual Bandwidth Growth

There will always be a few super-users who have advanced equipment that runs really, really fast. Nielsen's Law addresses the more normal high-end user who is willing to pay a premium but still wants well-tested equipment that can be bought in a regular shop. This is the kind of user who may have an ISDN line today.

The vast masses of users are low-end who will lag 2-3 years behind the high-end users. Bandwidth is one of the two most important elements in computing these days (together with screen quality), since computational speeds are almost always more than enough for non-engineering tasks. Unfortunately, I can argue as much as I want: most users still save on bandwidth and prefer a $20/month ISP over a $30/month one with better service.

Web design needs to cater to the masses. Only rarely can a site be successful if it is aimed at the most advanced 10% of users. Thus, even though high-end users may have ISDN these days, Web design must aim at optimal usability over a 28.8 kbps modem. International users have even slower connections and response times across the oceans will likely get worse over the next few years.

For the next five years, the Web will be dominated by users with so slow connections that any reasonable Web page will take much longer to download than the response time limits  indicated by human factors research. Thus, the dominant design criterion must be download speed  in all Web projects until about the Year 2003. Minimalist design rules.

Starting about 2003, high-end users will have speeds corresponding to a personal T-1 line. This will allow them to download pages in less than a second, meaning that they will be able to navigate the Web freely. The user experience will become radically more gratifying with subsecond response times.

Of course, low-end users will be on ISDN lines in 2003, so high-end users' megabit access will still not sanction bloated design. Looking even further ahead, Nielsen's Law does predict that the Web will be 57 times faster in ten years. At that time, even low-end users will be able to access multimedia designs, and the high-end users will be able to use very advanced sites. The future of the Web holds great promise for much richer designs. It is simply that the current Web is so horribly slow that it will take five years to achieve acceptable response times. Only after 2003 can Web design change direction and aim at higher bandwidth.

Update 2008

10 years after writing this article, I updated the chart with data points showing the bandwidths from the cable modems I have had during the period 1998-2008. My latest cable provider gives me 16 Mbps, which is 122 the speed of the ISDN modem I had when I wrote the original article.

Despite these seemingly large changes, the new data points fit the chart perfectly. (I got the ISDN modem in 1996, and a growth factor of 122 over 12 years equals 49% per year — incredibly close to the 50% per year I predicted in April 1998.)

Update 2010

2 years later — another upgrade. Now to 31 Mbps. This new data point also fits the prediction from 1998.