The Technium

How much does one search cost?

[Update. I have modified some figures below following feedback since I first posted.]

[Update, 2. I have posted alternative ways to calculate this value in this follow up post.]

This week Google’s stock passed $625, giving it a valuation of $195 billion. This immense value is primarily built on the importance and ubiquity of search. How ubiquitous is their search? According to a report just released today by ComScore, Google served 37 billion searches in August 2007. Assuming that Google averages that many searches per month for this calendar year, they will conduct 444 billion searches in a year. According to Google’s latest financial statements, nearly two thirds of Google’s revenue comes from search, the rest from “network” Adsense sites, and other content. This ratio suggests that 64% of Google’s value is in search, or 124 billion. If Google maintains this valuation over the same time period, this makes each search — each search you do — worth $0.27! Of course, this valuation, like all those based on stock prices, has built into it the expectation of future earnings, which may be unreasonable expectations, so it is a strange kind of “worth.” Nonetheless, this market price suggests that each time we google, we add 27 cents to their valuation.

If that is how much one search is worth, how much does one search cost? If we take that same total of 37 billion, we get 111 billion searchers per quarter. According to SEC filings, Google’s Q2 quarterly operating expenses were $319 million. Not included in this figure are sales and marketing expenses, taxes, finance, or research and development. It does include admin saleries. I wish they revealed their precise costs of running and programming their data centers but they don’t, so these calculations are very rough. They might be accurate to the nearest magnitude. Using these coarse budget figures we find it costs $319 million dollars to serve up 111 billion searches and related ads. That works out to $0.0028, or about 3 tenths of a cent per search. (Did you know you can calculate problems like this using Google search?) So each time you use Google to calculate a figure, or check the weather, or find a friend, you cost it .3 cents.

That price is just the upper limit, because undoubtedly the marginal cost of adding searches by adding more servers must be way lower. A third of a cent is nearly free, and that of course is how we treat it. I’m guessing, but have not done the research, that Yahoo, Baidu and the rest of search engines have a similar economics.

I have made a clumsy attempt to estimate the full value of search by using the market cap of the largest search provider. This is unorthodox to say the least. Some folks would say you have to value a service on its revenue or its price on the street. My logic goes like this. What is the value of a car? You can say it is whatever someone will pay for it. But a car is not only valuable to the person driving it. It is valuable to employers to have mobile employees. It is of value to real estate developers to have mobile residents who can drive to your subdivision. There is societal-wide economic value to the product that are not caught by its price tag on the window. (And there are societal-wide costs to automobiles — death and pollution — that are also not captured by the price tag.) I was attempting to use a company’s market value for that service as a surrogate for the value of that service in the marketplace. In some ways, this market cap does incorporate the product’s liabilities, risks, and downsides since those are concerns to investors, if not to its buyers. This number is crude, it’s flawed, it’s not kosher, but in the absence of another I’m using it.

A Google insider made this observation about my calculations. He pointed out that all Google’s server hardware is paid for via ads. “Since the advertisers are paying for the ads, they must find that search gives them additional profit. Since customers are clicking on the search-based ads, they must get additional value.” So the .3 cent/search expense is “paid for” by the improved efficiency in matching buyers and sellers on the Internet. The value which just one search action brings to advertisers and to users is the difference between the cost of search and its appraised value, or 27 cents minus .3 cents. Let’s be conservative and call this differential 26 cents. This makes the .28 cent cost 100 times less than 26 cent value. It is this 100-fold difference between extremely low marginal costs of producing and very high value in the world which is the source of progress.

According to the same ComScore research, people around the world searched the web — using all search engines — some 67 billion times in one month (August 2007). Taking this for a rough monthly average, humans now make 804 billion searches in one year. If each search increases the efficiency and serendipity of our lives by 26 cents worth (assuming Google is a guide and it may not be), then the total yearly worth of web search is $209 billion. That’s not web search investment, that’s the increase in intangible wealth to society yielded by the collective searching of humans in one year.

I don’t expect anyone to seriously believe these amateur calculations; they are only intended to provide some scale. The value of web search is in the hundreds of billions. But what is most interesting to me, is that this huge “industry,” this huge appetite, has materialized out of nowhere. If we ask 800 billion questions today and expect to get instant answers, where were those 800 billion questions 20 years ago?

No one would have believed 20 years ago there was a $200 billion business in answering peoples questions (for free!). There weren’t many MBAs dreaming to fill this need. The need for question/answers was latent and invisible. More accurately, it was driven up by supply. People didn’t know how valuable instant answers were until they had them. Twenty years ago there were answer businesses, and they were not all that small.

The largest answer business was phone directory assistance. Before there was search, there was 411. The number of 411 calls has remained steady for decades: about 4 billion per year. Despite the free web, people still make billions of 411 calls on their phones, even at $1.25 per call. Cell phones have increased that usage a billion plus, to a total of 6.6 billion calls in 2004.

The other search mechanism in the past was the yellow pages — I mean the paper version. According to the Yellow Page Association, 50% of American adults used the print yellow page at least once a weak, performing 2 lookups per week in the 1990s. Since the adult population in the 90s was around 200 million, that’s 200 million searches per week, or 104 billion questions asked per year. Nothing to sneeze at.

The other classic search route is the library. US libraries in the 1990s counted about 1 billion library visits per year. Out of those 1 billion, about 300 million were “reference transactions,” or questions. We could be generous and count any visit to the library a “search,” as in searching for a book or topic.

Our grand total pre-web Q&A tally in the US would thus sum up this way:

1 billion library visits

6.6 billion yellow page lookups

104 billion directory assistance


111 billion searches

So in fact, we’ve always had questions. But now that we have better quicker answers we gained less “costly” ways to answer them. One study conducted in 2000 determined that the average American adult seeks to answer 4 questions per day. If my own life is any indication, I am asking more questions every day. Google tells me that I asked it 349 questions last month, 10 per day, (and my peak hour of inquiry was 11 am on Wednesdays).

Since all these earlier search methods have not declined after the web (in fact they continue to rise) we can add them to web search for a grand total. First I would double the US figure to arrive at a conservative global amount. (The usual rule of thumb is to triple US figures to arrive at approximate global figures.)

222 billion non-web searches

804 billion web searches


1, 026 billion total searches

Over a trillion questions are answered each year. And we have just started getting good at giving answers. As search answers get better, and asking questions gets easier (think natural spoken English in the near future), we are headed to a future where we will ask several hundred questions per day. The “manufacturing costs” of each answer will be nano-cents, while the value of answering them continues to climb.

But do users really value search if they don’t pay for it? Yes. Oxygen has value even though it is usually free. We only pay for air when it is not abundant, as in underwater or high altitudes. We’d value air at a million dollars if we didn’t have it. But day to day we pay nothing for it. It’s free. Search is a little bit like this. It is the uncharged free but essential air of the internet.

Once you start to live with search, swim in it. Live in it, and depend on it, you can’t live without it. (See Clive Thompson’s essay on how dependent we are already.) I have already started paying for improved search. I pay for answers from Google Answers and Uclue, and hire other researchers. I would be willing to pay for search — by the search or subscription — if it were substantially better. I believe many people would pay for a voice activated AI search — an answer box — if it really did work like that. But few, myself included, would have ever considered paying 20 years ago.

Search has been considered a non-essential luxury. But once it becomes ubiquitous, good, and free, we will find it is no longer a luxury but essential. As long as there are biz models that will deliver this good for free, this value will be hidden. The free in fact disguises its value.

I believe we are at a cusp where we become increasingly dependent on search. It is possible we will never pay much or anything for search as searchers, except for premiums on superior search. But this doesn’t mean we will not value it tremendously. Both individually and societally. While it is true that if people actually had to pay for search our habit of search would at first be immediately and dramatically curbed. But you could say the same thing about roads. People’s driving habits would be dramatically curbed if they had to pay tolls for roads. Yet we value roads very highly. And toll roads make money.

Part of the increasing value in providing answers lies in the fact that past questions and past correct answers increase the value of the corpus. Each question we ask a search engine and each answer we accept refines the intelligence of the process, increasing the engine’s value for future questions. As I have argued before, as answers are perfected and become too cheap to meter, the real new value will shift towards great questions.

Perhaps we will reach the time when we share our thinking with this answer machine, so that “search” becomes synonymous with “think.” Cognitively, “think” is just search for a solution in a high-dimension of variables, so we can consider all thought as a type of search. I have often wondered what we would do with petahertz/petabyte computers. Or exahertz/exabyte computers after them. YouTube won’t max it out. Even with mashing hi-definition 3D virtual reality 24 hours a day, there may be a lot of spare cycles. I think we are going to fill that extra room with thinking-like search. Our 444 billion searches per year will happen in a few seconds.

I asked Google how many seconds in a year and it instantly told me: 31.5 million. That means that today 14,000 searches are performed on the web every second. Considering the web as its own global machine, search is running at 14 kilohertz. If we could audibly hear each click of the mouse as everyone searched, the resulting sound — vibrating at 14 kilohertz — would be a high pitch hum right at the edge of human hearing. Hear it, hmmmmmm?

Each of those search cycles cost a third of a cent. But the search function itself in this great machine (not counting the rest of the web) is alone worth 300 billion dollars. We can only imagine what the value of the web itself is.

If you have an estimate, even a rough one, I’d be interested to here it. Comments here are broken so email me: kk at kk dot org.


© 2023