The Search Engine Start-up Myth

or why Google’s hundreds of thousands of servers and billions of dollars a year in hardware costs are not required for a starting search engine

Google SkyNet

Google has an impressive amount of servers and data centers. The total number of servers is unknown but it is believed it runs into the hundreds of thousands. The Un-Official Google Datacenter FAQ (http://www.datacenterknowledge.com/archives/2008/Mar/27/google_data_center_faq.html) reports there are dozens of data centers and including data centers that are under construction, Google has 19 locations in the US, 12 in Europe, one in Russia, one in South America, and three in Asia. Not all of the locations are dedicated Google data centers, since they sometimes lease space in other companies’ data centers. In 2006, Google spent 1.9 billion dollar on its data centers and hardware. In 2007, that number was 2.4 billion dollar. Google has many projects and experiments going and it’s philosophy is that a combination of large data sets, processing power, and brilliant engineers will lead to new and competitive products in the future.

Barrier to entrance?

It is often heard that starting a competing search engine would be extremely expensive due to the costs of the massive amounts of required hardware and costs of operating that hardware. Who would be able to come up with at least the hundreds of millions of dollars required to run the server farm required for a large scale search engine? The reality however is that most of these operating costs are related to performing searches and not to building the search index. And searches also make money… In 2007, every search performed on Google generated about 12 cents on average. When counting the data center costs only towards search results (search still brings in almost all the money and carries the other initiatives) Google’s costs per search results page are roughly somewhere between 1.75 and 2 cents (for the calculation, see the overview provided by the table below. Some items are unknown but extrapolated from other years).

The amount of searches performed every month is not disclosed by Google but are estimates based on research of different organizations like ComScore and Nielsen. Combining the amount of searches performed and the revenues and costs reported by Google in their annual report leads to an estimated 12 cents revenues per search results page.

2005
2006
2007




US searches per month
2,700,000,000
4,177,694,576
5,877,036,708


Global searches per month (incl. US)
5,115,466,667
8,034,028,030
11,523,601,389


Part of revenues from abroad (non-US)
47%
48%
49%


















Revenues per search page (RSP)
$0.10
$0.11
$0.12


Costs per search page (CSP)
$0.04
$0.04
$0.05


Monthly revenues US
$270,000,000
$459,546,403.33
$705,244,405.00


Monthly revenues global
$511,546,667
$883,743,083
$1,382,832,167


Annual revenues global
$6,138,560,000
$10,604,917,000
$16,593,986,000


Annual costs
$2,577,088,000
$4,225,027,000
$6,649,085,000




Gross profit
$3,561,472,000
$6,379,890,000
$9,944,901,000








Google datacenter costs
$1,288,544,000
$1,900,000,000
$2,400,000,000


Percentage of costs of service
50%
45%
36%


Datacenter/Servers costs per page
$0.0210
$0.0197
$0.0174

Of course, there is a substantial investment related to collecting all the information captured in the search index and also to setting up the initial batch of computers storing that index and performing searches for users. This is because an index containing information about billions of websites requires a certain amount of computers that together store the index. These days, the newest generation of computers can easily store many times more information than the original Google computers did. In size, the investment in hardware to start serving users would only be a fraction of the amount of money that Google currently spends on its operations. And as user volumes increase, so do revenues. These new revenues then can be used to cover the costs of the additional hardware required to service the additional users.

The fact that the revenues from each search are about 5 to 10 times higher than the costs of that search (including the hardware costs of other projects!) also means that each additional user on average generates more additional revenues for Google than additional costs. For a new web search player, the initial costs of building a great search engine are a substantial though do-able investment and the hundreds of millions of dollars in service costs only occur when massive amounts of visitors are realized. At that time however, the revenues to cover the costs are also generated by these visitors. With a solid monetization model, new search engines can be started without an investment requirement of hundreds of millions of dollars as is sometimes heard as the reason why it is hard to start a new search engine. With a disciplined costs structure on hardware spending and efficiency (historically not Google’s focus, much more can be done on cost control) and a clear market focus and model to make money, a search start-up can still be successful. Having a successful media market strategy is a subject for another post.

(re-published from original post in 2008)