← All resources

What the world’s best-known websites actually cost, and why none of them are your benchmark.

One government website cost more than two billion dollars. One of the ten most-visited sites on earth spends $3.4 million a year to keep online. Both figures are published, both are true, and neither tells you what your project should cost — but the gap between them explains almost everything about how web budgets go wrong.

Nobody publishes what a website cost. Agencies sign NDAs, in-house teams bury the spend across payroll and cloud bills, and the few real numbers that reach daylight do so because a public auditor forced them out or a founder chose to publish them. That makes the handful of genuinely documented figures unusually valuable — not as benchmarks, but as a way of seeing which parts of a budget are actually load-bearing. What follows are the ones with a paper trail: audited government projects, a foundation’s published accounts, and a software company that put its infrastructure bill on the internet. The pattern they share is more useful than any of the individual numbers.

Websites whose real costs are a matter of public record

WebsitePublished figureWhat that number actually covers
HealthCare.gov — US federal insurance marketplace$840 million to $2.1 billionThe GAO counted $840M across 62 contracts to 2014; the HHS Inspector General put associated contract value at $1.7B; Bloomberg Government’s independent tally reached $2.1B. The spread is a definition problem, not a dispute
GOV.UK Verify — UK identity service£217M in benefits, against £837M forecastThe National Audit Office’s revised figure — 75% below the 2016 business case, with 3.6 million users against a 25 million target
GOV.UK — the UK government’s main website£61.5 million saved per yearThe Cabinet Office’s own accounting of annual savings from consolidating hundreds of departmental sites into one platform
Wikipedia — a top-ten global website$3.4 million per year, hostingThe Wikimedia Foundation’s 2025 internet hosting line, against roughly $180 million in total annual expenses. Serving the content is not where the money goes
37signals — Basecamp and HEY$3.2M/year, cut to under $1MPublished infrastructure spend before and after moving off the cloud onto owned hardware — about $2M saved annually, $10M projected across five years

Build costs and running costs are different measures and are not comparable line for line — HealthCare.gov’s figure is what it cost to create, Wikipedia’s is what it costs to keep serving. They are placed together deliberately: the contrast is the point.

The most expensive website ever built, and what the money actually bought

HealthCare.gov is the only consumer-facing website in history with a nine-figure audited price tag, and the range in that price tag is itself instructive. The US Government Accountability Office counted $840 million spread across 62 separate contracts through 2014. The Department of Health and Human Services’ own Inspector General valued the associated contracts at $1.7 billion. Bloomberg Government, counting the supporting systems the site depended on, arrived at $2.1 billion. All three are defensible; they simply drew the boundary in different places.

That boundary problem is not a Washington quirk — it is the single most common reason web quotes appear to differ wildly. A quote for “the website” might mean the pages a visitor sees, or it might include the CRM integration, the payment reconciliation, the data migration, the staff training and the year of support afterwards. When two quotes for the same brief are 3× apart, the usual explanation is not that one supplier is greedy; it is that they are pricing different scopes and nobody has said so out loud.

What actually failed was not the code

The GAO’s finding on HealthCare.gov is worth reading closely, because it does not blame the engineering. Its conclusion was that the agency “undertook the development of HealthCare.gov and its related systems without effective planning or oversight practices,” while facing changing requirements throughout. The money went not on building the thing once, but on building it repeatedly as the specification moved underneath the people building it.

The UK produced a cleaner illustration of the same failure. GOV.UK Verify was forecast in its 2016 business case to deliver £837 million in benefits; the National Audit Office later revised that to £217 million — 75% lower — with 3.6 million users signed up against a 25 million target. Nothing was technically broken. The premise was wrong, and no amount of delivery quality rescues a wrong premise. This is the argument for spending real money on discovery before spending any on build, and it is why a fixed scope agreed in writing is worth more to a client than a low hourly rate.

What it costs to run one of the ten biggest websites on earth

Now the number that surprises people most. Wikipedia — consistently among the ten most-visited websites in the world, serving billions of page views a month in hundreds of languages — spent $3.4 million on internet hosting in 2025. Against the Wikimedia Foundation’s total annual expenses of roughly $180 million, hosting is under 2% of what the organisation spends.

The implication for a normal business is direct and mostly reassuring: at any traffic level you are realistically going to reach, hosting is not your cost problem. A business site serving a few hundred thousand visits a month costs a rounding error to host. If a proposal makes hosting sound like a major line item, that is a signal worth questioning — it is where the money goes least. What Wikipedia spends its actual budget on is people, and so will you.

The one number here that maps onto a real business

The most transferable figure on the list comes from 37signals, the company behind Basecamp. They published their infrastructure spend before and after leaving the cloud: $3.2 million a year on cloud services, cut to well under $1 million after buying their own hardware — roughly $2 million saved annually, and a projected $10 million over five years. That is a real software company, running real products for paying customers, publishing real numbers.

The lesson is not “leave the cloud” — that decision depends entirely on scale and on having the staff to run hardware. The lesson is that the recurring bill deserves as much scrutiny as the build quote, and almost never gets it. Businesses negotiate hard on a one-off build price, then sign a monthly stack of subscriptions that quietly exceeds it within two years. Ask what the thing costs to run in year three, not just what it costs to launch.

Why none of these figures are a benchmark for you

Every project above carries costs that have nothing to do with the website. HealthCare.gov had to integrate with the IRS, the Department of Homeland Security and dozens of private insurers, under federal procurement rules, with legally mandated deadlines. GOV.UK consolidated hundreds of departmental websites while maintaining statutory accessibility obligations. Wikipedia’s spend is dominated by grants and staff, not the site.

Strip those away and the actual website underneath each is not exotic. That is the useful conclusion: the price of a website is set almost entirely by what it must connect to, who must approve it, and how much of it is genuinely bespoke — not by how it looks or how many pages it has. A brochure site with three integrations costs more than a fifty-page site with none.

What actually moves the number on a real quote

In practice, four things drive the difference between a €499 project and a €15,000 one. First, integrations: every external system — payments, CRM, ERP, booking, invoicing — adds work that scales with how badly the other system is documented. Second, content: a site with copy and photography ready to go is a fraction of the cost of one where those have to be produced. Third, languages: a Multilingual Website is not one site times three, but it is genuinely more than one site. Fourth, bespoke functionality — anything a visitor can do beyond reading and submitting a form.

What barely moves the number, contrary to expectation: page count, visual ambition within reason, and hosting. Ten well-built pages and thirty well-built pages are not three times apart in cost, because the expensive part is the system, not the repetition. This is why our Business Website and Landing Page are priced by what they do rather than by size.

What honest numbers look like at the other end of the scale

We publish our own prices for the same reason the figures above are useful: an unpriced quote invites the boundary problem that cost HealthCare.gov its reputation. Our full breakdown lives in the price guide, and the packages themselves are on Prices with what each one includes and excludes stated before you talk to anyone.

If you are trying to establish whether a number you have been quoted is reasonable, the most useful thing you can do is not compare it to another number — it is to ask both suppliers to write down the scope boundary in the same words. The gap will usually explain itself. If you would rather have someone else read the quote, a Strategy Session is a cheaper way to find out than starting the project and discovering the difference at month three.

Sources

Every figure above comes from a public audit, an official accounting or a company’s own published numbers. Checked August 2026.

Frequently asked questions

Want to know what your project would actually cost?

Tell us what the site has to do and what it has to connect to, and we will give you a fixed number with the scope boundary written down.