Web Hosting Statistics Updated November 2026

Human Written & Fact Checked · Updated

100 Website Performance Statistics for 2026

A fast connection does not guarantee a responsive page. This evidence separates browser experience, connection quality and observed business behaviour so you can see what each measurement actually supports.

Technical benchmarks use July 2025 page data and Cloudflare's 1 January–2 December 2025 reporting window. The commercial study covers 28 October–24 November 2019 and is retained as foundational evidence, not a current forecast. Its figures are relative modelled associations with a study-defined 0.1-second speed improvement, not guaranteed causal effects.

Headline figures

  • Eligible sampled web pages/sites: Mobile sites with good overall Core Web Vitals: 48% (July 2025). (HTTP Archive / Google CrUX, Jan 2026)
  • Eligible sampled web pages/sites: Desktop sites with poor FCP: 9% (July 2025). (HTTP Archive / Google CrUX, Jan 2026)
  • Eligible sampled web pages/sites: Mobile sites with good TTFB: 44% (July 2025). (HTTP Archive / Google CrUX, Jan 2026)
  • Eligible sampled web pages/sites: Desktop sites with good INP: 97% (July 2025). (HTTP Archive / Google CrUX, Jan 2026)
  • Eligible sampled web pages/sites: Mobile pages with good CLS: 81% (July 2025). (HTTP Archive / Google CrUX, Jan 2026)
  • Eligible sampled web pages/sites: Mobile lab median Total Blocking Time: 1,916 ms (July 2025). (HTTP Archive / Google CrUX, Jan 2026)
  • Spain: Spain speed-test average download speed: 318 Mbps (1 Jan–2 Dec 2025). (Cloudflare, Dec 2025)
  • Spain: Spain speed-test average upload speed: 206 Mbps (1 Jan–2 Dec 2025). (Cloudflare, Dec 2025)
  • Moldova: Moldova speed-test average loaded latency: 73 ms (1 Jan–2 Dec 2025). (Cloudflare, Dec 2025)
  • Participating European and US brand sites: Retail mobile conversion rate association: +8.4%. Modelled for the study-defined 0.1-second speed improvement (28 Oct–24 Nov 2019). (Fifty-Five / Deloitte, Mar 2020)
  • Participating European and US brand sites: Travel mobile conversion rate association: +10.1%. Modelled for the study-defined 0.1-second speed improvement (28 Oct–24 Nov 2019). (Fifty-Five / Deloitte, Mar 2020)
  • Participating European and US brand sites: Lead-generation mobile conversion rate association: -1.9%. Modelled for the study-defined 0.1-second speed improvement (28 Oct–24 Nov 2019). (Fifty-Five / Deloitte, Mar 2020)

How quickly do measured pages show their content?

Loading performance spans the network, server and browser. A single server-response measurement cannot describe the entire visible experience.

What do first-paint and overall experience measures show?

Overall Core Web Vitals and first paint are different assessments. The former combines loading, responsiveness and stability; the latter identifies the first visible response. These field results reflect eligible Chrome user experiences represented in the dataset instead of a census of every browser or website. Device categories should remain attached to comparisons. A page can display something quickly while still taking longer to show its main content or respond to input. The figures therefore identify distinct aspects of observed experience, not interchangeable scores for the quality of a web host.

How does first-byte performance vary by device?

Time to First Byte includes more than application processing. Redirects, connection establishment, network conditions and caching can influence the measurement. A poor result therefore does not independently identify a deficient server or hosting company. The source reports rounded shares in performance categories, which may not sum exactly to the full population. These are field measurements from eligible traffic, not timings from a single synthetic server benchmark. Retain the device category and dataset date when comparing them with operational monitoring for an individual website.

What do content rendering and image measurements reveal?

LCP describes the largest visible content element. Image prevalence and the Lighthouse audit come from crawl observations, while the performance categories use field data. Lazy loading can affect discovery of the main image. These measures describe separate rendering stages.

How responsive and stable are measured pages?

A page can appear promptly while responding slowly to an interaction. Field responsiveness, lab blocking and visual stability capture different experiences and should stay separate.

How does interaction responsiveness differ by device?

Interaction to Next Paint measures how promptly a page presents feedback after an interaction. These categories describe eligible real-user measurements, not the time needed to download the page. Desktop and mobile populations differ in devices, networks and browsing patterns. The observations do not isolate any one cause of their performance difference. They also cannot establish how an unmeasured page behaves or identify a specific third-party script responsible for delay. For an individual site, measurement needs to preserve the affected interaction and audience instead of substituting a global device-category average.

How common are visual-stability problems?

Cumulative Layout Shift measures unexpected movement instead of elapsed time. Its score has no millisecond unit, so it cannot be averaged with loading or responsiveness timings. The category distributions describe eligible pages in the source's field dataset. They do not establish which design choice caused a shift. Images, font changes, injected elements and other layout behaviour can contribute under different conditions. Rounded percentages can also leave a small difference from a complete total. Compare the same metric, device category and collection method before interpreting a change as an improvement.

How does lab blocking vary across the distribution?

Total Blocking Time is a controlled-test metric, not the same measurement as Interaction to Next Paint. Percentiles describe positions within a distribution and should not be added together. A high upper percentile shows a slower part of the measured population without describing the typical user of every website. The test conditions and page selection influence these values. They can help identify browser-workload problems, but they do not isolate the hosting layer. A site-specific investigation should distinguish script execution, resource delivery and the actual interactions users attempt.

How do connection speeds vary across countries?

Connection throughput provides delivery context, not a webpage load-time ranking. These country averages come from voluntary speed tests and cannot represent every household or visitor.

Which download averages appear in the European results?

  • Hungary speed-test average download speed: 305 Mbps. (Cloudflare, Dec 2025)
  • Portugal speed-test average download speed: 297 Mbps. (Cloudflare, Dec 2025)
  • Denmark speed-test average download speed: 247 Mbps. (Cloudflare, Dec 2025)
  • Romania speed-test average download speed: 237 Mbps. (Cloudflare, Dec 2025)
  • France speed-test average download speed: 237 Mbps. (Cloudflare, Dec 2025)

The measurements aggregate tests initiated through Cloudflare's speed-test service. People may run a test because a connection is unusually good, unusually poor or being investigated. That selection limits extrapolation to the national population. The reported averages are not medians or guaranteed service-plan speeds. They also do not tell us how quickly a particular website becomes usable. Resource size, server response, caching and browser processing remain outside this connection-throughput comparison. Geographic rankings should therefore retain the measurement method and reporting window instead of being presented as universal national internet quality scores.

What other download results and changes were observed?

  • Chile speed-test average download speed: 271 Mbps. (Cloudflare, Dec 2025)
  • South Korea speed-test average download speed: 260 Mbps. (Cloudflare, Dec 2025)
  • Brazil speed-test average download speed: 248 Mbps. (Cloudflare, Dec 2025)
  • Israel speed-test average download speed: 244 Mbps. (Cloudflare, Dec 2025)
  • Spain speed-test year-over-year change in average download speed: +25 Mbps. (Cloudflare, Dec 2025)
  • Hungary speed-test year-over-year change in average download speed: +46 Mbps. (Cloudflare, Dec 2025)

The country averages describe the same voluntary testing system, while the reported changes compare its annual observations. A change in the people and connections taking tests can influence the comparison alongside infrastructure changes. The year-over-year values are absolute throughput differences, not percentage increases. They should not be added to another country's average or interpreted as a controlled estimate of an upgrade's benefit. Faster connection throughput can coexist with slow page rendering when another part of the delivery or browser process is the limiting factor.

Which upload-speed thresholds did the source report?

  • South Korea speed-test average upload speed: over 130 Mbps. (Cloudflare, Dec 2025)
  • Macau speed-test average upload speed: over 130 Mbps. (Cloudflare, Dec 2025)
  • Singapore speed-test average upload speed: over 130 Mbps. (Cloudflare, Dec 2025)
  • Japan speed-test average upload speed: over 130 Mbps. (Cloudflare, Dec 2025)

The source describes these locations with a threshold instead of giving an exact average in the surrounding text. The inequality is retained instead of inventing precision. Upload performance concerns traffic sent from the client and is particularly relevant to sending files or interactive applications. It is not the same direction as downloading a page's resources. The tests remain self-selected, and their geographic averages do not establish the experience of every visitor. Compare upload and download measurements with their units and direction intact when assessing a connection's suitability for a workload.

How does latency change when a connection is busy?

Throughput and latency answer different questions. A connection can move large amounts of data while still taking time to exchange the small messages an interaction depends on.

Which idle-latency averages were reported?

  • Iceland speed-test average idle latency: 13 ms. (Cloudflare, Dec 2025)
  • Moldova speed-test average idle latency: 15 ms. (Cloudflare, Dec 2025)
  • Portugal speed-test average idle latency: under 20 ms. (Cloudflare, Dec 2025)
  • Spain speed-test average idle latency: under 20 ms. (Cloudflare, Dec 2025)
  • Hungary speed-test average idle latency: under 20 ms. (Cloudflare, Dec 2025)

Idle latency is measured from the speed-test application's perspective when it is not actively consuming bandwidth. The source reports exact averages for some countries and threshold descriptions for others; those levels of precision are preserved. These results do not measure the round trip to every possible hosting location. Network routing, distance and test endpoints influence what is observed. They also cannot predict browser rendering or application execution time. A website's audience may use a different path and connection mix from the people who chose to run these tests.

Which locations remained below the reported loaded-latency threshold?

  • Hungary speed-test average loaded latency: under 100 ms. (Cloudflare, Dec 2025)
  • Spain speed-test average loaded latency: under 100 ms. (Cloudflare, Dec 2025)
  • Belgium speed-test average loaded latency: under 100 ms. (Cloudflare, Dec 2025)
  • Portugal speed-test average loaded latency: under 100 ms. (Cloudflare, Dec 2025)
  • Slovakia speed-test average loaded latency: under 100 ms. (Cloudflare, Dec 2025)
  • Slovenia speed-test average loaded latency: under 100 ms. (Cloudflare, Dec 2025)

Loaded latency measures delay while the test actively consumes bandwidth. It describes a different condition from the idle measurement, not an alternative estimate of the same uncongested path. The source supplies a threshold for these country averages instead of exact individual values. That prevents ranking countries within this group. The observations are useful context for interactive workloads but cannot determine the latency of a specific hosted application. Endpoints, routing, concurrent traffic and test participation all matter. Compare measurements under equivalent load before attributing a difference to the server or delivery provider.

What did the historical retail and travel study observe?

The behavioural evidence below is historical and observational. It describes modelled associations in participating brands, not a promise that a hosting change will produce the same commercial result.

How did retail funnel progression relate to the speed model?

The percentages are relative modelled changes linked to the study-defined improvement in mobile speed. They are not percentage-point changes or the results of randomly assigning visitors to faster hosting. Different steps also have different denominators: moving to a product page is not the same outcome as purchasing. The negative early-stage association is retained alongside positive observations. The study's brand selection and historical period limit extrapolation to another retailer. These findings support measuring user journeys carefully, instead of treating a single funnel step as a substitute for completed transactions.

What did order value and bounce measurements show?

These outcomes describe separate retail and travel samples. Average order value is conditional on an order, whereas bounce rates concern visits to specified page types. They should not be combined into a revenue forecast without the missing traffic and conversion denominators. The report used naturally occurring speed variation and selected statistically significant model results. It did not establish that the same change would occur after a particular hosting migration. The source's detailed tables are used here because some narrative summaries contain inconsistent values or descriptions.

How did travel progression and engagement relate to speed?

The progression observations describe transitions between distinct stages of the travel journey. Page views and order value measure different outcomes and cannot be added to the progression percentages. These are relative associations with the report's speed model, not guaranteed effects for an individual booking service. The findings also predate the current dataset used in the technical sections. Keeping that period visible prevents an older behavioural study from being mistaken for a contemporary market benchmark. An operational decision needs direct measurement of the actual booking flow and its visitors.

What did the historical luxury and lead-generation study observe?

An action labelled conversion can mean different things across businesses. The report's luxury outcomes include adding to a basket or contacting a brand, while lead-generation outcomes concern completed forms.

How did luxury progression relate to the speed model?

These are stage-specific relative associations within the study's luxury-brand sample. Adding to a basket may function as saving an item for later, so progression does not necessarily indicate a completed purchase. The contact journey is distinct from the basket journey and should not be merged with it into a single conversion rate. The source also notes that progression beyond product details was small. Large relative changes can therefore describe a narrow baseline population. They do not establish the monetary value of a hosting upgrade or predict outcomes for another luxury website.

What did luxury bounce and conversion measures show?

The bounce results are separated by page type, while the conversion measure uses the study's luxury definition. These denominators cannot be substituted for one another. The observations come from naturally occurring changes in speed, analysed through the report's statistical model. They are historical associations instead of randomised causal estimates. The report's luxury page-view figure differs between its landing-page summary and detailed PDF, so that metric is excluded. Keeping this uncertainty explicit is more useful than selecting the more favourable number or blending the two into a new estimate.

How did lead-generation engagement and progression relate to speed?

  • Lead-generation mobile homepage-to-form-step-one progression association: +5.5%. (Fifty-Five / Deloitte, Mar 2020)
  • Lead-generation mobile form-step-one-to-submission progression association: +21.6%. (Fifty-Five / Deloitte, Mar 2020)
  • Lead-generation mobile homepage-to-information-page progression association: +11.1%. (Fifty-Five / Deloitte, Mar 2020)
  • Lead-generation mobile homepage bounce rate association: -3.9%. (Fifty-Five / Deloitte, Mar 2020)
  • Lead-generation mobile informational-page bounce rate association: -8.3%. (Fifty-Five / Deloitte, Mar 2020)
  • Lead-generation mobile page views per session association: +6.9%. (Fifty-Five / Deloitte, Mar 2020)

The source separates informational-page visits from the contact-form journey. Its model can associate speed with improved engagement even when the overall conversion result moves differently; the latter is retained in the takeaways. These results should not be interpreted as a universal increase in leads. They describe selected participating brands and their historical analytics definitions. Relative progression changes also cannot be added across stages to estimate a total funnel effect. Evaluate the completed business outcome separately from browsing activity when applying the study's questions to another website.

Glossary

Methodology and sources

Every figure on this page comes from the organisation that produced it: regulators, statistics offices, public bodies, research institutes, and companies reporting their own data or surveys. Each one links to the original report or dataset and carries its release date. Survey results keep their population and fieldwork period, and financial figures keep the measure the producer used. Figures from different producers are not added together or ranked against each other. The page is reviewed monthly and updated when a producer releases newer data.