Web Hosting Statistics 2026: Provider Market Share and IPv6 Adoption
More domain names point at GoDaddy's parking service than at any web host on the internet: 22,239,769 of them, against 20,981,382 on the largest host. Only 1.3% return a web page, because these are names held rather than used. GoDaddy runs that parking service on IP addresses Amazon owns, so a ranking built on who owns the address credits every one of those names to AWS. It is a large part of how AWS comes to look like a quarter of the web. Count who actually runs the service instead and no company holds more than 12.8% of the 292,860,014 domain names that publish an IP address.
- Rank hosting by the provider actually running the service rather than by who owns the IP address, and the largest is Cloudflare at 12.8% of every domain that resolves, 37,403,719 domains serving a page on 80.0%. Second is GoDaddy parking (on AWS) with 22,239,769, serving 1.3%.
- 55.6% of the domains resolving into Cloudflare use GoDaddy's nameservers, 41,548,631 domains, and 22,219,787 of those sit on the four IP addresses where at most 1.4% serve a page. The largest registrar runs its DNS and its landing pages on the largest cloud, so a ranking by IP address credits AWS for them. Ten networks hold 48.3% of everything that resolves and thirty hold 69.0%.
- IPv6 reaches 23.8% of the domains that resolve, 69,736,003 of 292,860,014. 97.5% of those also publish an IPv4 address, so only 1,773,477 domains are reachable over IPv6 alone.
- IPv6 is a decision your hosting provider makes, not one you make. Among networks holding a million domains or more it runs from 94.1% on Strato to 0.7% on XServer Cloud, a spread of 134.4 to one.
- 6.0% of domains that resolve publish an IP address where nothing answers: 15,449,376 never answered before the deadline and 1,384,497 refused the connection outright.
- 57.2% of domains with an A record publish exactly one IPv4 address, 166,409,314 domains with no second address to fall back to.
- Multi-location hosting barely exists at the domain level. 0.8% of domains publish IP addresses belonging to more than one company and 0.8% to more than one country, measured on the bare name rather than on any hostname below it. What looks like multi-location is almost always anycast inside a single network.
- The busiest IP address on the internet is 3.33.130.190, which 22,313,054 domain names point at and 1.4% of them get a page from. The busiest IP address that is genuinely a web host is 185.230.63.107 on Wix, with 7,131,754 domains and 97.0% serving.
- 263 IP addresses have more than a hundred thousand domain names pointing at them each, 53.9% of every A record on the internet. At the other end 7,919,817 IP addresses have exactly one.
- 59.1% of domains that resolve publish an IP address registered in the USA, but 21.6% of those sit behind a reverse proxy and more again are parked inventory. Take both out and 93,089,531 are left, 54.4% of the American total.
- GoDaddy answers DNS for 53,913,594 domains and Cloudflare for 44,985,559. Ten companies serve the DNS for 49.9% of every domain that publishes a nameserver.
- 14,997,760 domains publish a null MX, the record that exists to say this domain accepts no mail. That is 8.7% of every domain with an MX record.
How this report counts hosting
Six words carry this report and get used loosely everywhere else. Address here always means an IP address, never a domain name. A network is the company the address registry files the block to. A provider is the company whose service actually answers there. Those two diverge constantly, and the gap is the subject of most of what follows: GoDaddy's parking pages answer on addresses Amazon owns, so the network is Amazon and the provider is GoDaddy. An autonomous system is a third unit again, and because one company can hold several, the two network tables below do not match. Parked is the crawler's verdict on what the address returned, not an inference from the name.
Resolves is the sixth: it means the name publishes at least one IP address, through an A record or an AAAA record, and 292,860,014 domains do.
Two questions get confused in every hosting ranking and this report keeps them apart. Who owns the IP address a domain points at is a registry fact. Who runs the service answering there is a different fact, and no DNS record carries it. GoDaddy's parking pages answer on addresses Amazon owns, so the network is Amazon, the provider is GoDaddy, and a ranking that reports only the first credits Amazon for a business it is not running.
Every table below therefore names the provider wherever the two can be told apart. A platform on its own address block is lifted out of its landlord's row and marked, which is why GoDaddy's parking appears as GoDaddy's rather than as a quarter of Amazon's. Two things are not hosting at all and are marked as such: a reverse proxy runs no origin server, and a parked or aftermarket name is held rather than served.
That separation is uneven, and the reason is structural rather than a gap in the work. AWS comes apart cleanly because every tenant gets its own Global Accelerator endpoints. Cloudflare does not: apart from Shopify, which has a block of its own, its largest ranges are shared anycast pools with eight million domains each, fronting origin servers that DNS cannot see. So Cloudflare leading the next table is a statement about what can be named, not about market share. If its customers could be separated the way AWS's can, its row would come apart too.
| Platform | Domains | Serve a page |
|---|---|---|
 GoDaddy parking | 22,316,430 | 1.4% |
 GoDaddy aftermarket | 12,769,245 | 0.5% |
| GoDaddy hosting | 6,969,878 | 32.0% |
| GoDaddy Website Builder | 5,975,851 | 98.3% |
 Shopify | 5,016,894 | 87.8% |
 HugeDomains | 3,490,113 | 2.0% |
 Vercel | 3,310,449 | 96.3% |
 Google Sites and Blogger | 760,988 | 57.4% |
 Weebly | 719,869 | 62.8% |
Embed this figure
Nameservers cannot find a platform. A Shopify merchant or a Vercel developer keeps the DNS their registrar gave them and points a single A record, so a nameserver lookup sees Namecheap and never sees the platform: ranking by nameserver found three Shopify domains where the address block found 5,016,894.
Address blocks can. Rank every block by how many domains point at it, fetch a sample of each and read what comes back, and the row comes apart. 73.3% of AWS is identifiable this way, 54,831,966 domains across six products, and they behave nothing alike. GoDaddy alone accounts for four of them, running the largest parking operation on the internet at 1.4% serving and a site builder at 98.3% on neighbouring ranges inside the same cloud.
3,310,449 domains are Vercel, spread over four separate ranges and serving a page on 96.3%. 3,490,113 are HugeDomains, a reseller. And 5,016,894 are Shopify, which is not in AWS at all: its block is announced by Cloudflare, so every store on a custom domain has been counted against the CDN.
No one provider holds more than an eighth of the domains that resolve
| Network | Domains | Serve a page | Not tried | IPv6 |
|---|---|---|---|---|
 Cloudflare | 37,403,719 | 80.0% | 11.2% | 81.3% |
| GoDaddy parking (on AWS) | 22,239,769 | 1.3% | 39.8% | 1.0% |
 AWS | 20,981,382 | 46.4% | 21.6% | 6.5% |
| GoDaddy aftermarket (on AWS) | 12,312,207 | 0.5% | 3.3% | 0.9% |
 IONOS | 9,818,904 | 35.8% | 30.0% | 71.6% |
 Squarespace | 8,928,111 | 93.7% | 4.4% | 0.7% |
 Hostinger | 8,268,281 | 86.5% | 7.7% | 63.1% |
 Wix | 7,294,657 | 96.0% | 2.6% | 0.7% |
 Namecheap | 7,218,426 | 58.1% | 23.9% | 1.2% |
| GoDaddy hosting (on AWS) | 6,930,788 | 31.9% | 31.6% | 0.7% |
| GoDaddy Website Builder (on AWS) | 5,947,582 | 98.3% | 1.8% | 0.7% |
 OVH | 5,503,437 | 61.3% | 15.8% | 14.1% |
| Google | 5,455,642 | 77.5% | 13.8% | 6.3% |
 Hetzner Online | 5,279,826 | 70.6% | 11.3% | 26.4% |
| Shopify (on Cloudflare) | 4,959,588 | 87.8% | 9.6% | 38.7% |
| HugeDomains (on AWS) | 3,451,643 | 2.0% | 39.0% | 98.8% |
| Vercel (on AWS) | 2,930,411 | 96.4% | 2.9% | 1.1% |
 GMO Internet | 2,892,839 | 42.3% | 27.8% | 2.2% |
 Strato | 2,644,446 | 42.7% | 27.3% | 94.1% |
 Akamai Cloud | 2,632,467 | 72.7% | 22.3% | 15.8% |
 SEDO | 2,561,383 | 16.0% | 0.2% | 1.0% |
 Alibaba | 2,548,751 | 51.7% | 26.2% | 2.0% |
 Oracle | 2,345,648 | 88.8% | 7.5% | 1.2% |
| GoDaddy | 1,952,680 | 80.3% | 9.9% | 0.9% |
 DigitalOcean | 1,860,794 | 76.4% | 12.7% | 13.4% |
| Confluence | 1,718,040 | 74.5% | 22.9% | 1.1% |
 Automattic | 1,681,133 | 86.6% | 7.5% | 0.7% |
 All-Inkl | 1,504,059 | 76.7% | 13.0% | 0.7% |
 REG.RU | 1,444,449 | 51.3% | 26.3% | 39.2% |
 Unified Layer | 1,435,168 | 73.4% | 18.6% | 1.0% |
Embed this figure
The concentration is the finding, and it does not soften as you go down the table. Ten networks hold 48.3% and thirty hold 69.0%. That is a longer tail than extensions, whose largest ten hold 72.1% of every registration, or content management systems, whose largest five hold 88.1% of the sites running one. Hosting is the least concentrated of the three and still puts 20.4% into two companies.
The second column is what keeps the first honest. Cloudflare holds the IP addresses that the most domains point at, by a distance, and gets a page out of 80.0% of them.
Which of those networks are running websites
| Network | Domains | Serve a page | Not tried | IPv6 |
|---|---|---|---|---|
| Cloudflare | 37,403,719 | 80.0% | 11.2% | 81.3% |
| GoDaddy parking (on AWS) | 22,239,769 | 1.3% | 39.8% | 1.0% |
| AWS | 20,981,382 | 46.4% | 21.6% | 6.5% |
| GoDaddy aftermarket (on AWS) | 12,312,207 | 0.5% | 3.3% | 0.9% |
| IONOS | 9,818,904 | 35.8% | 30.0% | 71.6% |
| Squarespace | 8,928,111 | 93.7% | 4.4% | 0.7% |
| Hostinger | 8,268,281 | 86.5% | 7.7% | 63.1% |
| Wix | 7,294,657 | 96.0% | 2.6% | 0.7% |
| Namecheap | 7,218,426 | 58.1% | 23.9% | 1.2% |
| GoDaddy hosting (on AWS) | 6,930,788 | 31.9% | 31.6% | 0.7% |
| GoDaddy Website Builder (on AWS) | 5,947,582 | 98.3% | 1.8% | 0.7% |
| OVH | 5,503,437 | 61.3% | 15.8% | 14.1% |
| Google | 5,455,642 | 77.5% | 13.8% | 6.3% |
| Hetzner Online | 5,279,826 | 70.6% | 11.3% | 26.4% |
| Shopify (on Cloudflare) | 4,959,588 | 87.8% | 9.6% | 38.7% |
| HugeDomains (on AWS) | 3,451,643 | 2.0% | 39.0% | 98.8% |
| Vercel (on AWS) | 2,930,411 | 96.4% | 2.9% | 1.1% |
| GMO Internet | 2,892,839 | 42.3% | 27.8% | 2.2% |
| Strato | 2,644,446 | 42.7% | 27.3% | 94.1% |
| Akamai Cloud | 2,632,467 | 72.7% | 22.3% | 15.8% |
| SEDO | 2,561,383 | 16.0% | 0.2% | 1.0% |
| Alibaba | 2,548,751 | 51.7% | 26.2% | 2.0% |
| Oracle | 2,345,648 | 88.8% | 7.5% | 1.2% |
| GoDaddy | 1,952,680 | 80.3% | 9.9% | 0.9% |
| DigitalOcean | 1,860,794 | 76.4% | 12.7% | 13.4% |
| Confluence | 1,718,040 | 74.5% | 22.9% | 1.1% |
| Automattic | 1,681,133 | 86.6% | 7.5% | 0.7% |
| All-Inkl | 1,504,059 | 76.7% | 13.0% | 0.7% |
| REG.RU | 1,444,449 | 51.3% | 26.3% | 39.2% |
| Unified Layer | 1,435,168 | 73.4% | 18.6% | 1.0% |
Embed this figure
Sort the same networks by the share of their own domains that serve a page and the table rearranges completely. Wix and Squarespace sit at the top because a domain on either is a website by construction: you do not point a name at them for any other reason. The CMS report counts the same two platforms by what runs on the page rather than by where the name points.
At the bottom sit the parking and domain-monetisation networks, where a domain resolves because somebody wants to sell it rather than use it. The domain registration report puts that population at 31,975,287 names. That is a real business and these are real IP addresses, so they belong in the table.
The domain registration report measures the same gap from the other side, over every registered name rather than every resolving one, and finds 8.0% of registrations held for sale.
The largest registrar is not in that table, and that is the finding
Most of the domains counted under AWS are not AWS customers running websites. They are names whose DNS GoDaddy answers, resolving to addresses Amazon owns.
55.6% of the domains resolving into Cloudflare use GoDaddy's nameservers: 41,548,631 of 74,794,773.
GoDaddy serves the DNS for more domains than any other company, 53,913,594 of them. Its own network holds 1,953,119, and that network behaves like a web host, serving a page on 80.3% of what is on it. The rest is answered from AWS addresses, which is why the largest cloud's row is mostly somebody else's customers.
A nameserver says who runs the DNS, never who owns the name, so this cannot separate GoDaddy's own stock from a customer's registration that was never built on, and the second is much the larger group. Nor is all of it idle: 8,926,857 of those domains served us a page, 31.4% of them. What is measurable is where they sit, and 22,219,787 of them sit on the four IP addresses where at most 1.4% answer.
The one population that is definitely for sale is a different cut: 11,461,262 domains delegate to Afternic, GoDaddy's aftermarket, which is a name being advertised rather than a name being used.
| Whose nameservers answer | Domains on AWS | Share of AWS | Serve a page |
|---|---|---|---|
| GoDaddy | 41,548,631 | 55.6% | 31.4% |
| Everyone else | 16,634,087 | 22.2% | 37.9% |
| Afternic | 11,035,636 | 14.8% | 0.5% |
| AWS Route 53 | 1,767,676 | 2.4% | 74.4% |
| Cloudflare | 1,139,192 | 1.5% | 84.3% |
| Namecheap | 747,845 | 1.0% | 88.4% |
 IBM NS1 | 602,808 | 0.8% | 95.1% |
| Google Cloud DNS | 395,393 | 0.5% | 86.4% |
| No nameserver seen | 392,820 | 0.5% | 42.4% |
| Hostinger | 262,555 | 0.4% | 93.3% |
Embed this figure
Split the whole network by whose nameservers answer and it stops being one business. GoDaddy's registrar DNS and its Afternic aftermarket hold 70.3% of AWS's domains between them, and the Afternic slice serves a page on 0.5% of 11,035,636 names, which is what a for-sale listing looks like from the outside.
What is left is the cloud business people mean when they say AWS, and it is the minority of its own row: the domains using Route 53 or sitting behind Cloudflare serve a page on three quarters and more. Shopify, worth saying because it is the obvious guess, is barely here: 3 domains out of 74,794,773. It fronts on Cloudflare, as the platform table above showed.
By comparison 7.1% of Cloudflare's domains use GoDaddy nameservers, so the effect is not peculiar to Amazon. It is what a registrar's DNS does to any network it rents from.
The busiest IP address on the internet answers almost nothing
| IP address | Whose address | What runs on it | Domain names | Serve a page |
|---|---|---|---|---|
| 3.33.130.190 | AWS | GoDaddy parking | 22,313,054 | 1.4% |
| 15.197.148.33 | AWS | GoDaddy parking | 22,288,707 | 1.4% |
| 13.248.169.48 | AWS | GoDaddy aftermarket | 12,682,811 | 0.5% |
| 76.223.54.146 | AWS | GoDaddy aftermarket | 12,682,801 | 0.5% |
| 185.230.63.107 | Wix | Wix | 7,131,754 | 97.0% |
| 198.185.159.144 | Squarespace | Squarespace | 7,093,708 | 93.9% |
| 198.185.159.145 | Squarespace | Squarespace | 7,035,783 | 93.9% |
| 198.49.23.144 | Squarespace | Squarespace | 7,028,500 | 93.9% |
| 198.49.23.145 | Squarespace | Squarespace | 7,027,241 | 93.9% |
| 3.33.251.168 | AWS | GoDaddy hosting | 6,953,538 | 32.0% |
| 15.197.225.128 | AWS | GoDaddy hosting | 6,952,459 | 31.9% |
| 185.230.63.171 | Wix | Wix | 6,651,860 | 97.1% |
| 185.230.63.186 | Wix | Wix | 6,628,656 | 97.1% |
| 13.248.243.5 | AWS | GoDaddy Website Builder | 5,954,978 | 98.3% |
| 76.223.105.230 | AWS | GoDaddy Website Builder | 5,937,997 | 98.3% |
| 188.114.97.2 | Cloudflare | Cloudflare | 5,397,265 | 81.3% |
| 188.114.96.2 | Cloudflare | Cloudflare | 5,397,250 | 81.3% |
| 13.223.25.84 | AWS | HugeDomains | 3,455,127 | 2.0% |
| 54.243.117.197 | AWS | HugeDomains | 3,455,056 | 2.0% |
| 76.223.67.189 | AWS | AWS | 3,045,368 | 4.5% |
Embed this figure
22,313,054 domain names have an A record pointing at the single IP address 3.33.130.190. 1.4% of them get a page.
The four busiest IP addresses on the internet all behave this way, and they are the same infrastructure the domain registration report traced by hand to a parking lander. The third column is the point: every one of the twenty is owned by AWS, Cloudflare, Wix or Squarespace, and 10 of them are running a product belonging to somebody else entirely. They are not web servers under strain. They are the IP address a registrar hands out to every name nobody has done anything with yet.
The busiest IP address that is genuinely serving websites is 185.230.63.107 on Wix, with 7,131,754 domains and 97.0% of them returning a page.
The same domain names appear on more than one row of that table, so the column does not sum to anything. A name that publishes three IP addresses is counted once against each of them, and the busiest addresses come in matched sets that answer for a single population of names.
The top two rows are the clearest case. 3.33.130.190 and 15.197.148.33 list 44,601,761 domains between them, but only 22,315,970 different names: very nearly the same list, published twice so that a resolver has a second address to try. Read each row as its own answer to "how many names point here", and take any total from the rest of this report, where a domain is counted once.
Half the internet's A records point at 263 IP addresses
| Domain names on the IP address | IP addresses | A records |
|---|---|---|
| 1 domain | 7,919,817 | 7,919,817 |
| 2 to 10 | 3,843,158 | 13,892,387 |
| 11 to 100 | 841,557 | 25,335,306 |
| 101 to 1,000 | 215,778 | 82,758,405 |
| 1,001 to 10,000 | 20,979 | 40,267,083 |
| 10,001 to 100,000 | 1,892 | 45,061,198 |
| over 100,000 | 263 | 251,574,720 |
Embed this figure
There are 466,808,916 A records on the internet, pointing at 12,843,444 different IPv4 addresses. 263 of those IP addresses have more than a hundred thousand domain names pointing at them each, and between them they hold 53.9% of every A record there is.
At the other end, 7,919,817 IP addresses have exactly one domain name pointing at them, 61.7% of all of them. Both facts are true at once, and the distance between them is what a shared hosting market looks like when you measure it rather than describe it.
The busier a site is, the less likely it is on a shared host
| Tranco rank | Largest | Second | Third | Fourth | Fifth |
|---|---|---|---|---|---|
| 1 to 1,000 | AWS23.4% | Cloudflare16.2% | Google8.7% |  Fastly7.6% | Akamai6.2% |
| 1,001 to 10,000 | Cloudflare28.0% | AWS20.4% | Google6.1% | Akamai4.4% | Fastly3.6% |
| 10,001 to 50,000 | Cloudflare29.9% | AWS19.3% | Fastly3.3% | Akamai3.0% | Google3.0% |
| 50,001 to 100,000 | Cloudflare37.0% | AWS15.1% | Google2.5% | Fastly2.2% |  Microsoft2.0% |
| 100,001 to 500,000 | Cloudflare36.7% | AWS12.1% | Hetzner Online2.4% | Google2.3% | Microsoft2.1% |
| 500,001 to 1 million | Cloudflare37.2% | AWS9.7% | Hetzner Online2.6% | OVH2.3% | Google2.1% |
| 1 to 2 million | Cloudflare32.2% | AWS8.6% | Hetzner Online2.9% | OVH2.8% | Google2.1% |
| Over 2 million | Cloudflare26.7% | AWS12.4% | Hetzner Online4.2% | Google2.8% | Squarespace2.8% |
Embed this figure
That much is expected, and worth saying plainly rather than dressing up: a site with real traffic outgrows a shared machine and can afford not to be on one. The part that is not obvious is who it moves to, because the answer changes at every level of the list rather than converging on one or two names.
AWS leads exactly one band, the top thousand. Cloudflare takes the lead at rank 1,001 and never gives it back, and its share is not flat across the rest: it climbs from 28.0% to a peak of 37.2% around rank one million, then falls away again below that. Both ends of the list are less concentrated behind it than the middle is, which is a shape worth noticing rather than a rule about what large sites do.
The two enterprise CDNs live only at the top. Akamai and Fastly are 13.8% of the top thousand between them, and both are out of the top five below rank one hundred thousand. Hetzner and OVH do the opposite, neither appearing at all until the hundred thousand mark has passed, which is what budget European hosting looks like when you sort by traffic rather than by count.
Read the Largest column against the report's headline and the difference is the whole point of this section. Counting every domain, no provider holds more than an eighth. Counting the thousand busiest sites, AWS holds a quarter of them, and around the millionth rank Cloudflare is holding close to four domains in ten.
Company size predicts the tail, not the top
| Employees | Largest | Second | Third | Fourth | Fifth |
|---|---|---|---|---|---|
| 1 person | Cloudflare15.4% | AWS14.3% | Squarespace9.7% | Wix8.6% | Google4.0% |
| 2 to 10 | Cloudflare17.9% | AWS16.2% | Wix6.3% | Squarespace5.1% | Hostinger4.8% |
| 11 to 50 | Cloudflare18.8% | AWS13.7% | Hostinger5.4% | Google4.3% | Wix4.3% |
| 51 to 200 | Cloudflare19.4% | AWS12.0% | Hostinger5.2% | Google4.1% | Wix2.9% |
| 201 to 500 | Cloudflare18.8% | AWS12.1% | Hostinger5.0% | Google3.6% | GoDaddy2.5% |
| 501 to 1,000 | Cloudflare19.2% | AWS12.1% | Hostinger4.0% | Google3.3% | Microsoft2.7% |
| 1,001 to 5,000 | Cloudflare18.8% | AWS12.6% | Microsoft4.0% | Google3.0% | Hostinger2.7% |
| Over 5,000 | Cloudflare18.0% | AWS13.9% | Microsoft4.5% | Hostinger3.1% | Google2.8% |
Embed this figure
Matching domains to the headcount their company publishes on LinkedIn gives a much flatter picture than rank does. Cloudflare is first in all eight bands and never moves outside 15.4% to 19.4% of any of them, and AWS is second in all eight. The top of this table is the same whether a company is one person or five thousand, so size predicts almost nothing about who is first.
It predicts a great deal about who is third. Squarespace and Wix are 18.3% of the one-person band between them and are the third and fourth largest networks in it. Squarespace has left the top five by eleven employees and Wix by two hundred, which is the clearest measure available of what a site builder is for. Hostinger takes their place and is in the top five of every band except the one-person one. Microsoft appears nowhere until five hundred employees and then rises through every band above that, which is the same enterprise pattern that keeps Azure out of the headline table.
This table covers only domains whose company has a LinkedIn record carrying both a website and a headcount, so it is a sample of the web rather than the whole of it, and the methodology says how large.
IPv6 is mostly decided by your hosting provider
| Network | Domains | Have IPv6 |
|---|---|---|
| Strato | 2,644,457 | 94.1% |
| Cloudflare | 42,376,874 | 76.3% |
| IONOS | 9,819,117 | 71.6% |
 One.com | 1,334,346 | 64.7% |
| Hostinger | 8,269,187 | 63.1% |
 Host Europe | 1,386,748 | 48.0% |
| REG.RU | 1,444,457 | 39.2% |
 Aruba | 1,372,745 | 31.1% |
| Hetzner Online | 5,280,154 | 26.4% |
| Akamai Cloud | 2,637,701 | 15.8% |
| Unified Layer | 1,435,227 | 1.0% |
| SEDO | 2,565,194 | 1.0% |
 Newfold Digital | 1,399,770 | 1.0% |
| GoDaddy | 1,953,119 | 0.9% |
 Network Solutions | 1,093,206 | 0.8% |
| Automattic | 1,681,243 | 0.7% |
| All-Inkl | 1,504,093 | 0.7% |
| Squarespace | 8,932,563 | 0.7% |
| Wix | 7,295,092 | 0.7% |
 XServer Cloud | 1,056,353 | 0.7% |
Embed this figure
IPv6 reaches 23.8% of the domains that resolve. The interesting number is not that one, it is the spread behind it.
Among networks holding a million domains or more, IPv6 runs from 94.1% on Strato down to 0.7% on XServer Cloud. That is 134.4 to one between networks selling the same product to the same market.
Nothing about the site explains this. A small business site on one German host has IPv6 and the same site on a large American builder does not, and neither owner chose. It is the same mechanism the AI crawler report finds in robots.txt, where a provider's default decides for millions of sites that never opened the file. 7 networks in that group are below one per cent, and they hold 23,515,669 domains between them.
| Extension | Resolving domains | Have IPv6 |
|---|---|---|
| .nl | 3,075,879 | 54.9% |
| .pro | 1,043,891 | 45.6% |
| .ch | 2,123,112 | 42.2% |
| .de | 14,500,631 | 40.0% |
| .site | 1,677,156 | 39.7% |
| .store | 2,013,234 | 38.5% |
| .se | 1,196,832 | 35.5% |
| .fr | 2,286,559 | 33.0% |
| .eu | 1,911,930 | 32.2% |
| .info | 4,413,661 | 31.2% |
| .shop | 3,581,756 | 30.0% |
| .online | 3,228,290 | 29.6% |
| .app | 1,363,127 | 29.3% |
| .br | 1,896,983 | 29.1% |
| .es | 1,050,192 | 27.5% |
| .it | 1,970,359 | 26.5% |
| .top | 3,828,411 | 24.8% |
| .uk | 4,990,117 | 24.7% |
| .in | 1,192,922 | 23.6% |
| .com | 150,171,713 | 21.1% |
Embed this figure
By extension, .nl leads at 54.9%. Half the extensions above the average are European country codes and most of the rest are the cheap generic ones, which sit on the same few hosts. Both halves say the same thing: the extension is standing in for who hosts it.
6.0% of domains point at a server that is not there any more
| What the published IP address did | Domain names |
|---|---|
| Answered nothing before the deadline | 15,449,376 |
| Refused the connection | 1,384,497 |
| No route to the address | 378,935 |
| Closed the connection with no reply | 230,671 |
| Stopped resolving between the two passes | 4,403,790 |
Embed this figure
Every domain in this section works as far as DNS is concerned. Ask for the name and you get an IP address back, which is the whole job of a domain name. The address is where it fails: we opened a connection to it and found nothing listening. The name resolves and the website does not load, and from the outside those two things look identical until something tries to connect.
This is what a stale A record is, a line of DNS still giving directions to a machine that has moved or been switched off. Nobody has to do anything wrong for it to happen. A host is cancelled, a server is retired, a company folds, and the record goes on pointing at the address it was given.
Two different failures sit underneath the one figure. 15,449,376 domains publish an IP address that never answered us at all, and 1,384,497 publish one that actively refused the connection.
The second group is the more interesting one. A timeout can be a firewall, a slow host or our own network on a bad day. A refused connection is an answer: something is at that IP address, it is reachable, and it is not serving this domain.
A further 4,403,790 domains had an IP address when we scanned DNS and had stopped resolving by the time we fetched them, which is the same decay caught in the few weeks between the two passes.
Most domains have one IP address and no spare
| IPv4 addresses published | Domain names |
|---|---|
| 1 IP address | 166,409,314 |
| 2 IP addresses | 101,998,092 |
| 3 IP addresses | 10,330,542 |
| 4 IP addresses | 10,385,472 |
| 5 or more | 1,963,117 |
Embed this figure
57.2% of domains with an A record publish exactly one IPv4 address. If that address stops answering, there is nowhere else for a resolver to send the browser.
| How far the IP addresses spread | Domain names |
|---|---|
| More than one IP address | 124,677,223 |
| More than one company | 2,312,651 |
| More than one network number | 2,877,041 |
| More than one country | 2,318,802 |
Embed this figure
Publishing several IP addresses is common enough, but they almost always belong to the same host. Only 0.8% of domains publish IP addresses belonging to more than one company, and 0.8% publish addresses in more than one country.
The chart's third row is a looser test than the second and that is why it is larger. A company can hold several numbered pieces of network, so a domain with one address in AS16509 and another in AS14618 counts there while still sitting entirely inside Amazon. 1.0% of domains are split that way, and most of them have one provider.
This is the finding most likely to be misread, so it is worth being exact in two directions.
It does not say the web has no redundancy. It says redundancy has moved inside the network: Cloudflare answers on the same IP address from every city it operates in, so a domain behind it has one A record and hundreds of physical locations. Multi-homing at the domain level, the thing DNS was designed to express, is now a rounding error.
It is also a measurement of the bare domain name and nothing below it. Everything in this report is read at the apex, so this figure asks whether example.com publishes addresses in more than one network. It does not ask whether www.example.com points somewhere different, and a site split that way would not appear here. What the data can say about names below the apex is a different finding, and a large one: 5,088,501 domains publish no address at all on the bare name while www does answer, 11.4% of every domain that answers DNS with nothing to connect to. Those sites work, at one address of the two people type.
Three in five addresses are American, which is not three in five websites
| Country | Domains | Serve a page |
|---|---|---|
| USA | 171,264,097 | 58.1% |
| Germany | 27,885,912 | 52.2% |
| Hong Kong | 10,269,839 | 52.0% |
| Canada | 9,240,366 | 69.7% |
| France | 8,560,556 | 70.6% |
| Netherlands | 6,548,835 | 69.1% |
| Japan | 6,372,880 | 66.3% |
| Russia | 5,382,131 | 70.8% |
| UK | 4,331,371 | 75.8% |
| China | 2,715,893 | 52.7% |
| Italy | 2,684,673 | 78.2% |
| Poland | 2,302,631 | 86.6% |
| Singapore | 2,288,732 | 80.9% |
| Cyprus | 2,246,727 | 69.9% |
| Australia | 2,134,434 | 75.4% |
Embed this figure
59.1% of the domains whose IP address we can place have it registered in the USA, and 9.6% in Germany. No third country reaches four per cent.
That first figure is the one most likely to be repeated as "three in five websites are hosted in America", and it is not that. Holding the IP address is not hosting the website, and in this table two separate populations make the gap large enough to matter.
| Country | Domains | Not proxied or parked | Largest | Second | Third |
|---|---|---|---|---|---|
| USA | 171,264,097 | 93,089,531 | Cloudflare20.9% | GoDaddy parking (on AWS)13.0% | AWS9.8% |
| Germany | 27,885,912 | 25,237,254 | IONOS25.2% | Hetzner Online15.6% | Strato9.5% |
| Hong Kong | 10,269,839 | 10,225,516 | Power Line Datacenter10.5% | Alibaba10.5% | DXTL6.3% |
| Canada | 9,240,366 | 7,906,782 | Shopify (on Cloudflare)53.7% |  Team Internet14.2% | OVH11.1% |
| France | 8,560,556 | 8,558,577 | OVH44.0% | Hostinger10.6% | Host Europe10.2% |
| Netherlands | 6,548,835 | 6,535,977 |  Signet12.0% | Google9.4% |  Your Hosting7.1% |
| Japan | 6,372,880 | 6,160,935 | GMO Internet45.5% | XServer Cloud16.6% |  SAKURA Internet11.5% |
| Russia | 5,382,131 | 5,376,677 | REG.RU26.9% |  Beget12.9% |  DDoS-Guard10.4% |
| UK | 4,331,371 | 4,325,470 | IONOS13.9% |  20i10.6% |  Heart Internet5.6% |
| China | 2,715,893 | 2,715,666 | Alibaba45.7% | China Telecom17.1% |  Tencent12.4% |
| Italy | 2,684,673 | 2,684,605 | Aruba51.1% |  Register20.0% |  Server Plan S.r.l.5.7% |
| Poland | 2,302,631 | 1,475,901 | Cloudflare36.0% |  Cyber_Folks10.7% |  home.pl9.1% |
| Singapore | 2,288,732 | 2,214,175 | AWS13.7% | GoDaddy13.4% | Hostinger11.3% |
| Cyprus | 2,246,727 | 2,233,965 | Hostinger99.2% | Cloudflare0.6% |  Cyta0.1% |
| Australia | 2,134,434 | 1,714,502 |  Dreamscape Networks22.3% | Cloudflare19.7% |  Synergy Wholesale18.4% |
Embed this figure
The first is the reverse proxy. A domain behind one is filed to whichever country the proxy registered the address block in, while the machine serving the page can be anywhere. 37,064,002 American domains are in that state, 21.6% of the country's total, and nothing in DNS says where they are really served from.
The second is names that are not websites. Parked inventory and aftermarket listings resolve into whichever country their parking company's block is registered in. 41,110,564 of the American total are those.
Take both out and 93,089,531 domains are left, 54.4% of what the country chart credits to the United States. That is the closest this data comes to a count of sites actually hosted there, and it is the third column above. Across all fifteen countries the same subtraction leaves 180,455,533.
The subtraction is not even. Germany loses almost nothing, because Germany is IONOS, Hetzner and Strato running their own machines for their own customers. The United States loses almost half. Canada is the clearest case of all: over half of its domains are Shopify, whose address block the registry files to Canada while the stores on it sell from everywhere.
Two countries in the table are effectively one company each. Cyprus is Hostinger, at 99.2% of everything registered there, and Italy is Aruba at 51.1%. Neither is a statement about where those countries' websites live. It is a statement about where one hosting company chose to register its addresses.
Read the whole table as where the address block is registered rather than where the packet is answered. An anycast address belongs to one country in the registry and answers from fifty.
Who runs the mail
Two sections on a different layer follow, because the same DNS scan that says where a domain is hosted also says who handles its mail and who answers for its name, and the concentration looks nothing alike.
| Mail provider | Domains |
|---|---|
| Google Workspace | 20,785,465 |
| Microsoft 365 | 14,168,244 |
| GoDaddy | 9,420,686 |
| Namecheap | 7,910,718 |
| IONOS | 7,736,402 |
| Hostinger | 3,570,810 |
| Strato | 3,493,451 |
| OVH | 2,756,400 |
| Cloudflare | 2,056,577 |
 Zoho | 1,888,810 |
| hostedemail.com | 1,826,961 |
 Above.com parking | 1,506,790 |
| All-Inkl | 1,439,004 |
 jellyfish.systems | 1,264,203 |
 SiteGround | 1,196,490 |
 Proofpoint | 969,650 |
 united-domains | 912,629 |
 mailgun.org | 907,391 |
| one.com | 872,090 |
 qq.com | 822,788 |
Embed this figure
172,100,073 domains publish an MX record. Google Workspace takes 12.1% and Microsoft 365 8.2%, and ten providers carry 42.9% between them.
Those shares are lower than a mail market share usually reads, because this counts every domain with an MX record rather than every mailbox. Most domains with mail configured have no mailboxes behind it, only a registrar's default.
14,997,760 domains do the opposite and publish a null MX, a single record whose entire meaning is that this domain accepts no mail. It is 8.7% of everything with an MX, or 5.1% of every domain that resolves. The mail authentication report counts a slightly larger MX population from its own pipeline, so the two totals differ by about a million. The mail authentication report covers what the rest of them do about spoofing, and finds that most domains publishing mail records never set a policy behind them.
Who runs the DNS
| DNS provider | Domains |
|---|---|
| GoDaddy | 53,913,594 |
| Cloudflare | 44,985,559 |
| Afternic parking | 11,401,391 |
| Google Cloud DNS | 10,879,545 |
| Namecheap | 10,730,611 |
| IONOS | 8,464,806 |
| Hostinger | 8,302,950 |
| Wix | 6,789,008 |
| Alibaba | 5,408,938 |
| AWS Route 53 | 4,102,948 |
| share-dns.net | 3,906,521 |
| Alibaba Cloud DNS | 3,900,483 |
| Strato | 3,649,659 |
 NameBright | 3,537,272 |
| NS1 | 3,328,033 |
| OVH | 3,074,435 |
 Spaceship | 2,953,016 |
| Squarespace | 2,867,019 |
| REG.RU | 2,537,038 |
| julydns.com | 2,256,703 |
Embed this figure
GoDaddy serves the DNS for 53,913,594 domains and Cloudflare for 44,985,559. Ten companies serve it for 49.9% of every domain that publishes a nameserver, which sits between hosting's 48.3% and mail's 42.9%.
The top of it is tighter than that tail suggests, two companies serving nearly a third of every domain with a nameserver. DNS is the part nobody moves. A site changes host, changes builder, changes mail provider, and keeps the nameservers its registrar set on the day the name was bought.
One company can be several networks
| Network | ASN | Domains | Serve a page |
|---|---|---|---|
| AWS | AS16509 | 67,324,032 | 30.6% |
| Cloudflare | AS13335 | 40,235,460 | 80.6% |
| IONOS | AS8560 | 9,818,046 | 35.8% |
| Squarespace | AS53831 | 8,932,554 | 93.7% |
| Hostinger | AS47583 | 8,268,409 | 86.5% |
| AWS | AS14618 | 7,465,673 | 37.5% |
| Wix | AS58182 | 7,295,058 | 96.0% |
| Namecheap | AS22612 | 7,218,691 | 58.1% |
| OVH | AS16276 | 5,504,369 | 61.3% |
| Hetzner Online | AS24940 | 4,896,727 | 73.9% |
| Google | AS396982 | 3,324,133 | 74.6% |
| GMO Internet | AS7506 | 2,749,563 | 35.6% |
| Strato | AS6724 | 2,644,457 | 42.7% |
| Akamai Cloud | AS63949 | 2,637,736 | 72.5% |
| SEDO | AS47846 | 2,565,194 | 16.0% |
| Oracle | AS31898 | 2,344,658 | 88.8% |
| Cloudflare | AS209242 | 2,140,969 | 86.7% |
| DigitalOcean | AS14061 | 1,861,672 | 76.4% |
| Confluence | AS40034 | 1,718,903 | 74.4% |
| Automattic | AS2635 | 1,681,243 | 86.6% |
Embed this figure
The network table near the top of this report counts by company. This one counts the numbered pieces those companies announce to the rest of the internet, which is the unit most published network data uses.
The two do not agree, and that is the point. Amazon holds two of the twenty largest pieces, AS16509 and AS14618. Cloudflare holds two more, AS13335 and AS209242. Counted this way each of them is split in half, and a reader who took this table for a market share would put both companies lower than they are.
Most hosting rankings are built on one of these two units without saying which. The gap between these two tables is how much consolidation the numbered view hides.
Frequently asked questions
Who is the largest web host in the world? Not the largest row in this report's network table. 37,403,719 domains resolve to Cloudflare, which is a reverse proxy rather than a host: it runs no origin server, and whatever is behind it is hosted somewhere DNS cannot see. The largest network actually running the sites is AWS with 20,981,382 domains at 46.4% serving, ahead of IONOS at 9,818,904.
Why does GoDaddy look small in the hosting table? Because the names it answers DNS for resolve to AWS IP addresses. 41,548,631 domains on GoDaddy nameservers sit inside the AWS row, 55.6% of it, and 22,219,787 of those are on the four addresses where at most 1.4% answer. GoDaddy's own network holds 1,953,119 domains and serves a page on 80.3% of them.
What percentage of websites support IPv6 in 2026? 23.8% of domains that resolve publish an AAAA record, 69,736,003 of them. Almost all of those also publish an IPv4 address; only 1,773,477 domains are reachable over IPv6 alone.
How many domain names share one IP address? It depends entirely which IP address. 7,919,817 IPv4 addresses have exactly one domain name pointing at them, while 263 have more than a hundred thousand each. The busiest has 22,313,054.
Where are most websites hosted? 59.1% of the domains whose IP address we can place publish one registered in the USA, ahead of Germany at 9.6%. Holding the address is not hosting the site, though: 37,064,002 of the American total are behind a reverse proxy serving from an origin DNS cannot see, and 41,110,564 are parked or aftermarket names rather than websites. Removing both leaves 93,089,531 domains genuinely hosted on American addresses. The largest holder of them is Cloudflare.
What is a stale A record? An A record that still points at an IP address after whatever was there has gone. 6.0% of resolving domains are in that state, either timing out or refusing the connection.
Do most websites use more than one hosting provider?
No. 0.8% publish IP addresses belonging to more than one company, 2,312,651 domains. Redundancy today is built inside a single provider's anycast network rather than across two providers. That counts the addresses a registered name publishes for itself; a site serving www from a different provider than the bare name is not counted, and this data cannot say how common that is.
Who provides DNS for the most domains? GoDaddy, serving the DNS for 53,913,594 domains, ahead of Cloudflare at 44,985,559.
How many domains answer DNS but publish no IP address at all? 44,651,345 domains answer DNS without publishing an A or AAAA record, 13.2% of everything that answers DNS. They have nameservers and usually an MX or a TXT record, and nothing to connect to.
Methodology and sources
The population. Every figure here is a share of the 292,860,014 domain names that publish at least one IP address, through an A record or an AAAA record. That is a subset of the 399,851,676 registered domains the domain registration report counts, and the two measurements are independent: this one comes from DNS scanning on the crawl infrastructure, that one from the registration and resolution pipeline, and they agree on the resolving population to within 0.013%.
What counts as a domain name. A registration at the apex, meaning two labels, or three where the middle one is an ICANN second level such as co.uk. Under a wildcard suffix such as *.np every second level is itself a suffix, so registrations there are three labels. Hostnames below the apex are not registrations and are excluded.
A domain's IP addresses are taken at its latest scan. DNS was read between 15 June and 31 August 2026. 10,655,480 domains were scanned more than once in that window, and 41.6% of those changed how many IP addresses they published between one scan and the next, so counting every record ever seen would report a move from one host to another as two addresses held at once. Each domain's set is taken at its own most recent scan.
An A record shows the edge, never the origin. This is the hardest limit on everything above. 22,239,769 domains resolve to Cloudflare, and what sits behind Cloudflare is invisible to DNS: the origin may be Google Cloud, Azure, a rack in an office, anything. A site behind any CDN is credited here to the CDN and not to whoever runs the server.
Google's row is not Google Cloud. The address registry files every network Google runs against the same domain, so that row mixes the cloud anyone can buy with the infrastructure Google serves its own products from. Google Cloud Platform proper is one autonomous system inside it, AS396982, holding 3,324,117 domains and serving a page on 74.6% of them: 60.8% of the row. The remaining 2,141,163 are Google's own networks, which is where Blogger and Sites answer from. The same caution applies to every other row here: the unit is the company that holds the addresses, not the product it sells.
That is why the two largest enterprise clouds look small in these tables. Google's networks hold 5,465,280 domains and Microsoft's 1,252,352, against AWS's 74,794,773, and the gap is not the whole story: AWS's figure is 55.6% domains on GoDaddy's nameservers, while Google and Microsoft serve a page on 77.5% and 72.9% of theirs. The clearest measure of what DNS cannot see is in this report's own mail table: Microsoft 365 is the MX for 14,168,244 domains, 11.3 times the number that resolve into Microsoft's network. Read every network figure here as "the addresses domains point at", not as cloud market share.
Some nameserver rows carry a hostname rather than a company. Where the zone's SOA record names who is responsible, the row uses the company: worldnic.com answers as Network Solutions, namebrightdns.com as NameBright, spaceship.net as Spaceship. Three of the largest name nobody. share-dns.net points at itself, dyna-ns.net points at each customer's own hostmaster, and julydns.com's sample is an unrelated mix, so those rows keep the nameserver domain. Naming them would be guessing a brand off a hostname, which is how two wrong labels got into an earlier draft of this report.
The popularity bands are Tranco, a 4,783,135-row list taken on the crawl infrastructure. Each band excludes the ones above it, so "rank 1,001 to 10,000" is nine thousand sites and not ten. The sister reports use a list of their own taken on a different day, so a rank band there will not hold exactly the same sites.
The hosting column subtracts two named populations. Beside each country's total sits the count left after two groups are removed: domains whose address belongs to a reverse proxy with no platform identified behind it, and domains on a parking or aftermarket platform. A platform found on a proxy's own block stays in, because Shopify running stores on a Cloudflare-announced range is hosting them. The subtraction is a floor on what is locally hosted, not a precise count, because a proxy can also front an origin in the same country.
IP addresses are placed by registry data, not by measurement. Country, autonomous system and network owner come from the registration data for the address block. Every IPv4 address any domain publishes is covered bar four, so the network and country tables have no gap on the address side, but an anycast address is placed where it is registered and not where it answers.
Networks are grouped by their own domain, not their filed company name. Cloudflare files under two names and Amazon operates two of the twenty largest autonomous systems, so a raw ranking of either splits a company in half.
Mail and DNS providers are ranked by the registrable domain of the target, not from a provider list. Our own provider roster names twenty mail companies covering 60.4 million of the domains with an MX, so ranking from it would publish a market share with two thirds of the market missing. Where one company answers on several domains the fold happens before the distinct count and never by adding rows: IONOS answers NS on four ui-dns domains for the same 8,464,946 domains, and summing those rows would report 33.8 million.
A nameserver is not ownership. Where this report says domains use a company's nameservers, that is what it measured: which DNS servers answer for the name. It does not show who registered the name, who owns it now, or whether it is for sale, and no DNS record carries that. The only for-sale figure here is the count delegating to an aftermarket's own nameservers.
A serving rate is over the domains that carry a verdict. 50,649,094 resolving domains are recorded as failed with no failure reason at all: empty error, empty attempt history, zero response time. Sixty of them drawn at random and resolved live gave 37 serving a page and 45 answering something, so the group is a gap in the crawl rather than dead infrastructure. It is also uneven, 2.6% of one network's domains against 30.0% of another's, so leaving it in the denominator would not shift every network by the same amount, it would reorder them. Every serving share here is therefore a share of the 242,210,920 domains that were tried and gave an answer either way.
Company size covers a sample, not the web. The headcount table matches a domain to the LinkedIn record of the company that publishes it as its website, so it reaches only the domains that have such a record with both a website and a headcount on it. Read those rows as the shape within that sample.
Serving means the crawler got a page. Failure reasons come from the fetch, so a domain counted as publishing an IP address that answers nothing is one whose address was tried and did not work. Domains with no IP address to try at all are a separate state and are not in that figure.
The apex is the unit, and www is not measured except once. Every figure here describes a registered name itself, not hostnames under it, so a domain whose www resolves differently from the bare name is counted once at the bare name. The one place www appears is the count of domains that answer only there, and that count is a floor rather than a rate: the www lookups in this data were made as a fallback when the apex failed to resolve, so 98.8% of them belong to an apex with no address. Domains where both resolve are 1.2% of that set and are in it by accident of the crawl, which is why no "how often does www differ" figure is published. Answering that properly needs a www lookup for every domain, not only for the ones that failed.
What is not measured. Response time per provider was measured and dropped: it is recorded for 4.09% of the domains that served a page, and whatever decided that coverage could be related to what is being measured. Reverse DNS, hosting plan, server software and the physical location of an anycast address are all outside what DNS and address registry data can answer.
Sources. Address ownership, autonomous system and country from IPinfo, which updates daily and places an anycast address where its block is registered rather than where it answers. Second-level suffixes from the Public Suffix List, ICANN section. Null MX is defined in RFC 7505. Some country data elsewhere on this site is derived with MaxMind GeoLite2.