dns.pizza research

One outage from darkness: 90.6% of the web's top domains run DNS on a single provider

We resolved the nameservers of 99,262 domains in the Tranco top 100,000 to the companies that actually operate them. Only 9.4% use two or more independent DNS operators. Two companies answer for 51.2–51.4% of the total — and counting nameserver hostnames instead of owners hides the problem entirely.

Published October 15, 2026 · data collected July 2026–October 2026 · download the data

Want the names? For each major operator, a standing page lists which sites would go dark if it failed and which also list a second operator.

90.6%

run every nameserver at one operator — no DNS redundancy

32.8%

Cloudflare — the largest DNS operator

51.2–51.4%

Cloudflare and Amazon Route 53 combined

20,831

domains that look redundant by nameserver name but sit at one company

Key findings

  1. Nine in ten of the web's most popular domains have no DNS redundancy. 89,904 of 99,262 domains (90.6%) run every one of their nameservers at a single operator. If that operator's control plane fails, there is no second authoritative source to answer from. Only 9,358 domains (9.4%) split their nameservers across two or more independent companies. Because our method counts two zones owned by the same unlisted company as two operators (see limitations), 90.6% is a floor.
  2. Counting nameserver names instead of owners hides the concentration completely. By nameserver zone, only 69,073 domains (69.6%) look single-homed. Resolve those zones to the companies that own them and the number jumps to 89,904 — 20,831 more domains, 21% of the corpus. The reason is that the biggest operators spread customers across many zone names: amazon.com's four nameservers sit in awsdns-52.org, awsdns-21.co.uk, awsdns-01.net and awsdns-33.com — four domains, four TLDs, one company. microsoft.com does the same across four azure-dns.* zones. Any survey that ranks raw nameserver domains makes Route 53 disappear from the top of the table.
  3. Two companies answer for half the web's DNS. Cloudflare serves 32.8% of the corpus and Amazon Route 53 18.6% — 51.2–51.4% between them. Then the curve falls off a cliff: Akamai 2.9%, GoDaddy 2.1%, Google 2.1%, Azure DNS 1.5%. No one else reaches 2%. Behind them sits a very long tail: 14,794 distinct operators too small for a row of their own — between them the sole DNS of 23,103 domains (23.3% of the corpus) — most of them registrars and hosting companies running DNS as a side effect of selling something else.
  4. The biggest sites hedge — and they buy differently. 17.1% of the top 1,000 run two or more DNS operators, against 9.1% in the 10,001–100,000 band. The top 1,000 is also the one tier Cloudflare does not lead: Route 53 serves 23.1% of it against Cloudflare's 11.7%, with Akamai at 9.9% and NS1 at 5.8% — enterprise DNS vendors sold on exactly the multi-provider resilience the rest of the web skips. Cloudflare's share runs the other way, from 11.7% of the top 1,000 to 33.6% of the mid-tail. NS1 is the clearest tell: it is 1.2% of the corpus overall but 5.8% of the top 1,000, and most of its domains are multi-provider setups rather than sole deployments (462 sole of 1,186 total).
  5. Redundancy is a regional habit, not a global one. The TLDs that actually run multi-provider DNS are northern and central European: .fi (34.6%), .cz (34.4%), .de (34.2%), .se (33.1%). The ones that don't are the TLDs the modern web was built on: .live (0.4%), .app (1%), .xyz (1.5%). .de runs 34.2% multi-provider — more than thirty times .app's 1%. Some of that gap is real operational culture (two nameservers at two hosts is a long-standing convention among German, Dutch and Nordic hosters) and some is an artifact of how we attribute small operators, which the limitations below spell out. Leadership is far less varied than the mail map: Cloudflare is the largest operator in 46 of the 57 TLD rows and Route 53 in 9, leaving only 2 where a national provider comes first.
  6. 203 domains publish just one nameserver. RFC 1034 has asked for at least two authoritative nameservers since 1987, and 203 domains in the top 100,000 still publish exactly one hostname — a single name to fail, never mind a single company. The modal setup is the bare minimum: 52% publish exactly two. The biggest TLD is no exception: .com, 46,294 domains, runs 93.5% single-provider with Cloudflare alone at 36%.

Who runs it, by popularity tier

Every domain in each tier, split by the operator running ALL of its nameservers. The 2 largest operators are named; "another single operator" is every other sole-operator domain, and the last band is domains spanning two or more companies.

Source: dns.pizza nameserver records · snapshot October 1, 2026

Largest DNS operators by footprint

Share of the 99,262 measured domains using each operator for at least one nameserver. Multi-provider domains count for every operator they use, so these overlap and do not sum to 100%.

Source: dns.pizza nameserver records · snapshot October 1, 2026

Every operator with enough domains for a row

Every DNS operator serving at least 200 domains in the corpus. Sole counts domains where this operator runs every nameserver; footprint counts domains using it for at least one, so footprints overlap across operators. 14,794 smaller operators fold into "everyone else". Rows named as a bare domain are operators we did not map to a company name — the nameserver zone is reported as-is rather than guessed at.

OperatorSoleFootprintShareTop 1,00010,001 – 100,000
Cloudflare32,35532,53132.8%11.7%33.6%
Amazon Route 5318,24918,46018.6%23.1%18%
Akamai2,4592,8832.9%9.9%2.5%
GoDaddy2,0442,0782.1%0.4%2.2%
Google1,9772,0692.1%7.2%1.8%
Microsoft Azure DNS1,4571,5201.5%4.1%1.5%
Alibaba Cloud DNS1,3731,3871.4%1.3%1.4%
NS1 (IBM)4621,1861.2%5.8%1.1%
Tencent DNSPod9429651%1.7%0.9%
UltraDNS (Vercara)5819050.9%1.6%0.8%
RU-CENTER3787450.8%0.9%0.8%
Namecheap6496580.7%0.1%0.7%
CSC6016310.6%0.2%0.7%
DNS Made Easy4144840.5%0.5%0.5%
ClouDNS3224670.5%0.4%0.5%
Yandex3673750.4%0.7%0.4%
Gandi3173360.3%0.3%0.3%
Selectel2983030.3%0.3%0.3%
REG.RU2762800.3%0.4%0.3%
OVHcloud2522640.3%0.1%0.3%
pbsg.com2332330.2%0%0.3%
DigitalOcean2282320.2%0%0.3%
Network Solutions2262280.2%0%0.3%
Oracle (Dyn / OCI)1572180.2%0.4%0.2%
DNSimple1842010.2%0.4%0.2%
Everyone else23,103————

How many nameservers, how many companies

The same 99,262 domains counted two ways: by distinct nameserver hostname, and by the number of independent companies those hostnames resolve to. The gap between the two columns is the whole finding.

Nameserver hostsDomainsShare
1 nameserver2030.2%
2 nameservers51,63552%
3 nameservers7,8717.9%
4 nameservers30,21630.4%
5 or more9,3379.4%
Independent operatorsDomainsShare
1 operator89,90490.6%
2 operators6,5176.6%
3 or more2,8412.9%

By popularity tier and TLD

Each segment's domains, the share running a single operator, and the footprint of the three largest operators. All 57 TLDs with at least 200 measured domains are listed, largest sample first. Operator columns are footprints and overlap, so they do not sum to 100%.

SegmentDomainsOne operator2+ operatorsCloudflareRoute 53Akamai
All domains99,26290.6%9.4%32.8%18.6%2.9%
Top 1,00099682.9%17.1%11.7%23.1%9.9%
1,001 – 10,0009,24988.5%11.5%26.9%24.1%5.7%
10,001 – 100,00089,01790.9%9.1%33.6%18%2.5%
.com46,29493.5%6.5%36%23.3%3.3%
.net6,45291%9%28.5%17.5%2.4%
.org4,15093.9%6.1%41.8%14.1%1.1%
.ru3,80680.3%19.7%12%0.9%0%
.io1,91897.4%2.6%40.3%36.6%0.6%
.jp1,78281%19%5.8%36%7.3%
.de1,78265.8%34.2%14.8%9.7%4%
.br1,38394.7%5.3%30.2%22.4%5.4%
.uk1,33784.9%15.1%25.5%21.7%3.4%
.in1,16495.4%4.6%27.6%12.5%2.5%
.fr99474.7%25.3%18.6%9.9%3.2%
.cn94787.5%12.5%1.3%7.5%0.4%
.it86175.4%24.6%20.4%16%4.9%
.co84197.9%2.1%40.8%28.5%0.6%
.edu82575.9%24.1%10.1%10.5%3.2%
.pl76090%10%32.5%4.2%3.7%
.nl64568.2%31.8%21.6%10.7%10.7%
.es55379.7%20.3%21%15.4%6.5%
.au52890.9%9.1%19.1%23.3%8.3%
.tv52493.5%6.5%44.1%19.1%1.9%
.ca46082.8%17.2%20.2%14.8%10.2%
.me44094.3%5.7%49.8%18%0.7%
.gov43992%8%26%8%16.6%
.ai42498.1%1.9%49.1%28.1%1.2%
.ir42194.3%5.7%7.6%0%0%
.info40193.3%6.7%49.4%6%0.2%
.app40199%1%54.9%19.2%0.7%
.id39791.2%8.8%39%7.6%0.5%
.cz39365.6%34.4%30%3.8%1.5%
.ua37683.8%16.2%59.3%4.3%2.4%
.cloud35993.6%6.4%28.1%37.9%1.1%
.eu34581.2%18.8%27.8%16.8%3.2%
.xyz34498.5%1.5%50%16.3%0%
.tr34194.7%5.3%26.4%8.5%2.3%
.cc33196.1%3.9%55.9%10.9%0%
.biz31694.3%5.7%18.4%24.4%0.3%
.se31166.9%33.1%16.4%10.3%2.3%
.gr31089.4%10.6%49.4%5.2%4.5%
.tw30994.8%5.2%23%12%2.9%
.vn29884.6%15.4%33.9%6%0.3%
.mx28992%8%22.8%15.2%6.9%
.ro28789.2%10.8%47.4%7.3%2.4%
.kr28287.2%12.8%8.5%16.7%4.6%
.ar27391.9%8.1%26.7%15.8%1.5%
.hu26876.1%23.9%31.7%5.6%1.1%
.ch25378.3%21.7%20.2%12.3%5.9%
.be24179.3%20.7%14.9%15.8%12%
.us23492.7%7.3%21.8%24.4%8.5%
.at23375.1%24.9%18.5%8.2%1.7%
.za23385.4%14.6%39.5%8.2%0%
.pro22793%7%48%5.7%0%
.online22693.4%6.6%49.6%8%0%
.live22499.6%0.4%60.7%12.9%0%
.fi21765.4%34.6%13.4%28.6%1.8%
.cl21487.4%12.6%26.6%18.7%3.3%
.site20795.2%4.8%47.8%10.6%0%
.top20694.7%5.3%53.9%1.9%0%
Download the full dataset (CSV)

Methodology

dns.pizza records each domain's nameservers while crawling the Tranco top 100,000 and keeps the latest set per domain. This report reads those records for 101,638 ranked domains refreshed between July 31, 2026 and October 1, 2026. 99,262 of them had at least one parseable nameserver hostname and form the corpus; the remaining 2,376 — 2,356 with no nameservers recorded and 20 whose only values were bare IPs or malformed names — tell us nothing about who runs their DNS and are excluded from every denominator. The corpus is slightly larger than 100,000 domains because ranks accumulate across Tranco revisions, so a few ranks are held by more than one domain over time.

Each nameserver hostname is grouped to its provider zone with the same rule the dns.pizza nameserver hub uses — the last two labels, or three under a multi-part public suffix, so ns1.foo.co.uk groups to foo.co.uk and not co.uk. Zones are then mapped to the operator that owns them, which is the step that matters: Route 53 is spread over hundreds of awsdns-NN.{com,net,org,co.uk} zones and Azure DNS over four azure-dns.* zones, so a zone-level ranking reports those customers as multi-homed when they are not. Cloudflare's enterprise Foundation DNS zones are counted as Cloudflare.

Zones we do not recognise keep the zone itself as their operator identity, never a shared bucket. That matters for correctness: folding every unknown zone together would make a domain on ns1.foo.com plus ns2.bar.net look single-operator when it has two. A domain counts as single-provider only when every nameserver it publishes maps to one operator. An operator earns its own row at 200 domains of footprint; 14,794 smaller operators fold into "everyone else". That row shows no footprint: it folds thousands of separate zones, so a domain using two of them cannot be counted once, and an inflated number would be worse than none. Per-TLD rows need 200 measured domains, giving 57 rows; the gate leaves out 10,186 domains across 602 smaller TLDs, which remain in the overall and per-tier numbers. Every figure the page cross-references comes from a single SQL statement, so the segments cannot disagree with their own total.

Limitations. The 90.6% single-provider figure is a floor. When one company runs its own DNS across two of its own zones we count two operators: bbc.co.uk publishes nameservers under both bbc.co.uk and bbc.com, and lands in the multi-provider column despite being one organisation on one infrastructure. That bias inflates measured redundancy wherever the long tail of self-run and small-hoster DNS is thickest, which is exactly where the most-redundant-TLD table points — so read .fi figures as an upper bound on real redundancy, not a confirmed one.

The combined 51.4% for the two largest operators is a sum of two overlapping footprints, so it is an upper bound rather than a count of distinct domains: a domain using both is counted once in each. At most 176 domains can be in that overlap, so the share they serve between them is somewhere between 51.2% and 51.4%. Every single-operator figure on this page is a true distinct count and is unaffected.

"Single provider" is a statement about administrative dependency, not about servers. Every large operator here runs anycast across many sites, and a single well-run anycast network is more reliable than two poorly-run unicast ones; the risk a single operator concentrates is correlated failure of a shared control plane, routing or account — the class of event that has taken large parts of the web offline before. Nameservers are read as our resolvers see them, from one vantage point, which is the delegation in practice rather than a registry zone-file audit; a domain mid-migration is captured in whatever state it was in when last checked. One domain is one vote, so these are shares of popular domains, not of DNS query volume — and Tranco ranks popularity signals, not visits, so this describes the well-known web rather than all registered domains.

The data is free to cite and reuse (CC BY 4.0, credit "dns.pizza"). For questions or the raw per-domain data, get in touch.

How many companies answer for your domain?

Look up your NS records live — every nameserver, who operates it, and whether you have a second operator at all.

Check your nameservers