Summary
- Thirteen root server identifiers are operated by twelve organisations, and every count of the same service that I found uses a different unit: identifier, operator, address record, instance, node, site.
- For the one letter Netnod runs, the operators' association prints 93 sites and Netnod prints "more than 70 nodes". Both statements are on the record, neither defines the other, and the difference is not a discrepancy.
Three lines. That is what the root hints file, in the version marked 12 November 2026, gives a resolver for i.root-servers.net: an NS record, an A record at 192.36.148.17, an AAAA record at 2001:7fe::53. The file's own header says what it is for — it "holds the information on root name servers needed to initialize cache of Internet domain name servers" — and for that job three lines is the entire requirement.
Everything else about the same letter is counted somewhere else, in a unit someone else chose. The operators' association prints 93 sites for I-root. Netnod prints more than 70 nodes. The association's system-wide headline is 2046 operational instances, and the advisory committee's public FAQ says there are more than 1500 anycast instances in operation throughout the world. No document I read on 20 November 2026 converts any one of those figures into any other, and the units are defined well enough that the conversion cannot be assumed either.
Thirteen names, twelve operators
The logical layer is genuinely simple, and the standard states it without decoration. RSSAC001 Version 3, published 2 September 2026, opens with a sentence that could be printed on any slide about the root: "At the time of writing there are thirteen root server identifiers, operated by twelve different RSOs."
Thirteen names, twelve organisations, and one of those organisations — Netnod — holding exactly one letter. IANA's root servers page presents the same layer as a list of 13 named authorities and puts i.root-servers.net at the same two addresses the hints file carries, with Netnod named as the operator. The association's page repeats the structure in its own words: "The 13 root name servers are operated by 12 independent organisations."
So far every count agrees because every count is counting names. The trouble starts one level down, where the names are implemented.
What an instance actually is
The definitional work has been done, carefully, and it is worth reading rather than paraphrasing. RSSAC001v3: "A root server "instance", or an "anycast instance", is the portion of a root server's infrastructure that serves root data at one site." The older RSSAC lexicon of 12 March 2020 is more explicit about what a site is: "An instance, or an anycast instance, is the portion of a root server operator's infrastructure that serves root data at one site (i.e., topological location on the Internet) using an IP address associated with a root server identifier."
A site, in that phrasing, is a topological location on the Internet — not a building, not a rack, not a machine. And the operators' own documentation makes the remaining looseness explicit. The operational information document published alongside the association's map says a site "essentially corresponds to a geographic location such as a town or city", and then says the thing that matters more: an operator may group several instances at one place under a single site with the instance counter set above one, or split them into separate sites with the counter set to one. The per-site instance number is the operator's to set. It is not a server count, a rack count or a data centre count, and the document does not pretend otherwise.
That is not evasion. It is a schema honest about its own granularity. But it means the word "instance" in a headline is a unit of operator reporting, and the only party who knows what went into a given number is the operator that filed it.
The number the association prints
The association's page carries a live timestamp and a headline: "The root server system consists of 2046 operational instances operated by the 12 independent root server operators." I read that page twice on 20 November 2026 — at 06:22:25Z and again at 21:05:35Z the previous day by its own clock — and both retrievals printed 2046. The timestamp moves; the count did not.
Per letter, the page uses a different label set: Sites, Operational, Suspended, Decommissioned. For I-root, the row reads 93 sites, 93 operational, 0 suspended, 0 decommissioned, beside IPv4 192.36.148.17 and ASN 29216. Those are sites, not instances, and the page does not present them as a decomposition of the 2046.
Here is the part that took two retrievals to be sure of. The page is the closest thing to a reconciling inventory that exists in public: it organises the system by letter and operator, it labels sites and statuses, and it reports instance counts at individual site entries. And in the text I received, the headline 2046 is not accompanied by an arithmetic that sums those per-site values. The total is printed; the addition is not. Both retrievals of the long page were truncated, so I am describing what I read, twice, and not what the page might carry further down.
What Netnod prints about its own letter
Netnod's own I-root page is shorter and uses a word nobody has defined. It states that since 2000 Netnod has operated i.root-servers.net, "one of the 13 logical Internet DNS root name servers"; that the service is provided by distributed nodes using IPv4 and IPv6 anycast; and that "There are currently more than 70 nodes distributed across the world."
Node. Not instance, not site. The page does not define it, in the text I retrieved, and that is the whole of the definitional gap — one word carrying an unbounded quantity inside a sentence that also carries a qualifier. More than 70 is a floor. A floor is not a count, and a node is not a site.
So put the two statements side by side and resist the arithmetic everyone reaches for. The operator says more than 70 nodes. The association says 93 sites for the same letter. Those are not two answers to one question; they are two answers to two questions that share a subject. The operator's figure is a lower bound in an undefined unit belonging to the party that runs the deployment. The association's figure is a status-tagged site count belonging to a schema that lets the operator decide how many instances sit at each site.
Anyone who subtracts one from the other, or divides one by the other, has invented a fact about I-root's physical footprint that neither publisher stated.
The hints file, and IANA's "hundreds"
The hints file is the smallest population in play and the only one with a dated, machine-checkable form. In the version marked 12 November 2026 each letter carries an IPv4 and an IPv6 address record; the excerpt I can quote verbatim is the I-root block above. The wider totals of thirteen NS records and twenty-six address records in the retrieved file are the retrieval provider's count of that file, not my own line-by-line enumeration, and I am flagging that because the difference between an address record and an instance is the entire subject of this piece.
IANA sits above all of it with a sentence that refuses to be a count: "The authoritative name servers that serve the DNS root zone, often known simply as the "root servers", are a network of hundreds of servers in many countries around the world. They are configured in the DNS root zone as 13 named authorities, as follows."
Hundreds of servers. It is the plainest description in the set and the least convertible: no date, no total, no definition of server, and no way to reconcile it with 2046 instances or with 93 sites. It is also the sentence most likely to be repeated in a briefing, because it is the one a non-specialist can hold in their head.
What the advisory committee says, and does not publish
The RSSAC FAQ resolves any lingering hope that a standards body keeps the register. "While today there remain 13 RSIs, there are in fact more than 1500 anycast instances in operation throughout the world." Approximate, undated in the text I retrieved, and offered as background rather than as inventory: the same FAQ sends readers to the operators' own data for the detail.
That is the shape of the disclosure perimeter, and it is not a scandal. The definitions live with the advisory committee. The publication expectation lives with the operators, individually. The aggregation lives with an association whose schema is documented, whose per-site values are operator-supplied, and whose system total arrives without a printed derivation. Nothing in that arrangement hides a number. It simply distributes the authority to say what is being counted.
Eighteen boundaries
- Every figure here is reproduced as its publisher printed it on 20 November 2026 between 06:23 and 06:26 UTC, and nothing is converted between populations.
- An operational instance is not a server, a rack or a data centre. The definition is per-site serving infrastructure, and the per-site instance value is operator-determined.
- The 2046 headline's arithmetic is not published in the text I received: the page prints the total and the per-letter counts, and no sentence summing sites or per-site values into 2046 appeared.
- The association's page was read twice with two different printed timestamps and the same 2046, so that figure is a reading taken at a moment rather than a standing condition.
- The I-root site row and the ASN 29216 detail come from that page's per-letter summary, identical in both retrievals; the long page was truncated both times, so rows beyond the I-root summary and any methodology further down were not read.
- Netnod's "more than 70 nodes" is a lower bound in a unit its page does not define. It is not converted into 93 sites, 2046 instances or any other figure here, and the two numbers are not in conflict.
- The root hints I-root lines are quoted verbatim from the version marked 12 November 2026; the wider totals of thirteen NS records and twenty-six address records are the provider's count of the retrieved file, not my own enumeration.
- The 2020 lexicon says of itself that it does not intend to redefine technical terms and that its examples were accurate at publication time; its definition is quoted as printed, not as current registry policy.
- The RSSAC001v3 quotation carries an ellipsis in my evidence record and was read as provider-extracted PDF text, not as a page image.
- The statement that operators are expected to publish locations, addressing and routing information comes from the provider's summary of RSSAC001v3 rather than from a sentence I can quote, and no argument above depends on it.
- The advisory committee's "more than 1500 anycast instances" is approximate and undated in the retrieved text; it is printed beside 2046 and never averaged with it.
- The advisory committee's landing page did not resolve during retrieval. No fact here comes from it, and that gap describes my access on 20 November 2026, not anyone's publication practice.
- No claim is made that any operator, association or advisory committee misstates its footprint, conceals capacity or publishes a wrong number. The finding is that the printed numbers count different things.
- Netnod AB's directory record — entity, region as stored, company subtype — is the directory's record; the operator's own pages are the evidence about I-root, and the two are not interchangeable.
- "Since 2000" is the operator's own account of its tenure. No designation document, contract or registry approval was read for it.
- IANA's "hundreds of servers" carries no date and no count and is not converted into any figure here.
- Prior BTW coverage of Netnod is named rather than repeated: the Swedish-krona price list of 13 November 2026, the directory-card reading of 4 November 2026, the record-card and three-surface research of 29 October 2026, the three-service-lines briefing of 18 October 2026, the July exchange-and-I-root draft, and the 2024 Cross-IX profile. None of them took the root-server system itself as the counted object.
- No BGP session, looking glass, zone transfer or query tool was run. Nothing here reports the state of any resolver, instance or route at any instant beyond what the publishers themselves printed.
What a root-server number is worth
The practical consequence is dull and useful in equal measure. A root-server figure is a statement about the instrument that produced it, and the instrument is visible if you look. Thirteen and twelve are names. Ninety-three is a site count inside a schema that lets the operator decide how many instances sit at each site. Two thousand and forty-six is a system total of those operator-chosen values, printed without its addition. More than seventy is an operator's own floor in a word it never defines. Hundreds is a description with no date.
None of these are wrong, and the people who publish them are not hiding anything. But the next time a procurement document, a resilience claim or a policy paper asserts that an operator runs a certain number of root servers, ask which of the six things it means. The honest answer is usually that the author took the biggest number they saw, and the biggest number counts something nobody promised to keep steady.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






