How Big Is the Dark Web? What Metrics Show
Tor metrics for late 2026: about 3.3 million daily users, ~650,000 v3 onions seen, ~9,200 relays. Why iceberg charts are wrong and what Chainalysis can count.
The honest answer is a table of measurements, each with a hole in it. There is no census of "the dark web." There is Tor's public telemetry, a few academic crawls, and blockchain firms counting coins they know how to cluster. Those numbers are enough to kill the iceberg slide. They are not enough to say how many shops exist this week.
Figures below use Tor Project metrics for October 2026 unless noted.
What you can actually count
| Metric | Late Oct 2026 (approx.) | What it can tell you | What it cannot |
|---|---|---|---|
| Daily users (relay clients, all countries) | ~3.0–4.0 million; October mean ~3.3 million | Order of magnitude of Tor use | Unique people; many users are bridges, NATs, or brief sessions |
| v3 onion services seen by HSDirs | ~590,000–720,000; October mean ~650,000 | How many v3 addresses directory caches observed | How many are real sites vs. generated spam; they are not enumerable from the outside |
| Relays / bridges | ~9,200 relays, ~2,400–2,500 bridges (1–27 Oct) | Network capacity and censorship-resistance surface | "Number of dark-web sites" |
| Onion-service traffic share | On the order of tens of percent of relayed cells in recent years (varies) | That hidden services are not a rounding error | What those services are |
| Chainalysis-style market revenue | Hundreds of millions USD in peak years, methodology-dependent | Lower bound on visible Bitcoin-heavy markets | Monero volume; cash; shops the firm has not tagged |
User counts are estimates derived from directory requests, not logins. A café Wi-Fi of fifty people can look like one or like fifty depending on how the client behaves. Country tables are worse: a VPN exit in Germany inflates Germany.
Onion counts are the metric people want and the one that lies most politely. v3 addresses are 56-character hashes. You cannot scan the namespace. Tor estimates how many unique descriptors the directory system saw. A botnet spinning empty onions inflates the number; a quiet SecureDrop does not look different from a scam shop in that count.
Users are not "dark-web users"
Most Tor clients are fetching ordinary HTTPS: news, Wikipedia, a blocked social network. Onion-service traffic is only a slice. The myths page already buries the "everyone there is a criminal" line; the metrics bury the "everyone on Tor is on the dark web" line.
October 2026 country means from the same dataset (relay users, events off): United States ~640,000, Germany ~300,000, Iran ~53,000, Russia ~20,000. Iran and Russia are the interesting pair. National blocks push people onto bridges, and bridge users are counted separately and less completely. A collapse in "Russia users" on the public-relay chart can mean censorship, not a loss of interest.
The history of the overlay is a volunteer network that happened to host hidden services. Size debates that ignore that split produce the iceberg.
The iceberg and the 96%
BrightPlanet-era deep-web figures (the unindexed clearnet: paywalled DBs, intranets) got pasted onto Tor by a decade of infographics. The deep web is enormous. The dark web is a small overlay. Mixing them is how you get "96% of the internet is the dark web." It is also how you get the opposite panic, "70,000 sites of pure crime." Both are category errors. See dark web vs deep web and do not repeat the chart.
Older public estimates of "50,000–70,000 active onions" were plausible for earlier protocol eras and for crawls that only saw a subset. v3 estimates in 2026 are an order of magnitude higher on the directory-seen metric — and still say nothing about quality.
What blockchain firms measure
Chainalysis and peers publish yearly darknet-market figures by watching Bitcoin (and some other) clusters they have tagged as markets. That is a revenue estimate for the parts of darknet markets that still touch a transparent chain. It is one of the better sourced numbers in this space and it is still a lower bound: Monero-heavy shops, direct vendor deals, and untagged clusters are undercounted. The firm's 2024–2025 public write-ups described a market landscape that had shrunk from Hydra-era highs and then re-formed; use their latest table, not a tweet of the 2020 number.
Those dollars are not "the size of the dark web." They are the size of a business model that happens to sit on Tor. Forums, SecureDrop instances, and hobby onions do not show up in a Bitcoin cluster.
Academic work (Jardine, Owen/Savage, and later crawls) has tried to classify onion pages by content. Percentages of "illicit" vs "benign" swing with the classifier and the year. Quote a specific paper and year if you use one; do not invent a consensus 60%.
How to read a size claim
Ask: is this users, addresses, bytes, or dollars? Ask: which network (Tor, I2P, something else)? Ask: did they count v2, v3, or a crawl? If the graphic is an iceberg, throw it out. If the graphic is a single "number of dark-web sites," it is a crawl or a guess.
For a human-scale picture: a few million people a day use the overlay; on the order of hundreds of thousands of v3 addresses are seen; a thin slice of that is commercial crime that analytics firms can price; the rest is unused keys, mirrors, experiments and the same internet activities people do everywhere else, plus the newsrooms and sources who need the transport.
Frequently Asked Questions
How many people use the dark web?
Nobody knows "the dark web." Tor had on the order of three million relay users per day in October 2026. Many of them never open an onion. Unique humans are fewer than that estimate.
How many .onion sites are there?
Tor's directory-seen v3 estimate in October 2026 averaged around 650,000. That is not a count of working websites and not a count of shops.
Is the dark web bigger than the surface web?
No. The surface web and the deep web (unindexed clearnet) dwarf Tor. The iceberg chart that says otherwise is mixing categories. The myths page has the short version.
How much money do darknet markets make?
Analytics firms publish yearly Bitcoin-visible estimates in the hundreds of millions of dollars for recent years, with big swings after takedowns. Monero and untagged flow are mostly outside those totals.