What this is. An original dataset, built from one snapshot of the Companies House register and published with the rules that produced it. It counts companies whose registered name contains AI. It is not a count of AI companies, and the difference is the subject of the piece.
Using these figures. Free to cite with a link to this page. The method, the rejected rules and the caveats are all below, because a number without them is not worth quoting.
Want the data itself? The derived dataset and the scripts that produce it are yours on request. Email contact@krawl.digital and they will be sent over.
What AI in the name actually finds
The selection rule picks up companies like Stability AI, Perplexity AI and Humanloop AI, whose names clearly advertise an AI focus, alongside firms whose activities lie elsewhere, such as AI Barber Ltd (hairdressing), AI Building & Roofing Ltd (construction) and AI Commercial Cleaning Limited (specialised cleaning).
It also reaches back to AI (Bullers) PLC, incorporated in 1902 and registered as non-trading, and would have swept in long lists of people and places such as Ainger Road, Airconditioning, Desai, Acai, Aitken and Airdrie if the rule had matched AI inside longer words.
How common AI naming has become
Among new companies incorporated in 2026 to 1 September, 2,860 of 525,664 companies incorporated that year have AI in the name under this rule, which makes AI branding a routine choice rather than a curiosity.
Across the survivor stock of the register, the shift is captured as one in 1,526 companies incorporated in 2019 and one in 184 of those incorporated in 2026 to 1 September, showing how a once-rare naming choice has become part of mainstream company formation.
The numbers behind this chart
| Period | Value (%) |
|---|---|
| 2017 | 0.05 |
| '18 | 0.058 |
| '19 | 0.066 |
| 2020 | 0.087 |
| '21 | 0.079 |
| '22 | 0.083 |
| '23 | 0.211 |
| '24 | 0.274 |
| 2025 | 0.481 |
| 2026* | 0.544 |
The numbers behind this chart
| Row | Value |
|---|---|
| 2017 | 108 |
| '18 | 145 |
| '19 | 190 |
| 2020 | 241 |
| '21 | 231 |
| '22 | 269 |
| '23 | 862 |
| '24 | 1317 |
| 2025 | 3522 |
| 2026* | 2860 |
What was actually measured
The cohort here is companies on the UK register on 1 September 2026 whose registered name contains AI as a separate word or one of eight explicit AI phrases, which means every finding in this article is about naming rather than about what the companies really do.
On that basis, 10,011 of 5,689,367 companies on the register match the rule, but this is not a census of AI businesses, because the register records no company descriptions, no products and no websites.
The same rule that finds Stability AI, Perplexity AI and Humanloop AI also misses DeepMind, Synthesia, Wayve, Faculty and Quantexa, so the measurements here are a count of AI-named companies, not of AI companies trading in the UK.
Why the names look like this
Registering a company is cheap and choosing a name is free, so putting AI in the name is an easy way for founders to signal what they hope to build long before they have a product or a customer.
Because the name is the only visible claim at the moment of incorporation, it carries marketing, investor signalling and search visibility all at once, which helps explain why AI appears in sectors from software to hairdressing.
Formation agents and the addresses behind AI names
23.1% of AI-named companies against 10.4% of the register as a whole sit at addresses that already host hundreds of other companies, which makes AI naming far more likely at formation-agent maildrops than across the register as a whole.
Those addresses are treated as formation agents once they hold 500 companies, because at that point the registered office is clearly a service address rather than a workplace.
The single address with the most AI-named companies is 71-75 Shelton Street, WC2H 9JQ, which hosts 768 of 10,011 AI-named companies and is one of several London maildrops that also carry tens of thousands of other firms.
The numbers behind this chart
| Row | Value |
|---|---|
| 71-75 Shelton Street | 768 |
| 128 City Road | 406 |
| 20 Wenlock Road | 169 |
| 167-169 Great Portland Street | 167 |
| 3Rd Floor, 86-90 | 105 |
| 66 Paul Street | 97 |
How easily one of these is made
On the UK register, a company can be incorporated quickly with a small fee, standard documents and a registered office that is allowed to be a mail‑forwarding or formation‑agent address, and the data here shows that the newer the companies on the UK register on 1 September 2026 whose registered name contains AI as a separate word or one of eight explicit AI phrases cohort is, the more of it sits at one of those high‑volume service locations 13.0% of those incorporated in 2022 or earlier and 26.1% of those incorporated since the start of 2025.
The same register also records that a share of these AI‑named companies is already flagged for strike‑off 577 of 10,011 AI-named companies, but the filings cannot show whether anyone is working at any of them, what they do day to day, or whether they have ever traded at all.
What cleaning the addresses changed
Removing formation-agent addresses and placing companies in local authorities leaves 7,544 of 10,011 AI-named companies, and that one step is enough to move Herefordshire from 11th in Britain before cleaning and 120th after and Hillingdon from 17th before cleaning and 70th after in the per-capita ranking.
The numbers behind this chart
| Row | As registered | Agent addresses removed | Gap |
|---|---|---|---|
| Camden | 467.1 | 95.08 | -372.02 |
| Islington | 318.41 | 74.82 | -243.59 |
| Hackney | 215.59 | 31.55 | -184.04 |
| West Suffolk | 61.24 | 25.87 | -35.37 |
| Herefordshire | 31.16 | 8.83 | -22.33 |
| Cambridge | 36.7 | 36.7 | +0 |
Where AI-named companies actually cluster
On the cleaned base, 2,721 of 7,544 companies placed in a local authority are in London local authorities, but this still reflects where companies are registered rather than where they employ people or build products.
Outside inner London, Cambridge stands out with 36.7 per 100,000 residents, and it is the one place whose per-capita position barely shifts when formation-agent addresses are removed.
Most local authorities hold very small numbers of AI-named companies, with the median at 11, which is why no full per-capita league table is published: a ranking built on such thin counts would invert whenever the cleaning rule changed.
The numbers behind this chart
| Row | Value |
|---|---|
| London | 2721 |
| South East | 1131 |
| North West | 732 |
| East of England | 714 |
| West Midlands | 507 |
| South West | 376 |
| Yorkshire and The Humber | 373 |
| Scotland | 324 |
| East Midlands | 282 |
| Wales | 158 |
| North East | 129 |
| Northern Ireland | 85 |
What this data cannot show
The counts by year, such as 2,860, are survivor stocks rather than true formation flows, because the register omits dissolved companies and 2026 only runs to 1 September, so earlier years are depleted and later ones incomplete.
Methodology
Derived from the Companies House free company data product snapshot dated 1 September 2026, which lists every company then on the UK register. A company counts if its registered name contains AI as a separate word written without full stops, or one of eight explicit phrases: artificial intelligence, machine learning, deep learning, computer vision, NLP, generative AI, large language model, neural network. Three further rules were tested and rejected: AI glued inside a longer word, A.I. written with full stops, and neural on its own. Postcodes were resolved to local authorities through the ONS postcode directory, and population figures are ONS mid-2025 estimates. Geographic figures exclude companies registered at any address holding 500 companies or more, and Northern Ireland is absent from the per-capita comparison because the ONS series used here does not cover it. The bulk file was read once and deleted. The derived dataset and the three scripts that produce it are available on request: email contact@krawl.digital and they will be sent over.
Source
Companies House free company data product and ONS postcode and population data, both under the Open Government Licence v3.0. Snapshot dated 2026-09-01.