Which AI model providers UK companies use

Summary  ·  We collected 1,377 references in public evidence to UK companies, across industries, using AI models. OpenAI accounts for 48.7% of them, twice Anthropic’s 23.4%. Open-source models come third at 17.1%, ahead of Google at 10.9%. A company that uses several providers is counted once per provider, so these are shares of references, not of companies.

Market research report · November 2026 · 1,377 company references.

Key numbers

• Concentration. OpenAI and Anthropic account for 72.0% of references (992). Adding Google takes it to 82.9% (1,142); adding open source instead takes it to 89.1% (1,227).
• Corroboration is flat across the named providers. 69.9% of OpenAI references, 68.5% of Anthropic references and 65.7% of Google references are corroborated use. The gaps are within sampling noise.
• The caveat load is large. 31.0% of references to the named providers (354 of 1,142, projected) are not corroborated use. They are conditional or stale statements, offers, pilots or plans.
• Open source lifts the blended rate. Overall, 74.3% of references are corroborated use (1,023 of 1,377). All 235 open-source references are corroborated, which lifts the rate from 69.0% across the three named providers.

Data and method

We collected public evidence of AI model use by UK companies across industries: case studies, supplier disclosures, privacy notices and processor lists, and company statements. Each reference links one company to one model provider. Two datasets do two different jobs:

Dataset	Contents	Used for
Headline provider totals	4 entries: OpenAI 670, Anthropic 322, open-source models 235, Google 150. Total 1,377.	All shares, the league table and concentration
Coding sheet (15 providers)	734 references coded Supported, Qualified only or Other only; 509 Supported	Support rates for the named providers and the smaller providers

Each reference falls into one of three codes:

• Supported: disclosed use or an explicit AI supplier assignment, with a corroborated UK company link. The company’s or supplier’s own statement counts as evidence; it is not independent runtime verification.
• Qualified only: conditional, illustrative, stale or identity-unresolved evidence.
• Other only: offered integrations, pilots, plans, historical use, research or rejected attribution.

Two reading rules. References overlap: a company using OpenAI and Anthropic appears under both, so shares are not company counts or market shares. And the sample is companies with public evidence, not a representative sample of UK businesses, so no national adoption rate follows from it.

A projection. The headline totals have no Supported, Qualified or Other split. For OpenAI, Anthropic and Google we apply the coding-sheet rates to the headline counts and label the results derived. The scaling is not uniform: OpenAI’s headline total is 2.08 times its coded count, Google’s 1.52 times and Anthropic’s 1.39 times. The derived counts are an assumption, not a measurement.

Open source. “Open-source models” covers companies using open-weight models, directly or through routers. It is one provider row of 235 references, all Supported and deduplicated against the named providers. A router counts as supporting the models it serves.

Provider league table

OpenAI is the most-referenced provider by a wide margin, with 670 company references, 2.1 times Anthropic’s. Even so, it has 37 fewer references than the other three combined (707).

Model provider	References	Share	Support rate	Supported (derived)	Qualified only (derived)	Other only (derived)
OpenAI	670	48.7%	69.9%	468	108	94
Anthropic	322	23.4%	68.5%	221	51	50
Open-source models	235	17.1%	100.0%	235	0	0
Google	150	10.9%	65.7%	99	24	27
Total	1,377	100.0%	74.3%	1,023	183	171

The support rates for the named providers come from the coding sheet. Their three count columns are projections, rounded so each row sums to its total. The shares are solid; the derived counts are not raw counts.

How much of the evidence is corroborated use

Of all 1,377 references, 1,023 are corroborated use (74.3%). Another 183 are qualified evidence (13.3%), and 171 are offers, pilots, plans or historical use (12.4%).

The named providers carry 354 exceptions on 1,142 references, or 31.0%. On the raw coding sheet the same three carry 204 exceptions on 653 references (31.2%), so the projection does not distort this figure. Open source adds no exceptions, which lifts the blended rate from 69.0% for the named providers (788 of 1,142) to 74.3%.

Concentration

References concentrate on one dominant provider and one strong second, with open source a clear third.

Providers	References	Share
OpenAI alone	670	48.7%
OpenAI + Anthropic	992	72.0%
OpenAI + Anthropic + Google	1,142	82.9%
OpenAI + Anthropic + open source	1,227	89.1%
All four	1,377	100.0%

Anthropic leads open source by 87 references, or 6.3 share points. Google is the smallest of the three named providers, and on the coded sample it has the highest Other-only share (18.2%, against 16.2% Qualified). More of the evidence around Google is offers, pilots or plans rather than use than for any other named provider.

Support rate by provider

The support rate is the share of a provider’s company references that are corroborated use. The three named providers overlap almost completely. Mistral’s 76.7% looks higher, but on 30 references its interval runs from 59% to 88%. Only Amazon, at 25 of 25, sits clearly above the majors. Open source (235 of 235) is not shown, because it sits outside the coding sheet.

Smaller providers

Twelve providers appear in the coding sheet but not in the headline totals: Mistral, Amazon, IBM, xAI, Meta, Alibaba / Qwen, Cohere, Stability AI, DeepSeek, Midjourney, Stable Diffusion and Voyage AI. Together they have 81 company references, 60 of them corroborated use (74.1%). That is 5.9% of the 1,377.

Leaving them out of the headline is a coding decision. Including them would take the total to 1,458 and lower every headline share by about 5.6% in relative terms: OpenAI to 46.0%, Anthropic to 22.1%, open source to 16.1% and Google to 10.3%.

Findings by industry

Software, IT and AI services account for 40% of references (392 of 969) and 37% of corroborated use. Open source holds a near-constant 28–33% of corroborated use in every sector group, while support rates range from 63.5% in marketing, media and creative to 93.7% in education.

This cut uses a different base from the headline shares. It covers 969 references: the 734 coded references plus the 235 open-source references, each assigned to the company’s sector. On this base open source is 24.3% of references, against 17.1% of the 1,377 headline figure. The 31 sectors with references are grouped into nine sector groups.

Sector group	References	Supported	Support rate
Software, IT and AI services	392	278	70.9%
Financial services	138	109	79.0%
Business and professional services	95	81	85.3%
Education	79	74	93.7%
Marketing, media and creative	74	47	63.5%
Industry, energy and transport	57	47	82.5%
Retail, food and hospitality	56	43	76.8%
Legal services	47	40	85.1%
Healthcare and life sciences	31	25	80.6%
Total	969	744	76.8%

How the corroborated use splits by provider in each group:

Sector group	Open source	OpenAI	Anthropic	Google	Other
Software, IT and AI services	32%	32%	18%	9%	9%
Financial services	33%	25%	18%	16%	8%
Business and professional services	31%	30%	28%	4%	7%
Education	32%	32%	18%	9%	8%
Marketing, media and creative	32%	28%	28%	9%	4%
Industry, energy and transport	32%	36%	13%	9%	11%
Retail, food and hospitality	30%	23%	37%	5%	5%
Legal services	30%	35%	25%	5%	5%
Healthcare and life sciences	28%	28%	28%	8%	8%

• Open source is flat across sectors. Its share of corroborated use stays between 28% and 33% in all nine groups.
• OpenAI leads the named providers in most sectors, but not everywhere. Anthropic leads in retail, food and hospitality (16 Supported references against OpenAI’s 10), and ties OpenAI in marketing, media and creative and in healthcare.
• Google is strongest in financial services. It has 15.6% of corroborated use there, its highest share, and only 3.7% in business and professional services.
• Marketing, media and creative has the weakest evidence. 27 of its 74 references are qualified, offers, pilots or plans rather than corroborated use.
• The smaller groups are directional only. Healthcare (31 references) and legal services (47) are too small for firm provider comparisons.

Conclusions

• Two providers dominate UK company references. OpenAI and Anthropic account for 72.0%, and OpenAI alone for nearly half.
• Open source is a confirmed third. It has 17.1% of references, and every one of them is corroborated use.
• Corroboration does not separate the named providers. They sit at 65.7–69.9%, within each other’s confidence intervals. What separates them is how many companies reference them.
• About 31% of references are not corroborated use. Nearly a third of references to the named providers are conditional statements, offers, pilots or plans. That belongs in the headline, not in a footnote.
• Only one smaller provider clearly beats the majors. Among providers with enough references to tell, only Amazon (25 of 25) does.

Limitations

• References, not companies. A multi-provider company is counted once per provider. Shares describe references, never company counts, adoption or spend.
• Public evidence only. Companies that say nothing in public are absent, so no UK-wide adoption rate follows.
• Self-reported evidence. A company’s or supplier’s own statement counts as Supported; it is not runtime verification.
• Derived counts. The Supported, Qualified and Other counts for OpenAI, Anthropic and Google are projections from the coding sheet, not raw counts.
• Small groups. Providers and sector groups with few references are directional only.

---

Compiled November 2026. Shares cover 1,377 references to UK companies using AI models; the sector cut covers 969 references.