UK SME AI spending hits a new high as Anthropic takes 54.1% and OpenAI 27.3%
Capital on Tap card data shows UK SMEs buying AI rising from 1.1% to 12.8% of cardholders, with Anthropic at 54.1% of that spend against OpenAI’s 27.3%.
Capital on Tap is a London-based business credit card company for UK small and medium-sized firms. On September 10, 2026 it published “How SMEs in the UK are adopting AI,” an analysis of card transactions by its own cardholders between Q1 2023 and Q2 2026. Two numbers carry the story: the share of SME cardholders buying AI tools rose from 1.1% to 12.8%, and Anthropic took 54.1% of that AI spending against 27.3% for OpenAI — the first time Anthropic leads in this dataset. The blog dates the data collection to August 27, 2026.
Key points
- The lead flipped inside one quarter. OpenAI still led in Q1 2026 with 41.8% of AI spend to Anthropic’s 34%; a quarter later Anthropic held 54.1% and OpenAI 27.3%. Year-on-year spend growth was 3,930% (about 40x) for Anthropic and 140% (about 2.4x) for OpenAI.
- Reach and wallet share have come apart. OpenAI shows up in 65.2% of AI-buying businesses, Anthropic in 48.9%, yet Anthropic collects nearly twice the money. Meanwhile 28.2% of AI buyers pay two or more providers, up from 11.6% in Q1 2023.
- AI is still a rounding error in the budget. AI purchases account for 0.146% of total card spend in the dataset, and the typical single purchase barely moved, from £16.65 to £18.10 — the growth is more purchases, not pricier ones.
Background
AI was a tiny line item at the start of this window. Across 2024, Anthropic accounted for between 1.6% and 6.6% of SME AI spend while OpenAI held more than 70%. City AM reported the same dataset exclusively on August 14 (updated August 19), noting OpenAI still had 70% of the market as recently as Q2 2025 against 8% for Anthropic, and describing the sample as “hundreds of thousands” of business customers of a company that processes tens of billions of pounds a year. The same data point: at the start of 2025, only one in twenty cardholders was paying for AI at all.
Both vendors’ recent moves land on the same invoice. Anthropic bundled Claude API credits into its subscriptions on October 7 — up to $500 a month on Team, usable for apps and agents you build but not for Claude Code, as we covered in Claude Max and Team plans now ship monthly API credits. OpenAI’s comparable story is a company-level revenue figure, not SME card spend: OpenAI’s annualized revenue is near $50 billion counts global revenue, so the two numbers cannot be used to check each other, only to show both firms pushing into enterprise budgets.
A third thread is coding tools. Indeed data after Claude Code’s launch looks at hiring; this dataset looks at cards.
The facts
Every figure below comes from the Capital on Tap blog post, data collected August 27, 2026 and published September 10, 2026. The company defines “buying AI” as the share of cardholders with any transaction at an identified pure-play AI provider in that quarter.
| Metric | Q1 2023 | Q2 2026 |
|---|---|---|
| Share of cardholders buying AI | 1.1% | 12.8% |
| AI as a share of total card spend | 0.005% | 0.146% |
| Typical quarterly AI spend per buyer | £33.07 | £75.60 |
| Typical spend per purchase | £16.65 | £18.10 |
| AI buyers using 2+ providers | 11.6% | 28.2% |
| Coding tools as a share of AI spend | 0 | 11.6% |
Provider shares for Q2 2026:
| Provider | Share of AI spend | Share of AI buyers | YoY spend growth |
|---|---|---|---|
| Anthropic | 54.1% | 48.9% | 3,930% (40.3x) |
| OpenAI | 27.3% | 65.2% | 140% (2.4x) |
| Cursor | 4.4% | 3.1% | 500% |
| Replit | 3.9% | 2.3% | 760% |
| Lovable | 3.1% | 4.2% | 1,010% |
| Perplexity | 1.9% | 2.4% | 1,050% |
| Manus, ElevenLabs, Fireflies, HeyGen, Midjourney | 4.2% | 10.1% | up to 710% |
| Nine providers below 1% reach | 1.1% | 2.7% | 110% |
The coding-tool line needs care: its share fell from 15.8% of AI spend in Q1 2026 to 11.6% in Q2, yet the company’s coding spend index reached 2,214 in Q2 2026 against a baseline of 100 in Q1 2025. The share dropped because everything else grew faster, not because coding spend shrank. By sector, publishing leads adoption at 26%, radio and television and consulting both sit at 21.6%, and telecoms at 20.3%; restaurants trail at 4.8%, with transport and food retail at 4.9%. By region, London reached 16.6% adoption (up from 1.8% in Q1 2023) while Yorkshire and the Humber sits lowest at 11.2% — but that region’s AI buyers post the highest median spend.
The methodology carries real caveats. The data comes from one company’s cardholders; Capital on Tap does not disclose the sample size, the number of cardholders or the transaction count, and it does not say how it identifies AI merchants in card records. The blog states plainly that “all figures reflect purchasing behaviour within the Capital on Tap SME cardholder dataset and should not be interpreted as a measure of total UK SME AI adoption.” City AM adds that the data excludes AI services sold by software companies such as Google and Microsoft, because non-AI spending cannot be separated out.
Two spending figures do not agree. The blog reports typical quarterly AI spend per buyer rising from £33.07 to £75.60. City AM, on August 14, reported average quarterly AI spend across Capital on Tap’s customer base of £288, pushed up substantially by the top percentile of customers spending over £3,000 a quarter. The methodology note mentions median, mean, p90 and p99 values without saying which one £75.60 is; £288 is closer to a mean. They differ by roughly 3.8x. This article reports each at its own stated basis and does not average them.
What others say
Capital on Tap CEO Damian Brychcy is the source of the official framing. In the blog he writes that “a three-person firm can now do its own research, marketing, bookkeeping and customer service to a standard that used to need a department,” and that “more than a quarter of AI buyers now pay for two or more providers, and the balance between those providers can flip inside a single quarter.” In the same passage he tells customers to “review what you pay for, cut the duplication and be honest about whether each tool is earning its place.” Speaking to City AM, he said adoption could triple from here — “every indicator is up and to the right” — and raised the price question: if a £25 subscription became £100, would companies still buy?
City AM (Simon Hunt) was the first outlet to report the data, on August 14, and read it as OpenAI’s “waning stranglehold” on the AI market, crediting Anthropic’s focus on enterprise customers.
TechRadar Pro carried the dataset as a news story on October 16 (Christian Cawley). The report puts the emphasis on duplicate subscriptions, writing that “there is a suspicion that SMEs are trying one AI — perhaps on a trial that they forget to cancel — then moving onto another,” and it carries Brychcy’s advice to “cut the duplication and be honest about whether each tool is earning its place.” The piece also notes the analysis cannot show whether SMEs are equipped to judge AI risk, since few “will have dedicated IT personnel who are completely across the details.”
TechRound’s Les-Leigh A, on September 16, asked three harder questions: whether Anthropic’s share reflects genuine preference or the heavy usage of agentic tools like Claude Code inflating the bill; whether multiple subscriptions are strategy or forgotten recurring payments; and whether a 1,000% adoption jump has produced any return.
ChannelX (Chris Dawson, October 13) reproduced the vendor tables in full, which is how the dataset reached the ecommerce-seller audience.
Our take
For anyone building for small businesses, the useful part is how the buyers are shaped rather than who ranks first. OpenAI reaches more firms while Anthropic collects more money, which means SME decisions are shifting from “install a chat entry point” to “pay for the tool we use daily” — and spend share follows intensity of use, not seat count. If you sell AI to SMEs, your competitor is less likely to be another vendor than the two or three unaudited subscriptions already on the customer’s card.
The honest catch: this is one card company’s cardholders, and only pure-play AI vendors. The 12.8% is a share of cardholders, not UK SME adoption, and the vendor says so itself. National statistics are the place to look for adoption rates; what this dataset supports is the direction and the structure.
What is still unknown: the blog states that “the data cannot explain why businesses are spending more with Anthropic,” so the Claude Code explanation remains outside inference; whether the Q2 flip holds in the next data point; and whether a price rise bends the adoption curve — a question Brychcy raised himself and left open.