American AI application companies that signed up with Stripe between January 2023 and June 2025 were 2.2 times as likely as their EU counterparts to reach $1 million in revenue within 12 months, according to an analysis published today by Marisa Rama on Stripe Economics, the payments company's economics publication. The essay, titled "Europe vs the US: how big is the AI gap?", argues that the distance between the two economies is wide at the bottom of the AI stack, where electricity, chips, data centres and models sit, but much narrower among the companies that build products on top of models. Among the consumers and new businesses that use those products, the gap shrinks further still. The subtitle states the conclusion plainly: "It's big, but not across the entire AI stack."

In Short

Stripe, which handles payments for many AI companies, used its own records to measure how far Europe trails the United States in artificial intelligence. Europe is far behind on power prices, data centres and the most advanced AI models, but the gap is smaller for companies that build AI apps, and Europeans and new European businesses use paid AI tools at rates closer to Americans than many people assume. The biggest US advantages come from a small group of very fast-growing app companies that reach $1 million in sales in their first year, and from established firms that use AI to make their products. All of it covers only businesses and people who pay through Stripe, so it is a large sample of the AI economy rather than the whole of it.

A cake with four layers

Rama builds the essay around a metaphor she credits to Nvidia chief executive Jensen Huang: AI development as a layer cake. The bottom layer is infrastructure - electricity, chips and data centres. Above it sit the models, which she describes as the weights that turn input tokens into output. The top layer is applications that use models without training their own, with AI shopping agents for consumers and AI legal assistants for businesses given as examples. "Finally, businesses and consumers eat the cake. This is adoption," she writes.

Why split the question this way? Because, in Rama's telling, treating AI as a single race hides where Europe is competitive. The essay frames the stakes in growth terms. Citing a 2006 paper by Daron Acemoglu, Philippe Aghion and Fabrizio Zilibotti, Rama writes that wealthy economies require frontier innovation to grow, not just the adoption of technology produced elsewhere. Innovation in AI, she argues, includes innovation at the application layer, and adoption carries its own weight. "If artificial intelligence is increasingly commoditized, value will accrue to the individuals and companies that make best use of it," she writes, linking to an Epoch AI publication titled "The plunging price of thought".

She also points back to earlier Stripe Economics research. "The age of the solopreneur", which Rama co-wrote with Ernie Tedeschi and Chris Cruickshank and published on June 22, 2026, argued that AI and AI-augmented software now cover tasks such as coding, pricing, marketing and sales that founders once hired people to do.

Where Europe trails: power, capacity and models

Rama does not contest the gap at the bottom of the stack. Electricity prices in Europe are more than double those in the US, according to the essay, which links to the International Energy Agency's Electricity 2026 report. According to the IEA, average EU industrial electricity prices in 2025 remained roughly double US levels and more than 50% above those in China and India, a gap similar to 2024.

The US has more data centre capacity than the next 14 countries combined, according to the essay, which links that figure to a chart posted on X by Ernie Tedeschi, who also writes for Stripe Economics. The essay does not name the dataset behind the chart.

On models, Rama is blunt: "despite some exciting recent announcements, Europe has no model that breaks the top 20 of either the main Artificial Analysis or Epoch AI rankings." The announcements she links to are posts on X from three European developers - Mistral AI, Aleph Alpha and Black Forest Labs. None of them, on Rama's account, has yet placed a frontier model in either top 20. The essay does not link to the two rankings or say on what date they were checked. For a fuller account of Europe's position at the lower layers, it points to a report hosted at transformative-ai.eu, describing the lag as "well-documented".

PPC Land has tracked those lower layers for more than a year. When Mario Draghi returned to Brussels a year after his competitiveness report, he warned that EU natural gas prices remained nearly four times US levels and that data centre electricity demand was projected to rise 70% by 2030. In the Netherlands, more than 11,900 businesses were waiting for grid connections in July 2025, with waits of up to 10 years. Investment has not stopped: on October 20, Amazon said it would put €4.8 billion into its AWS Frankfurt region in 2027 as part of a German plan worth more than €16 billion.

The model gap has also had commercial consequences this month. Berlin search engine Ecosia said on October 8 that it was replacing Mistral with open-weight models, among them Chinese models Qwen, GLM and Kimi, after chief executive Christian Kroll described Mistral's models as about a year behind. Mistral had released its 1-trillion-parameter Large 4 model two days earlier.

The application layer: more companies, not bigger ones

The essay's own data begins with the layer above the models. The comparison covers AI companies on Stripe that do not train their own models, with revenue approximated by payment volume on Stripe.

At the aggregate level, American AI application companies have achieved roughly 7x as much revenue as those in the EU and 15x as much as those in the UK, according to Rama. She then adjusts for the size of each economy. The US economy is about 50% larger than the EU's in nominal terms and about 8x the size of the UK's, so once that difference is removed, US aggregate AI application revenue is 5x that of the EU and twice that of the UK. By PPC Land's check, the arithmetic holds: 7 divided by 1.5 is about 4.7, and 15 divided by 8 is about 1.9.

The underlying chart, sourced to Stripe internal data and the World Bank, follows companies that signed up between January 2023 and June 2025 over their first 14 months on Stripe. It includes only companies with at least $100,000 in cumulative revenue, and it carries no vertical scale, so the absolute revenue behind the ratios is not disclosed.

Where does the gap come from? Rama decomposes it into the number of firms and their average size. There are roughly 5x as many AI companies in the US as in Europe, according to the essay, but US AI application companies are on average only 50% larger than their EU or UK counterparts. "This is not exactly cause for celebration but it is much better than the gulf at the model layer," she writes.

A second chart shows what that 50% means in dollars. By PPC Land's reading, after 14 months the average cumulative revenue per company stood at about $630,000 in the US, $430,000 in the UK and $420,000 in the EU27, again counting only companies above the $100,000 floor. The three lines barely separate for the first three months.

The $1 million cohort

The average revenue gap, according to Rama, is "largely driven by the lower prevalence of European 'blockbuster' companies", which she defines as those that have achieved over $1 million in revenue 12 months after sign-up. These companies make up close to 0.5% of AI application companies on Stripe in the US and account for a disproportionate share of revenue. In Europe and the UK, they make up 0.2%, and their contribution is far smaller. The essay does not say what share of total revenue the blockbusters represent on either side.

The milestone chart shows how the US advantage builds with scale. Against EU27 companies, US AI application companies were 1.1 times as likely to reach $1,000 within 12 months, 1.2 times as likely to reach $10,000, 1.2 times as likely to reach $100,000 and 2.2 times as likely to reach $1 million. Against UK companies, the multiples were 1.1, 1.3, 1.5 and 2.7.

One detail sits awkwardly with the text. The essay says the average US AI product company "is about as likely as a European one to reach $100,000 in revenue". For the EU, the chart's 1.2x is close to that description. For the UK, the chart prints 1.5x at the same threshold, a wider gap than the sentence suggests.

The application-layer argument has a precedent in European debate. In July 2025, SAP chief executive Christian Klein said Europe should focus on AI software applications rather than data centre construction, a position at odds with Huang's call for more European infrastructure investment. Stripe has also published broader comparisons before. In November 2025, co-founder Patrick Collison shared an index of software startup revenue in which the US reached about 1,400 against roughly 600 for the EU and 500 for the UK, from a common base of 100 in January 2020. In September 2026, Stripe founder surveys found that only 25% of founders in Spain would incorporate there again, against 85% in the US.

Consumers: fewer Europeans pay, but those who pay spend more

For consumer adoption, Rama turns to Stripe's position as a payment processor for AI companies. Stripe powers close to 90% of the Forbes AI 50 and tens of thousands of smaller AI companies, according to the essay. A Stripe blog post dated August 11, 2026 puts the figure at 88% of the 2026 Forbes AI 50. The essay says Stripe's infrastructure lets these companies sell in 195 countries.

The analysis covers 34 of the most popular paid consumer AI tools across the US, UK and EU27 since the 2022 launch of ChatGPT. A footnote sets out the method. Stripe excludes spend on business-focused products, such as plans with "business," "team," or "enterprise" naming, and excludes customers for whom it has business signals, such as a VAT ID. It then deduplicates payer IDs across providers to count unique consumers paying for AI in each market. The adoption chart counts consumers once even if they hold several subscriptions, includes free trials and discounted plans, and assigns location by payment card issuer or payer country, with population data from the UN's World Population Prospects 2024. The 34 tools are not named.

The headline finding: EU consumer paid AI adoption is about 60% of the US level. "This may sound like a large gap, but it is only equivalent to a few percentage points," Rama writes. The essay does not publish the absolute adoption rates, so the size of those few points cannot be checked from the text. UK adoption, according to Rama, tracks the US almost exactly.

The chart adds a time dimension. By PPC Land's reading, EU27 adoption started 2023 at about 31% of the US level, climbed to roughly 56% by early 2024 and 61% by early 2025, peaked near 66% in mid-2026 and stood at about 59% at the last data point. The UK line moved between roughly 88% and 104% of the US level across the period and ended at about 89%, so "almost exactly" describes the long-run pattern better than the latest month.

Why paid adoption

Paid AI is a small slice of total use. Rama puts overall consumer AI use close to 50% in both markets, linking to two surveys. The European Commission found that around 54% of individuals in the 18 member states covered by a February and March 2026 survey module use AI technologies, based on 21,207 responses. Pew Research Center found that 49% of US adults use AI chatbots, in a survey of 5,119 adults conducted from February 17 to 23, 2026. The two questions are not identical: one asks about AI technologies in general, the other about chatbots.

Rama's case for counting payers is that willingness to pay reflects consumers getting more use out of AI, whether through productivity gains or enjoyment. Paid adoption also carries more friction, which makes it, in her words, where a difference between EU and US adoption "would be most likely to emerge".

Spend per paying user

Adoption is only half of the consumer picture. Rama writes that "using an AI chatbot is very different from using an AI coding agent", so the essay uses average spend per paying customer across all tools as a proxy for intensity of use. On that measure, Europe and the UK are keeping pace. "If anything, dollar spend per paying user is higher in the UK and Europe than it is in the US, though this may reflect a selection bias toward a more enthusiastic early-adopter population in these markets."

The spend chart excludes free trials and fully discounted plans and spreads annual plans across their duration. By PPC Land's reading, both the UK and the EU27 sat around 110% of US average spend through 2023 and 2024. The EU27 line dipped to about 104% in early 2025 and again in early 2026, before both lines rose at the end of the series, to roughly 123% for the UK and 116% for the EU27. The essay does not explain how euro and pound payments are converted to dollars, or whether European consumer prices, which include VAT, are compared gross or net of tax. Either could move a comparison measured in dollars.

Countries and states

Neither side of the Atlantic is uniform. A scatter chart for August 2026, indexed to the US average, plots adoption against spend for 13 European countries and 10 US states. According to Rama, some EU countries, such as Denmark and Ireland, look more like high-adoption US states, while some large US states, such as Ohio and Pennsylvania, look more like the EU average.

The chart bears that out. By PPC Land's reading, Denmark sits at about 1.08 on adoption and Ireland at about 1.06, both above the US average. Ohio, at about 0.67, and Pennsylvania, at about 0.73, sit closer to the EU27 average of about 0.59 than to California at about 1.27. Italy has the lowest adoption in the chart, at roughly 0.37, followed by Spain at about 0.55. Germany and France both sit near 0.65.

Spend tells a different story. Every European market in the chart is above the US average on spend per paying user, from Spain at about 1.07 to the Netherlands at about 1.26. Most of the US states shown, including Florida, Georgia, North Carolina, Michigan and Ohio, sit below 1.

Switzerland is the outlier on both axes. Rama writes that its adoption rates are "almost twice the US average for certain paid AI tools in our sample". Across the full portfolio, the chart places Switzerland at about 1.4 times the US average on adoption and about 1.25 on spend.

Businesses: founders level, established firms split

The essay's business section opens by arguing that the US-EU divide is also less stark among companies. Measuring it is harder. Many AI products are served directly through cloud providers, Rama notes, and billing blends AI spend with other cloud spend, which makes AI adoption among existing businesses difficult to isolate when it runs through the hyperscalers.

Stripe uses a different signal: whether businesses used AI tools to integrate Stripe itself. Rama treats this as a more advanced use than chatbots, since it involves agentic coding, a branch of agentic AI in which software agents write and change code. Among businesses that made their first charge on Stripe in September 2026, 22.2% in the US, 22.5% in the EU27 and 19.1% in the UK showed agentic coding signals, according to the chart. Company location is based on Stripe and third-party classification. The essay does not say how agentic coding signals are detected.

"Stripe sign-ups are a strong signal of entrepreneurial activity, and these results suggest that entrepreneurs on either side of the Atlantic are using coding tools in equal measure," Rama writes. On this measure, the EU edges the US by 0.3 percentage points.

Established firms

Existing companies are another matter. Rama acknowledges that entrepreneurs tend to be more innovative than incumbents, and public data, according to the essay, shows more of a gap among established businesses.

Overall business AI adoption is around 20% in both economies. The chart prints 19.4% for the US and 20.0% for the EU27 among firms with 10 or more employees in 2025, drawing on the Eurostat ICT usage in enterprises survey and the US Census Bureau's Business Trends and Outlook Survey. US estimates were adjusted to firms with 10 or more employees to match the EU survey, following a method by Bick and co-authors. The questions differ: Eurostat asks whether an enterprise uses any of a list of AI technologies, while the Census Bureau asks whether a business used AI in any of its business functions in the last two weeks.

The type of use is where the two diverge. As of the first quarter of 2025, only 4% of EU firms with more than 10 employees were using AI in the production of goods or services, compared with 6% of similar US firms, according to Rama. More recent EU figures are not available. US adoption in production, she writes, has climbed to over 8% in the last 12 months, so the gap may have widened further.

By PPC Land's reading of the production chart, the EU27 figure rose from about 2.0% in 2023 to 3.2% in 2024 and 4.1% in 2025, one survey point a year. The US line, built from the more frequent Census survey, begins in the autumn of 2023 at about 2.7%, peaks close to 9% in mid-2025 and ends at about 8.4% around October 2025. The chart does not extend into 2026.

Rama spells out why the gap matters: "The development of new production processes is how technology drives productivity increases, so this gap, if it persists, could be very consequential."

The link from AI use to productivity is not yet visible in US corporate results. On October 24, Apollo's chief economist Torsten Slok argued that AI has produced no measurable payoff outside the technology sector, with S&P 493 profit margins stuck between 9% and 10% since 2023.

What the data does not show

Rama closes with a caveat of her own. "Nothing in this analysis negates the challenges that Europe faces at the AI infrastructure and model layer, nor do we seek to minimize their significance," she writes. Focusing only on those layers, she argues, risks missing what is happening in application development and in business and consumer adoption, where "the US lead is less marked and Europe is less far behind than people may think."

The evidence has limits that the essay leaves open. Everything rests on Stripe's own customers. Consumer AI subscriptions bought through Apple's App Store or Google Play are processed by those platforms, not Stripe, and the essay does not discuss them. It gives no count of AI application companies, no sample sizes, no absolute adoption rates and no list of the 34 consumer tools. Revenue is approximated by payment volume. Consumer location depends on card issuer or payer country. And Stripe earns fees on the payments it measures.

The essay also says nothing about regulation, the explanation for Europe's tech gap that dominates much of the policy debate. PPC Land has covered both sides of that argument: a 2024 study by Columbia Law School professor Anu Bradford found no clear evidence that EU privacy and antitrust rules significantly hurt European firms, while a 2025 NBER working paper found that EU deals led by US investors fell 20.63% after the GDPR became enforceable. Under a provisional agreement reached on May 7, 2026, high-risk obligations under the AI Act moved to December 2, 2027.

Why this matters for marketing

For the advertising and marketing industry, the application layer is the one closest to daily work. Shopping agents, legal assistants and many marketing tools build on models rather than train them. Rama's data suggests European companies at that layer reach early revenue milestones at rates close to American peers, and that the US advantage more than doubles between the $100,000 and $1 million marks.

The consumer findings matter for anyone selling AI subscriptions or advertising around AI products. Fewer Europeans pay, but those who pay spend more per head. Spend per payer is only part of subscription economics, though. RevenueCat found in March 2026 that AI apps earned 41% more per user than other subscription apps but churned 30% faster, with annual retention of 21.1% against 30.7%. Stripe's essay says nothing about how long European or American payers keep their AI subscriptions. Paid use, by Rama's account, is a tiny fraction of the total. Edison Research found in May 2026 that 65% of Americans used AI weekly, up from 52% in February.

There is a further question the essay does not answer: whose tools are Europeans paying for? Rama measures adoption, not the nationality of the provider. In July, France's competition authority found that OpenAI, Google and Anthropic held more than 84% of the AI agent market in May 2026, based on Sensor Tower data. If most of the 34 tools in Stripe's sample are American, European subscriptions add to the revenue of US companies, which would help explain how European consumers can spend more per head while European application companies earn a fraction of US revenue.

Timeline

Summary

Who: Marisa Rama, writing on Stripe Economics, the economics publication of payments company Stripe, using Stripe internal data alongside figures from the IEA, the World Bank, Eurostat, the US Census Bureau, the European Commission and Pew Research Center.

What: An analysis of the AI gap between Europe and the US across four layers. Europe trails badly on electricity prices, data centre capacity and models. At the application layer, US AI companies earn 7x the EU's revenue and 15x the UK's, or 5x and 2x after adjusting for economic size, mainly because the US has about five times as many companies; US firms are 2.2x as likely as EU firms to reach $1 million within 12 months. EU paid consumer AI adoption is about 60% of the US level, but spend per paying user is higher in Europe and the UK. About 22% of new US and EU businesses on Stripe used AI coding tools to integrate, against 19% in the UK, while 4% of EU firms used AI in production in early 2025 against 6% in the US.

When: Today, October 28, 2026. The revenue data covers sign-ups from January 2023 to June 2025, the consumer series runs from 2023 to August 2026, and the coding data covers September 2026.

Where: The United States, the 27 EU member states and the United Kingdom, with country-level data for Switzerland, Denmark, Ireland, Italy, Spain and others, and state-level data for US states including California, New York, Ohio and Pennsylvania.

Why: Rama argues that judging Europe only by infrastructure and models misses the layers where its gap is smaller, and that adoption matters for growth as AI becomes cheaper. For marketers, the findings bear on where AI application companies scale, how much European consumers spend on AI, and how fast established firms put AI into production.