AI for Restaurants in 2026: Adoption Is Stuck Near 10%. What That Means If You're Building for Them

Every second pitch deck for restaurant software now has "AI" on the first slide: AI phone ordering, AI menus, AI scheduling, AI reviews. The assumption underneath is that restaurants are about to adopt AI the way offices did. Official surveys from three regions say otherwise. Restaurants are among the slowest adopters of any industry, they are not speeding up, and the reasons they give are not the ones most founders design for.

The adoption number: about one in ten, and flat

The US Census Bureau asks a sample of businesses every two weeks whether they used AI in any business function. For food services and drinking places the answer has barely moved over ten months: 8.5% in November 2025, 7.6% in December 2025, 9.4% in April 2026, 9.8% in June 2026 and 9.1% in September 2026 (US Census Bureau BTOS, 2025–2026).

For comparison, across all US sectors 17.9% of businesses used AI in the November 2025 to February 2026 window (BTOS, 2026). By September 2026 the Information sector was at 49.0%, arts and recreation at 22.6% and construction, not known as a tech-forward trade, at 15.4% (BTOS, 2026). Restaurants sit well below all of them.

Forward intent is low too. In September 2026, 12.0% of US restaurants expected to use AI within six months, and 60.5% said they did not (BTOS, 2026). US hotels, the other half of hospitality, behave differently: their current use jumped from 9.2% in June to 15.7% in September 2026, and 18.9% of US hotels expect to use AI within six months (BTOS, 2026). Same guests, same travel economy, different buyer.

Europe shows the same pattern from a different survey. In 2025, 11.98% of EU enterprises in accommodation and food service (NACE I, 10+ employees) used at least one AI technology, against 19.95% for the whole economy (Eurostat, 2025). The series is rising (3.27% in 2021, 6.09% in 2024) but the sector stays at roughly 60% of the economy-wide rate (calculation from Eurostat data). Hotels alone (NACE I55) were at 21.31% in 2025 (Eurostat, 2025). Eurostat does not publish restaurants separately, but because the combined sector is far below the hotel figure, food service is the part pulling the average down (inference from Eurostat data, not a published number).

Japan, with a very different restaurant culture, lands in the same place. In a March 2026 survey of small restaurants by Japan Finance Corporation, 11.6% used generative AI in some operations, 11.8% were considering it, and 51.1% had no plans (JFC, 2026).

Three countries, three methodologies, one answer: about one restaurant in ten.

Where it is higher: the country spread

Inside Europe the spread is wide, which matters if you are choosing a launch market. In 2025 the share of accommodation and food service enterprises using AI was 25.2% in Denmark, 23.74% in Sweden, 23.72% in Austria and 23.53% in Finland, but 3.95% in Bulgaria, 3.39% in Greece and 1.71% in Romania (Eurostat, 2025). Germany was at 14.63% (Eurostat, 2025).

The Nordic and Alpine markets already run at roughly twice the EU sector average. A founder who needs early reference customers will find them faster there than in Southern or Eastern Europe, where tourism is large but software buying is thin.

Why they don't buy: it is know-how, not price

This is the most useful finding for product design. When EU hospitality businesses that do not use AI are asked why, 69.01% cite lack of relevant expertise (Eurostat, 2025). Only 22.92% of EU non-users say AI is not useful for them, and that EU share has been falling, from 30.28% in 2023 (Eurostat, 2023–2025).

Japan gives a near-identical ranking. Among small restaurants, the top barrier to generative AI is lack of know-how (58.6%), then lack of staff who can use it (42.0%), then doubts about accuracy (40.6%). Implementation and maintenance cost comes fourth, at 23.2% (JFC, 2026).

US data shows the same skills effect by firm size across all industries: among businesses not using AI, 63.3% of firms with 1–4 employees say AI is "not applicable" to their business, against 42.2% of firms with 250+ employees (BTOS, 2025–2026). Most restaurants are small, so they sit at the high end of that curve.

Read together: the restaurant owner is not saying "too expensive" or "useless". They are saying "I don't know how, and nobody on my team does". That is a go-to-market problem, not a model-quality problem.

How much margin is there to pay you

Restaurants are a thin-margin buyer, and this caps what any tool can charge.

Among US-listed restaurant companies with revenue above $50M, the median GAAP operating margin in 2025 was 4.3%, with the 25th percentile at 0.2% and the 75th at 10.9%; 21.2% reported an operating loss, and the median net margin was 1.9% (SEC filings, 2025). Listed hotel and casino-resort companies had a median operating margin of 14.7% (SEC filings, 2025). These are large chains; independents are not in the sample.

In the EU, the gross operating rate of food and beverage service activities (NACE I56) was 11.28% of turnover in 2023, and labour costs took 57.38% of value added (Eurostat, 2023). In Spain the operating rate was 8.99% and labour 61.1% of value added (Eurostat, 2023). In Australia, labour costs were 30.4% of sales in the food and beverage services industry (ABS, FY2025).

The pattern is consistent: labour is the largest controllable cost, and little profit is left after it. A tool that saves labour hours speaks to the biggest line in the P&L. A tool that adds a new monthly fee for a "nice to have" competes for an operating margin that is 4.3% at the median listed US chain.

This also explains what restaurants do say they want from AI. Among Japanese small restaurants using or considering generative AI, 72.1% want it for operational efficiency and 62.5% for flyers and social media promotion; 26.2% mention labour shortage directly (JFC, 2026).

What a working model looks like

The restaurant software company with the most public data is Toast. It had 180,000 locations on its platform in mid-2026, up 22% year on year (Toast Q2 2026 earnings release). Its subscription software carried a 71.8% gross margin in 2025, but its blended gross margin across all revenue lines was 25.9%, because most revenue comes from payment processing (Toast 10-K, 2025). It processed about $1.19M of payment volume per location per year (Toast 10-K, 2025).

The lesson is not "become Toast". It is that the restaurant software business that scaled did so by taking a small cut of a flow the restaurant already had (card payments) rather than by asking for a large new subscription. Delivery platforms work the same way: DoorDash's net revenue was 13.4% of marketplace order value in 2024 (DoorDash 10-K, 2025).

What this means for your product

1. Sell a result, not an AI. With know-how as the main barrier, a product that needs the owner to configure prompts or learn a dashboard is fighting the top objection. Done-for-you outcomes (the phone gets answered, the rota gets built, the social posts go out) remove the skills problem instead of teaching around it.
2. Price against labour hours, and keep fixed fees small. At a median operating margin of 4.3% for listed US chains, the ROI story has to be in hours saved or covers added, and a usage or revenue-linked fee is easier to accept than a large flat subscription.
3. Ride an existing flow. POS, payments and ordering already sit in the restaurant. Distribution through those platforms, or pricing as a cut of a flow, follows the one model the data shows working at scale.
4. Start where adoption is already double. Denmark, Sweden, Austria and Finland give reference customers; hotels, which adopt faster than restaurants in both the US and EU, can be an easier first segment for a hospitality AI product.
5. Do not plan on a fast S-curve. US restaurant adoption moved from 8.5% to 9.1% in ten months while hotels moved much faster (BTOS, 2025–2026). Plan sales cycles and burn for a slow market.

What the data does not tell you

• Definitions differ. BTOS asks about AI use in any business function in the last two weeks; Eurostat counts at least one of several AI technologies over a year and covers only firms with 10+ employees; JFC asks about generative AI specifically. The levels agree, but they are not the same measure.
• Restaurants vs hospitality. Eurostat's AI figures are for accommodation and food service together; there is no published restaurant-only EU number.
• Listed chains are not independents. The margin figures cover US-listed companies with $50M+ revenue. Small independent restaurants are not in that sample.
• No willingness-to-pay data for AI tools. None of these sources measures what a restaurant will pay for an AI product. The margin data sets a ceiling; it does not set a price. Test that with your own customers.
• Short series. The BTOS restaurant series covers about ten months. A flat line over that period is a fact; it is not proof that adoption will stay flat.

The buyer the data describes is slow, thin-margin and held back by skills rather than by price or scepticism. Build for that buyer.