Lithuania doubled its enterprise AI use to 21%. Denmark is at 42%.

Eurostat published the 2025 enterprise AI numbers on December 11, 2025. The headlines: 19.95% of EU enterprises with 10 or more employees now use at least one AI technology, up from 13.48% a year earlier. Denmark leads at 42.03%. Lithuania jumped from 8.76% to 21.30%, the third-largest annual gain in the bloc after Denmark itself and Finland. We sit roughly at the EU average and roughly half of where Denmark already is.

If you run a service business in Vilnius, Kaunas, or Klaipėda, that’s the market you’re trading inside. The data doesn’t say all of your customers have switched to AI-powered competitors. It says enough of them have started that the curve is now obvious in official statistics.

I read the Eurostat news release, the underlying tables, and the OECD’s December 2025 SME report so you don’t have to. Here’s what they actually say, and what it means for a 5-30 person Lithuanian business.

The 2025 numbers, country by country

Eurostat’s isoc_eb_ai table shows where every member state lands. The leaders, with the EU average and Lithuania for reference:

Country	2024	2025	Change
Denmark	27.58%	42.03%	+14.45 pp
Finland	24.37%	37.82%	+13.45 pp
Sweden	25.09%	35.04%	+9.95 pp
Netherlands	23.06%	33.21%	+10.15 pp
Germany	19.75%	25.97%	+6.22 pp
Estonia	13.89%	23.40%	+9.51 pp
Lithuania	8.76%	21.30%	+12.54 pp
Spain	11.31%	20.27%	+8.96 pp
EU27	13.48%	19.95%	+6.47 pp
France	9.91%	18.16%	+8.25 pp
Italy	8.20%	16.40%	+8.20 pp
Latvia	8.83%	12.21%	+3.38 pp
Poland	5.90%	8.36%	+2.46 pp

Two things stand out.

Lithuania jumped further than everyone except Denmark and Finland. We went from one of the slowest movers in the EU to a country that pulled past France, Spain, Italy, Latvia, and Poland in a single year. Whatever’s driving it — cheap cloud credits, ChatGPT in the office, the GovTech Lab AI sandbox the Innovation Agency piloted in 2024, or a wave of small-business owners reading the same articles you are — it moved faster than the official strategy expected.

Denmark’s lead got bigger, not smaller. They added 14.45 percentage points to an already-leading base. The Nordic countries took the top four spots, all above 33%. Whatever conditions Denmark, Finland, Sweden, and the Netherlands have built — and I’ll come back to what those are — they keep producing wider gaps every year.

Inside those numbers, the SMB story is uglier

The country splits hide the part that matters most for the businesses I work with. Eurostat also breaks the same data down by company size:

• Small enterprises (10-49 employees): 17.00% use AI
• Medium (50-249): 30.36%
• Large (250+): 55.03%

A large company is 3.2 times more likely to be using AI than a small one. The OECD’s December 2025 AI adoption by SMEs report — written for the G7 by Flavio Calvino’s team at the OECD — opens with the same gap and calls it the central problem the report is trying to solve: “The share of large firms using AI (40%) is more than three times that of small firms (11.9%).”

The European average masks all of this. When the headline says 20% of EU enterprises use AI, what you’re actually looking at is a market split into halves — a top tier of large companies racing ahead, and a small-business tier that has barely cleared the starting line.

If you run a service business with 5 to 30 employees, you live inside the small-firm bucket. Your reference point isn’t the 19.95% EU figure. It’s the 17% small-firm figure. Inside that 17%, the firms moving early are quietly pulling away from their slower neighbours.

I wrote about the AI arbitrage window a few weeks ago — the period where early movers capture outsized returns before the rest of the market catches up. The Eurostat data is the cleanest measurement I’ve seen of how wide that window is right now. In Lithuania, four out of five small firms haven’t moved yet.

What Denmark is doing that Lithuania isn’t

The OECD report and the Eurostat methodology don’t directly explain why Denmark crushes everyone. If you read them alongside the European Commission’s State of the Digital Decade 2025 report and the Lithuania country note from October 2025, three structural differences stand out.

Denmark’s small businesses get direct subsidy access. Through programmes like SMV:Digital, a Danish small business can claim cash vouchers to bring in an external consultant for a digital or AI project. Fill in a form, get matched, get reimbursed. The Lithuanian equivalents — Innovation Agency Lithuania programmes that flow through Horizon Europe and Digital Europe channels — are real but slower and harder to reach for a 12-person dental practice. The OECD’s country note for Lithuania confirms there’s no dedicated AI funding envelope. The €160 million in the National Digital Decade Roadmap is for digital skills broadly, not AI implementation by small firms.

Denmark has more skilled people per capita who can implement AI without a dedicated CTO. The OECD’s four-pillar framework for SME AI adoption — connectivity, AI-enabling inputs, skills, and finance — flags skills as “consistently identified by SMEs as one of the main hurdles.” Eurostat’s ICT-specialist indicators put Denmark near the top of the EU on advanced technical specialists per worker. Lithuania has a rising pool — the CAST AI unicorn and the €238 million of 2025 startup funding say so — but most of it is concentrated in Vilnius and Kaunas tech firms, not distributed across the SMB owner base who needs it on a Tuesday afternoon.

The third difference is cultural, and it’s the one Lithuania is closing fastest. Danish business press, Børsen and Finans, has been running practical “what we automated this quarter” pieces for two years. The permission to say “yeah, our front-desk admin uses ChatGPT every day” landed in Denmark before it landed in Lithuania. We’re catching up there too. The +12.54 pp jump shows it. The starting point still matters when you’re measuring market position right now.

What this means if you’re an accounting firm in Vilnius

The high-level numbers compress into one question: are your direct competitors in the 21% or in the 79%?

For most service businesses I work with — accounting, dental practices, auto shops, law firms, restaurant groups, property managers — two or three of your direct competitors have already started using AI in some form. They probably aren’t running self-hosted agents on a Mac Mini. They’re using ChatGPT for client emails, Claude for proposal drafting, an AI receptionist for after-hours calls, or an invoice-extraction workflow that cuts data entry by 70%.

That’s where the gap lives. In daily operational tasks where the cost-benefit math became obvious in 2024, not in some research-grade application a Forbes piece imagined.

A 12-person Vilnius accounting firm that adds invoice extraction and a tax-prep coordination workflow can pull off an 80%-90% reduction in manual data entry inside a month. The €3,500-€10,000 setup and €300-€800/month ongoing, as I broke down separately, pays back inside a quarter on the labour numbers alone, before any new-client lift.

The Eurostat data says four out of five firms your size in this country haven’t done that yet. The OECD data says firms three times larger than yours are doing it at three times the rate. The gap isn’t really between Denmark and Lithuania — it’s between you and the small slice of your direct competitors who already moved.

The compliance footnote nobody quoted from the December release

One thing the headline coverage missed: the same week the December 11 numbers landed, the EU AI Act’s August 2026 enforcement window became visible enough to start mattering. Most of the 21.3% of Lithuanian firms now using AI are doing so without a Data Processing Agreement, without a Transfer Impact Assessment, and without an updated Record of Processing.

The risk isn’t that the Valstybinė duomenų apsaugos inspekcija sends you a fine letter on Monday. The risk is that a client asks you a pointed question about where their data is processed, and the honest answer is “I don’t know.” I wrote up the practical SMB checklist for the AI Act here.

Adopting AI without thinking about compliance is the failure mode that turns the +12.54 pp gain into a regulatory mess two years from now. Adopting AI thoughtfully — with data tier choices, documentation, and a clear human review checkpoint — is how you close the gap to Denmark without the compliance debt.

Where to start if you’re in the 79%

Four moves, in roughly the order I’d take them with a client.

Map what’s already running, and write it down as you go. The first surprise in most discovery calls is that two or three employees are already running ChatGPT or Claude on personal accounts, processing client data through them, and not telling you. That’s both a compliance risk and a sign of where the workflow value is hiding. Map it, then keep documenting every new tool you bring in. Five pages, not two hundred. Future-you in an audit will thank present-you.

Pick one workflow that wastes the most hours per week. Invoice processing, appointment scheduling, after-hours calls, proposal drafting, supplier email triage. Whichever one is eating the most time — that’s the candidate. Don’t pick “the most strategic” workflow. Pick the most boring one with the clearest before-and-after.

Decide your data tier honestly. For most service businesses, a cloud AI provider is fine for non-sensitive workflows and dramatically cheaper than self-hosting. For sensitive client data — patient records, tax filings, attorney-client work — a self-hosted setup or a dedicated GPU server is worth the extra cost and complexity. I wrote out the three tiers and where each one actually fits.

Move now, not in 2027. The 12.54 percentage point Lithuanian jump in 2025 wasn’t evenly distributed. The firms that moved in 2024 are now sitting on a year of refinement. The firms that move in 2026 will be three years behind. The arbitrage window doesn’t close all at once, but it does close, and the closing accelerates as more of your customers learn what to expect.

Want to see where your business sits

I do free 30-minute discovery calls. We look at your specific workflows, figure out which ones would actually benefit from AI, and what the realistic timeline and cost are. If the answer is you’re not ready yet, I’ll tell you that.

The Eurostat data is interesting, but your numbers — your hours per week on data entry, your missed-call rate, your invoice backlog — are what actually decide whether moving makes sense right now.