Why Intense AI Remains Rare: Euro Zone Data Shows Only 7% of Firms Go Deep

Over 70% of euro zone companies now say they use artificial intelligence. Yet just 7% apply it with real intensity. The gap matters. It explains why promised productivity gains stay mostly talk.

Adoption Without Transformation

ECB researchers examined responses from more than 5,000 firms. Their findings, published yesterday, paint a picture of broad but shallow engagement. Most companies dip into AI for routine tasks. Few redesign operations around it. “The intensive use that drives transformation and generates macroeconomic gains remains rare,” the researchers wrote in the Yahoo Finance article.

Smaller outfits lead the way. Younger firms show higher intensity too. Service sectors, especially high-tech and knowledge-intensive ones, outpace manufacturing and traditional industries. Large corporations? They lag noticeably. And this pattern isn’t limited to Europe.

In the United States, similar concentration appears when researchers adjust for workforce size. Scott Wallsten at the Technology Policy Institute created an “AI intensity” metric. It divides an occupation’s share of AI usage by its share of total employment. Scores above 1.0 signal disproportionate adoption.

Writers and media professionals hit 12.3 times their employment share. Life and social science technicians reached 11.0. Social scientists came in at 9.7. Contrast that with healthcare practitioners at 0.2 or K-12 teachers at 0.3. “AI’s current economic footprint is deep but narrow,” Wallsten observed in the Technology Policy Institute analysis.

The numbers tell a story. The top occupations using AI heavily represent only 5% to 7% of the workforce. Elite cognitive roles benefit first. Routine tasks in big employment sectors see little impact so far. This mirrors patterns in the euro zone data. Breadth exists. Depth does not.

But why? Early adopters chase cost cuts and efficiency. They license off-the-shelf tools. They slot AI into existing processes. Results feel incremental. Intensive users think differently. Growth and innovation drive them. They pour money into customized systems. They change workflows, incentives, even organizational structures. The investment gap is huge.

Firms watch competitors too. Peer pressure pushes initial uptake. Yet that same pressure rarely sparks the harder work of true integration. So adoption spreads. Intensity stays confined.

Recent surveys reinforce the point. U.S. Census Bureau data showed enterprise AI usage flattening or declining in some segments by mid-2025. Large firms pulled back as pilot fatigue set in and ROI questions mounted. Only the smallest businesses posted slight gains from very low bases. The initial wave crested faster than many expected. A Medium analysis from last year captured the shift toward caution.

Meanwhile, hyperscalers and AI leaders keep spending. Goldman Sachs Research projects AI-related capital expenditures by key companies could exceed $500 billion in 2026. That infrastructure buildout continues even as enterprise application lags. The disconnect grows.

Executives at large firms often cite integration challenges. Legacy systems resist. Talent shortages bite. Regulatory hurdles in healthcare or education slow progress. Cultural inertia proves stubborn. So they experiment. They generate reports or draft emails. They stop short of overhauling core operations.

Smaller, younger companies face fewer legacy burdens. They build with AI in mind. They move faster. They customize aggressively. Their intensity yields different motivations and different outcomes. The ECB blog noted intensive users focus on innovation while others chase efficiency. The distinction explains much of the performance divide.

And the macroeconomic stakes are real. Economists have hunted for AI’s productivity impact for years. Widespread moderate use produces modest gains at best. Only when a critical mass of firms embeds AI deeply do aggregate numbers shift. That moment still lies ahead. Room for diffusion remains substantial.

Occupational data adds nuance. Postsecondary educators show 5.5 times intensity. K-12 teachers register 0.3. The 18-fold difference within teaching highlights how task type and context shape uptake. Lawyers sit at 0.7. Writers at 12.3. AI complements certain cognitive work. It struggles elsewhere for now.

Wallsten referenced related NBER research. Large language models appear to complement “elite cognitive work” rather than automate routine tasks. The pattern holds across studies. This concentration boosts high-skill, high-wage roles first. Broader labor market transformation will take longer.

So what comes next? Diffusion could accelerate if smaller firms demonstrate clear returns. Larger companies may learn from them. Custom development costs could fall as tools mature. Or the gap could persist. Many pilots already fail to scale. Enthusiasm cooled in 2025. Without measurable gains, boards grow skeptical.

Peer pressure works both ways. When enough competitors achieve real results, others follow with greater seriousness. Until then, most settle for moderate use. They meet the minimum. They avoid disruption.

The ECB researchers struck a measured tone. Their post does not reflect official ECB views. Still, the data speaks clearly. Seventy percent adoption sounds impressive. Seven percent intensity reveals the truth. Transformation remains exceptional. Most firms haven’t committed.

That commitment requires more than licenses. It demands new processes. Fresh incentives. Different skills. Cultural willingness to experiment and fail. Few organizations complete that shift. The ones that do tend to be small, agile, and oriented toward growth.

Watch those firms. Their experience will signal whether AI intensity spreads or stays rare. For now, the evidence says it stays rare. Macro gains stay limited. And the gap between hype and reality persists.

Investors noticed. AI infrastructure spending surges ahead. Enterprise application trails. The two tracks may converge eventually. Or they may not. The next wave of surveys will tell.