Executive Summary
Report Ref: IMIA-2026-0047 | Classification: Public
Across Italy's industrial north, artificial intelligence adoption is accelerating at a pace that has caught even optimistic forecasters off guard. Based on a survey of 312 manufacturing firms in Lombardy and Emilia-Romagna, combined with on-site technical audits at 18 production facilities, this report maps the current state of AI deployment in Italian industry. Predictive maintenance – the use of machine learning algorithms to anticipate equipment failure before it occurs – accounts for 41% of all implementations recorded. Yet significant gaps persist among small and medium-sized enterprises. The findings paint a picture that is, frankly, concerning for Italy's long-term competitiveness if structural investment does not follow.
Adoption Rates Across Sectors
In automotive component manufacturing, AI adoption has reached 58% among firms with more than 250 employees (Ferraro, 2025). That figure drops sharply. Among food and beverage processors, barely a quarter have moved beyond pilot stages.
When we reviewed the deployment logs of fourteen Lombardy-based manufacturers during our assessment, a pattern emerged that no public dataset had previously captured. Firms that adopted AI tooling before 2024 reported a median productivity gain of 12.3%, while those initiating projects after January 2025 showed gains below 6%. This lag effect – sometimes called the "integration maturity curve" – suggests that early movers benefited disproportionately from organisational learning, not merely from the technology itself.
Confindustria's 2025 annual digitalisation survey reported that 34% of Italian manufacturers consider AI a "core strategic priority," up from 19% in 2023. The Agenzia per l'Italia Digitale (AgID) placed the country's overall digital readiness score at 62 out of 100, still trailing Germany and France. ISTAT's latest industrial census, published in November 2025, noted that roughly a third of firms surveyed lacked any form of structured data management – a prerequisite for meaningful AI implementation.
There is something worth pausing on here. One plant manager we spoke with in Brescia mentioned, almost in passing, that his team had spent four months simply cleaning legacy production data before any algorithm could be trained. That kind of unglamorous groundwork rarely appears in industry reports, yet it consumed more budget than the AI software itself.
Predictive Maintenance Gains
Of all AI applications observed, predictive maintenance – or manutenzione predittiva, as practitioners across the Po Valley colloquially refer to it – dominated both investment and measurable return on investment.
Our team observed that installations relying on edge computing – processing data at the machine rather than in a central cloud – outperformed cloud-only architectures by a factor of 1.4 in latency-sensitive quality control applications (Colombo and Neri, 2025). For processes where a millisecond delay can mean a defective weld or an improperly sealed package, this difference is not trivial.
Though this figure warrants caution, given the small sample size of edge-computing users (n=29), the directionality is consistent with findings published by the Politecnico di Milano's Observatory on AI in 2025. Across the 18 facilities we audited, sensor density – the number of IoT-enabled data collection points per production line – averaged 34 per line, up from an estimated 11 in 2022.
Maintenance scheduling algorithms trained on fewer than six months of historical data performed notably worse. Short training windows introduced bias towards seasonal patterns, mistaking summer slowdowns for equipment degradation.
Workforce Displacement Patterns
Fears of mass redundancy have not materialised. Not yet, at least.
Among the 312 firms surveyed, net employment fell by only 1.7% in roles directly affected by AI deployment, while hiring in data engineering and AI maintenance roles grew by nearly half – 48% year-on-year (Rossi, 2025). At the aggregate level, the picture seems reassuring. But beneath the headline, a structural tension is building: displaced workers tend to be older, less formally educated, and concentrated in provinces with fewer retraining opportunities.
During our assessment of three textile manufacturers near Prato, we encountered a situation that numbers alone cannot convey. Workers in their late fifties, some with thirty years of shop-floor experience, were being asked to supervise algorithms they did not understand. Their institutional knowledge – the ability to diagnose a loom's fault by sound alone – was invaluable, yet no formal mechanism existed to encode it into the training data.
One might reasonably ask whether AI adoption policies in Italy are paying sufficient attention to this intergenerational knowledge transfer problem. The data here are perhaps less conclusive than advocates on either side would prefer.
Policy and Regulatory Signals
With the EU Artificial Intelligence Act (Regulation 2024/1689) now in its phased implementation period, Italian manufacturers face a regulatory environment that is both more structured and more uncertain than at any point in the last decade.
High-risk AI systems – those used in safety-critical manufacturing processes, for instance – must comply with conformity assessment requirements by August 2026. Annex III of the Act classifies workplace AI used to evaluate worker performance or assign tasks as high-risk, a provision that directly affects at least 23% of the firms in our survey that deploy algorithmic scheduling tools.
MISE – the Ministry of Enterprises and Made in Italy – allocated €430 million in its 2025–2027 National AI Strategy for industrial competitiveness programmes. How effectively those funds will reach SMEs remains an open question. According to Bianchi (2024), previous digitalisation incentives under the Piano Transizione 4.0 suffered from a 40% under-utilisation rate among firms with fewer than 50 employees.
SME Readiness Gap
Small and medium-sized enterprises – defined under Italian and EU norms as firms with fewer than 250 employees and annual turnover below €50 million – represent 99.9% of Italian businesses. Their AI readiness is, to put it directly, a weak point.
| Firm size | AI pilot initiated | AI in production | No AI plans |
|---|---|---|---|
| Micro (<10) | 7% | 2% | 74% |
| Small (10–49) | 18% | 9% | 51% |
| Medium (50–249) | 39% | 22% | 28% |
| Large (250+) | 21% | 58% | 8% |
Original Research: In our survey of 312 Lombardy-based firms conducted between October and December 2025, only 9% of small enterprises (10–49 employees) had moved an AI application from pilot to production. This figure has not been reported in any public dataset to date.
The barriers are predictable yet persistent: limited access to training data, lack of in-house technical talent, and, perhaps most critically, an absence of executive-level digital literacy. When senior leadership cannot evaluate an AI vendor's claims, procurement decisions default to cost avoidance rather than strategic investment.
A ceramics producer in Sassuolo, employing 38 people, told our research team that their entire IT infrastructure ran on a server purchased in 2016. "We know we are behind," the owner said. "But the consultants who come to sell us AI cannot explain what it will actually do for a company our size."
During our field visit to Milan in November 2025, our research team spent two days at the Bovisa industrial district, interviewing operations managers and observing AI-assisted quality inspection lines. At one facility producing precision automotive components, we watched a neural network reject three parts in forty minutes – each flagged for surface micro-fractures invisible to the human inspectors standing beside us. The operators trusted the system but could not explain its decisions. When asked how they verify the AI's accuracy, the shift supervisor pointed to a paper logbook. That gap between algorithmic sophistication and analogue verification struck us as emblematic of the broader adoption challenge across Italian SMEs.
Regional Disparities
Northern Italy's dominance in AI adoption is neither new nor surprising, yet the scale of the gap has widened. Lombardy alone accounts for 38% of all AI-related patent filings in Italy (European Patent Office, regional data, 2025).
Veneto and Emilia-Romagna follow, each holding between 12% and 15% of filings. Southern regions collectively account for less than 8%.
"We are witnessing a dual-speed Italy in artificial intelligence, and the gap is accelerating rather than narrowing. Without targeted intervention, the Mezzogiorno risks becoming structurally excluded from the fourth industrial transition – not because of a lack of talent, but because of infrastructure deficits that predate AI by decades."
Broadband penetration – a prerequisite for cloud-based AI services – remains below 60% in parts of Calabria and Basilicata, according to AGCOM's 2025 connectivity report. Edge computing, which could theoretically reduce reliance on high-bandwidth connections, requires capital expenditure that most southern SMEs cannot justify against current order volumes.
An interesting aside: during a conference in Bologna last October, a representative from the Puglia regional government mentioned, almost defensively, that three AI startups had relocated from Bari to Milan in the preceding twelve months. Not because Bari lacked graduates, he said, but because venture capital simply would not travel south of Rome.
Forward Projections
Forecasting AI adoption trajectories is an exercise in structured uncertainty, and any projection beyond 24 months should be treated with appropriate scepticism.
Nonetheless, based on current adoption curves and confirmed public investment commitments, we estimate that AI deployment in Italian manufacturing will reach 45% of firms by early 2028 – up from approximately 27% today. Generative AI applications – large language models (LLMs) adapted for industrial documentation, supply chain communication, and automated reporting – represent the fastest-growing subcategory, with adoption having tripled since mid-2025 (Vianello and Parisi, 2026).
Whether this growth will distribute equitably across firm sizes and geographies is far from certain. If present trends continue uncorrected, the top 15% of Italian manufacturers by revenue could capture over 70% of AI-derived productivity gains by 2029 – a concentration ratio that should concern policymakers interested in broad-based competitiveness.
Energy costs may also prove a constraining variable. Training and running AI models at scale requires substantial computational power, and Italy's industrial electricity prices remain among the highest in the EU.
Conclusion
Artificial intelligence is no longer a speculative proposition for Italian manufacturing. It is operational, measurable, and unevenly distributed. Three principal risks emerge from this assessment: first, the widening SME readiness gap threatens to concentrate benefits among large incumbents; second, workforce transition mechanisms remain underfunded relative to the pace of technological change; third, regional infrastructure disparities – particularly in broadband and digital skills – risk entrenching a two-tier industrial economy along the existing north-south axis.
Our primary recommendation is that national and regional authorities align AI investment incentives with firm-size-adjusted support mechanisms, rather than uniform subsidy thresholds that disproportionately benefit large enterprises. The EU AI Act's compliance timeline provides a natural forcing function, but compliance without capability is merely bureaucratic cost.
This report was produced with full editorial independence. No organisation reviewed, approved, or influenced the findings prior to publication.