Leonardo Foundation: Italy in the Era of AI - Part 2
In the first episode, we mapped the Italian system in artificial intelligence: a 1.2 billion market in strong growth but split between large companies and SMEs, two supercomputers among the top five in Europe, sovereign language models unique on the continent, and an AI law that makes Italy the first EU country to have one. Real excellence, with a paradox at the center: hardware and cloud remain dependent on foreign countries, and the regulatory advantage is only valid if the implementing decrees arrive. In this second episode, we enter the most concrete fabric: companies, public administration, talent flight, and the roadmap for 2030.
The private ecosystem: verticalizing to survive
The Italian private sector shows contrasting dynamics that the report analyzes with the same precision reserved for the market. The most visible national champions share three characteristics: competitive verticalization in specific niches, attention to the Italian context in terms of language and regulation, and integration between frontier technologies and traditional skills.
Bending Spoons, valued at 11 billion dollars after a 710 million capital increase led by international investors, does not produce artificial intelligence technology: it uses it as a tool to optimize acquired digital platforms. The report is precise in its qualification: it is not a company that develops AI, but a digital operator with a high intensity of AI that has built its success on the optimization of products like Evernote and other acquired platforms. The distinction matters so as not to confuse industrial success with the capacity for sovereign technological development. Domyn, renamed in June 2025, reached unicorn status with a valuation of around 1.7 billion euros, focusing on autonomous technological development for regulated sectors and international cooperation, as demonstrated by the partnership with G42 of the United Arab Emirates. Intesa Sanpaolo, with the CENTAI research center and the AIxeleration program, has explored applications ranging from operational efficiency to algorithmic fairness in credit. Translated demonstrates how automation can enhance skilled work instead of replacing it, with adaptive neural translation that is perfected thanks to the feedback of professional translators: a philosophy that embodies the principle of human-centricity that runs through the entire report.
On the biotech front, the mega-rounds of AAVantgarde Bio (122 million euros, Series B in 2024) and Nanophoria (83.5 million euros, Series A) position Italy as an emerging hub for the convergence between artificial intelligence and life sciences, a dimension often overlooked in media attention concentrated on generative AI but which represents, according to the report, one of the sectors with the greatest growth potential. Overall, the Confindustria report collected for the first time over 240 artificial intelligence use cases already implemented and active in more than 70 companies operating in key sectors, a systematic documentation that allows companies to identify relevant applications and learn from the experiences of others.
Public administration as a laboratory
One of the less celebrated but most significant aspects documented in the report is the role of public administration as an experimental laboratory for responsible artificial intelligence. The Chamber of Deputies is among the first parliamentary institutions in the world to have developed systems based on language models to improve the management of legislative information: in July 2025 it presented three prototypes, Norma for the analysis of legislative production in natural language, the Amendments Writing Machine to assist deputies in drafting, and Depuchat to allow citizens to query official information on parliamentary activity. The total cost, about 500 thousand euros, is contained, but the value is well beyond the economic one: Italy is among the first countries in the world to experiment with artificial intelligence to improve the transparency of the democratic process.
INPS's virtual assistant for the Universal Single Child Allowance reduces waiting times while guaranteeing traceability and the possibility of switching to a human operator at any time. These cases share an emphasis on human-centricity, data protection, and inclusion that the report presents as a distinctive Italian approach to public artificial intelligence. It is no coincidence that the FAIR project is explicitly cited among the initiatives that embody this philosophy. The report recommends the creation of a national repository of public administration use cases, to scale successful experiences instead of leaving them isolated in individual entities.
The geographical distribution of these experiences, however, tells of a country that is still deeply asymmetric. Milan concentrates about 47% of national startup investment operations. Emilia-Romagna, with the Tecnopolo of Bologna and the ecosystem built around the Leonardo supercomputer, shows how the integration between infrastructure, local skills, and administrative solidity can generate scalable solutions: it is the model that the report indicates as a reference for regional strategies. The South remains largely on the sidelines of this transformation, and the report proposes dedicated spokes in the South and two-way exchange programs between Northern and Southern institutions as a structural response to a gap that, if not addressed, risks becoming permanent.
Human capital that flees
World-class supercomputers, sovereign language models, a cutting-edge law, a public administration that experiments. And then there is the datum that scales back optimism with the precision of a thermometer: a salary gap of 40-50% compared to Germany and the United Kingdom in the artificial intelligence sector. A specialized engineer earns in Italy between 30 and 35 thousand euros a year, against 55-60 thousand in Germany. No appeal to national pride stops a career when the difference is this, and the report is not under the illusion that it can.
The phenomenon is analyzed as a "leaky pipeline" that does not only concern salaries. The Milan hub shows a female presence of 30.7% in the tech ecosystem, a relatively positive figure by industry standards, but women occupy less than 14% of senior roles globally. The pipeline loses pieces at every hierarchical level, and no inclusion strategy proves effective if it does not address the structural causes of abandonment, which the report identifies as cultural obstacles, lack of role models, and absence of flexibility policies in the most competitive work contexts.
The barriers for those who would like to return to Italy from abroad are equally documented. The tax regime for "impatriati," or those who have worked or studied abroad for a certain number of years, currently planned for five years, is insufficient to compete with German and British incentives that extend over longer time horizons. The report recommends an extension to ten years specifically for profiles specialized in artificial intelligence, citing the Bending Spoons Fellowship, which guarantees 50 thousand euros a year to ten selected students, as a virtuous example of the private sector that, however, cannot replace structural public measures.
The problem of human capital has deep roots in the training system. The report dedicates a specific section to education at all levels: only 33% of Italian teachers feel prepared to teach digital skills, and the literacy of citizens, not just specialists, remains a serious criticality. Among the companies that have evaluated adoption without implementing it, 59% indicate the lack of internal skills as the main barrier, followed by regulatory uncertainty at 47.3% and data quality problems at 45.2%. The National PhD Program in Artificial Intelligence, with 318 doctoral students funded since 2021, is one of the structural investments in long-term human capital, but the horizon of specialist training is not enough: widespread literacy is needed, from primary school to professional training, to build the collective capacity to evaluate, use, and consciously govern artificial intelligence systems.
Global positioning: differentiate, don't chase
Italy cannot compete frontally with the United States and China at the foundational levels of the artificial intelligence ecosystem, and the report does not hide it. The USA maintains overwhelming leadership with annual private investments exceeding 100 billion dollars and control of the entire technological chain: hardware with NVIDIA dominating the accelerator market, cloud with AWS, Azure, and Google Cloud, foundational models with OpenAI, Anthropic, Google, and Meta, and application platforms. Italy's dependence on these platforms is structural and cannot be eliminated in the short term: the strategy must accept this reality and focus on minimizing risks, capturing value in vertical applications, and building autonomous capabilities in specific niches. China competes with massive state investments, an application ecosystem that counts over a billion users generating data, and significant progress despite hardware restrictions imposed by the USA, as demonstrated by DeepSeek, Baidu ERNIE, and Alibaba Qwen.
The strategy documented in the report is one of differentiation, not direct competition. Priority areas are responsible and human-centered artificial intelligence supported by the regulatory advantage, vertical applications in Italy's sectors of excellence such as precision manufacturing, agrifood, cultural heritage, and healthcare, linguistic and cultural sovereignty with models optimized for Italian and the Mediterranean context, and artificial intelligence for sustainability, from energy transition to regenerative agriculture.
The report also analyzes Asia-Pacific models as partially transferable references. Japan excels in the integration between artificial intelligence and industrial robotics, with the Society 5.0 program offering a framework for human-centered AI that resonates with the Italian approach. South Korea massively integrates artificial intelligence into semiconductors and automotive. Singapore has positioned itself as a regional hub through favorable regulation and attraction of global talent. Israel represents a unique case of AI startup density relative to population, with a model that is not directly replicable, given the different geopolitical context and entrepreneurial culture, but offers insights on mechanisms for technological transfer and connection with the international diaspora.
The most relevant European comparison concerns France and Germany. Mistral AI reached a valuation of 11.7 billion euros post-money in September 2025 after a 1.7 billion round led by ASML, a comparison that makes the different scale of national champions evident. On supercomputing and the regulatory framework, Italy's leadership is real and documented. On the overall size of the market and the startup ecosystem, the gap is equally real.
International cooperation in AI Safety is the field in which regulatory leadership also translates into negotiating weight. Italy actively participates in G7 processes and European AI Safety initiatives, from the 2024 Seoul AI Safety Summit onwards, contributing to the definition of standards that then also bind the domestic choices of companies. It is a participation that is not just diplomatic: international safety standards for artificial intelligence systems are also a barrier to entry for competitors who do not respect them, and Italy has an interest in making them as consistent as possible with its own internal regulatory framework.
The challenge is organizational
The report formulates eighteen operational strategic recommendations, each with measurable KPIs, assigned institutional responsibilities, and analysis of barriers to implementation. Some priorities emerge with urgency for the short term.
The issuance of the implementing decrees for Law 132/2025 by April 2026 is identified as the absolute priority: without the operational guidelines, legal certainty remains incomplete and the regulatory advantage does not convert into a competitive advantage. The Presidency of the Council of Ministers is indicated as the responsible subject. The operation of the 1 billion euro fund in Article 23 requires the definition of allocation criteria for artificial intelligence, quantum computing, and cybersecurity: the resources exist on paper but must be unlocked with clear criteria. Strengthening CDP Venture Capital with a dedicated fund of at least 500 million euros for seed phases is the lever to close the capitalization gap compared to European competitors, with MEF and CDP indicated as responsible. The extension of the tax regime for impatriati from five to ten years for profiles specialized in artificial intelligence, under the responsibility of the MEF, is the structural response to the problem of the salary gap. A strategy to reduce hardware dependence, developed in coordination between MIMIT and ACN within the framework of the European Chips Act, completes the immediate priorities.
The report also introduces an explicit alarm against AI-washing: the risk that public funding is dispersed in the renaming of obsolete legacy systems passed off as artificial intelligence. Funds must be tied to outcome-oriented KPIs, not to the adoption of labels. The incremental resources estimated for implementation in the 2026-2028 three-year period amount to a range between 800 million and 1.2 billion euros: about 500 million for the venture capital fund, 150 million for tax incentives for the return of talent, 100 million for the strengthening of university Technology Transfer Offices, the remaining quota for training, data infrastructure, and institutional coordination.
The objectives are articulated over three time horizons with precise indicators. For 2028: market growth beyond 2.5 billion euros, adoption in SMEs exceeding 30%, issuance of implementing decrees, and operation of the main financing mechanisms. For 2030: market at 5 billion euros, overall adoption between 65% and 70%, positioning among the top ten EU countries, reduction of the salary gap by 25 percentage points, and five Italian AI unicorns. For 2035: systemic integration of artificial intelligence in the production system, public administration, and education, with Italy stably in the European Tier 1 of artificial intelligence.
The cards are there, the game is not
The most honest metaphor to describe the Italian position in artificial intelligence is not that of the latecomer who chases, nor that of the first-born who sets the pace. It is that of a player with a respectable hand of cards who has not yet found the tactical coherence to play it well: excellence exists, but it often remains isolated, doesn't scale, and they don't talk to each other.
Supercomputing infrastructures are among the first in the world, already aligned with future energy standards. Sovereign language models exist and work. A cutting-edge law is in force. Public administration experiments with approaches that set a standard internationally. Globally competitive industrial champions demonstrate that the system knows how to produce excellence. The Leonardo Foundation report, in its documentary meticulousness, provides the most precise map available of where we are and where we could arrive, with the frankness of someone who does not hide the structural difficulties that risk nullifying the excellence: the salary gap, the leaky pipeline of talent, the digital divide between large companies and SMEs, the geographical concentration in the North, hardware dependence on abroad, and the fragmentation of resources that disperses energy instead of creating critical masses.
The challenge, as Floridi summarizes in the conclusion, is no longer technological but organizational and implementative: consolidating the conditions so that the country captures a significant share of the growth of the artificial intelligence market, transforming it from a competitive risk into an opportunity for industrial growth. A compass, indeed. The direction is traced, the cardinal points are clear, the obstacles are mapped with honesty. Deciding to move, and with what speed, is up to the ruling class.
The full report "Italy in the Era of AI. Growth, Challenges, and Perspectives of an Ongoing Revolution" is available for free on the website of the Leonardo Foundation ETS. Edited by Luciano Floridi and Micaela Lovecchio, March 2026.