The Bank of Italy Says AI Can Revitalize Italy. But I Have a Problem with the Word 'Potential'

A few days ago, the Governor of the Bank of Italy, Fabio Panetta, read the Final Considerations for 2025, and the part that made headlines is where he states that "artificial intelligence can become a decisive lever for reviving the productivity of the Italian economy". Newspapers headlined this similarly, some with enthusiasm like "AI Can Launch Italy", and up to this point, there's nothing strange; it's their job, and it’s somewhat true. My problem lies elsewhere; I know this discourse almost by heart because I've been hearing it repeated, in different words with the same substance, for years.

In 2013, I spent a day at CINECA in Bologna in front of a machine called Eurora. It was a supercomputer built around NVIDIA's GPUs, the company where I worked at the time, and at that moment, it was the most energy-efficient in the world, leading the Green500 ranking. For those who aren’t familiar, GPUs are the processors designed for running video games that turned out to be perfect for training neural networks, and this is largely why NVIDIA is valued as it is today. We were there in Italy, with one of the most advanced computing infrastructures in Europe. Thirteen years later, Panetta lists among the country's strengths "computing infrastructures among the most advanced in Europe", which is almost the same phrase I had in my head back then, with the difference that thirteen years have passed, and on true adoption, we are more or less at the same point.

I want to say this right away because otherwise it might seem like I'm just contrarian: I agree with Panetta’s diagnosis on several points. However, I have a fundamental reservation regarding the word that recurs throughout his talk and all the reports, which is "potential".

Let's start with the numbers. Panetta says that the share of Italian companies using artificial intelligence has risen to 30%, but only 5% makes "intensive use" of it. This distinction matters much more than it seems. Using AI non-intensively means, in practice, that someone in the office opens ChatGPT to rewrite an email or transcribe a meeting, and then goes back to working exactly as before. Using it intensively means integrating it into business processes, changing how information flows and decisions are made. The first task is now accessible to anyone with a smartphone and a twenty-euro monthly subscription; the second requires months of work and someone who really knows what they're doing, and indeed very few can afford it.

Another number gets little attention in the headlines: 70% of companies state that artificial intelligence has not improved labor productivity as expected, and disappointment remains high, around 59%, even among those who use it intensively. Keep together the 5% that is truly using it and the majority of discontents within that 5%, and the narrative about potential waiting to be unlocked starts to falter.

My explanation, drawn from visiting real companies more than reading reports, is quite prosaic. AI does not produce productivity gains if you try to pair it with a process that is already flawed. If the Monday meeting was pointless before, now you'll have the automatic minutes of a pointless meeting, and you’ll even think you’ve innovated. What really matters is understanding how work actually flows within an organization before integrating any tool, and this is a tedious job of mapping and cleaning processes, removing unnecessary steps, and it should be done before purchasing any license, not after it has already been installed on all the company’s computers.

And here lies the issue, because in Italian small and medium enterprises, this task is usually dumped on the employee who is "good with computers", the one who the company jokingly calls "the office nerd", yet no one removes their existing responsibilities, and no one provides the expertise to redesign a business process. Effectively, you assign them a consulting job while leaving everything else they were doing before on their desk, and the result is as predictable as the 70% of companies that say nothing has changed after a year.

Panetta sees part of this very clearly. He writes that the potential of AI "will not be realized automatically", it depends on how much it enters small and medium enterprises and how well these companies integrate it into their production processes, and he proposes that the State act as the primary client for innovation, steering public demand towards sectors like healthcare or energy. On paper, this is reasonable, and I’m not one of those who scoff at the idea of public intervention as a matter of principle. However, the same discourse contains the admission that struck me the most, the one reminding us that in the 1990s, with information technologies, Italy accumulated delays that slowed productivity for decades, and that this time "there is time to avoid" making the same mistake again. I read that sentence in reverse: if a country already missed a technological train almost identical thirty years ago and has since repeated every spring that it now has all the strengths not to miss it, the honest question becomes another: why has all this potential not translated into much so far?

Then there is the estimate on productivity: with slow AI adoption, labor productivity in Italy would grow by 0.2 percentage points per year, with rapid and widespread adoption of over one point. Labor productivity essentially measures how much value you can produce in the same number of hours, and it’s the variable upon which salaries and the standard of living in a country depend over time. It’s worth pausing for a moment on the figures circulating in the press these days. "La Repubblica" headlines that only one in twenty companies has initiated the revolution, and that "one in twenty" refers to the 5% of companies with extensive AI integration; elsewhere you read that 16% of enterprises have entered "decisively" into the algorithmic age, 4 points below the European average; and in Panetta's remarks, he mentions the 30% that uses it in some form. They don’t contradict each other; they measure different things, which is why when they throw a percentage regarding artificial intelligence at you, it’s always worth asking: percentage of what, and calculated how.

Five major American companies alone control about three-quarters of the world’s computing capacity, meaning the data centers full of processors where AI is trained and run, and Europe is lagging in this race while China is catching up rapidly. Translated: the muscle of the sector is almost entirely in the hands of a handful of companies overseas, and this is a real strategic problem that Italy can do little about alone, and Europe for now talks a lot and decides slowly.

The cybersecurity statistics are quite striking: between 2023 and 2025, cyber incidents affecting Italian banks have increased by 80%. In this area, artificial intelligence is a double-edged sword, as the same models that help defend against threats also know how to find system vulnerabilities faster than ever, and this is one reason Panetta urges banks to invest much more in security and specialized personnel at a time when their finances could actually allow it.

Panetta concludes by stating that the ultimate criterion for success will be the ability to provide opportunities and a future for young people, and he is right, noting that between 2020 and 2024, over one hundred thousand graduates have left. However, the potential of a country is not a number locked in a drawer that someone will eventually open. It consists of people who need the time and competence in the morning to change how they work, and above all a concrete reason to do so, and that time and competence, in the companies I visit, simply do not exist today, and no one is funding them. As long as this remains a discourse from the end of May, Italy’s strengths will continue to be exactly that: strengths, i.e., things that could be useful but are currently stagnant where they are.