Having grown up in Italian region of Veneto in the nineties, I learned to know the warehouse (“el capanòn” in the local language) as the universal unit of measurement for Efficiency and Prosperity. Did you have a plot of land and were tired of being a farmer? “Put up a capanòn,” which you could then rent out at most. The fever, which began at the end of the eighties—the real economic boom in the region came late—faded irreversibly with the 2008 crisis: the warehouses remained deserted for several years before being demolished or rediscovered by different companies. Some still remain empty, abandoned, like menhirs erected by forgotten civilizations.
In the past two years, Silicon Valley seems to have finally found its warehouse: no longer the mythological garage where startups were born to then conquer the world, but the data center, which upon closer inspection would be an enormous warehouse, full of technology and above all GPUs, preferably those produced by Nvidia, the company that has a monopoly on ultra-fast graphics processing units (or GPUs, precisely), essential for developing large-scale language models. In short, chatbots; in short, artificial intelligences.
Since this story has mainly to do with them, the data centers, it’s worth establishing a precise definition. Data centers are enclosed spaces, often quite large, full of racks—metal stacks slightly taller than an average person that contain the servers that make the digital world we now live in function. From a certain point of view, we could say that these racks are the digital world: when we talk about data clouds, or the cloud, we think mainly of these dark and noisy rooms. To make chatbots like ChatGPT and all the generative artificial intelligences that have sprung up from nowhere in these three years work, however, even more powerful and expensive equipment is needed.
Sam Altman, head of OpenAI, the company that developed ChatGPT and the epicenter of the new gold rush in the sector, confirmed this even in August: it will take thousands of billions of dollars to build all the data centers we need, as he had already announced a few months ago at the White House, presenting the “Stargate” project, made up of private investments. OpenAI is not the only one doing this, quite the contrary. According to consulting firm Gartner, about 475 billion dollars will be spent in 2024 (+47 percent over the previous year) on AI-dedicated infrastructure: Microsoft, Alphabet, Amazon, and Meta alone will invest more than 300 billion dollars. The trend is projected to increase by McKinsey, another consulting firm, according to which this type of spending is destined to exceed 5 trillion by 2030.
To try to give proportion to such large numbers, the Manhattan Project cost the US government the equivalent of about 30 billion dollars today. Pocket change, in short. Unlike the project that led to the construction of the first atomic bomb, however, these efforts are not entirely centralized in the United States, nor are they being carried out in mysterious New Mexico locations: there is no state or region in the world that doesn’t want data centers. Even in Europe.
At the moment, there are 10,511 active data centers worldwide (source: Data Center Map), 3,968 of which are in the United States; the European Union counts just over two thousand (plus the UK’s 479); China, meanwhile, follows with 368. Staying on our continent, however, it’s interesting to observe how much each member state contributes to the total: today Italy counts exactly 200 data centers, less than half of Germany’s (476) and a third fewer than France’s (313). Of these, almost half (84) are in Lombardy, mostly in Milan and surroundings, and only 21 in Rome; some regions are even without any (Basilicata, Molise, Abruzzo) while it’s surprising that Campania, among Italy’s most populated regions, has only five.
At this point the answer would seem simple: let’s build more! But it’s not that simple. Beyond the typically Italian impediments, made up of bureaucracy and lack of both public and private funds, not everyone agrees on the fundamental question: build more data centers… for what? And in what way?
In many cases, pre-existing data centers can be updated and enhanced to make them “AI data centers,” making it pointless to start from scratch with a new one. Not to mention that technological progress has the habit of making technologies more efficient—and therefore less costly for companies and less energy-intensive, for example. The data center race therefore concerns this strange present in which the Future is immanent and has caught us unprepared, naked, without Nvidia GPUs to feed to some language model.
And this is a good thing, given that data centers already represent 2-3 percent of global energy consumption today, a percentage that is destined to increase. Not to mention the water that these centers “drink,” because it’s used to cool the elements in continuous use. In a continent where the danger of drought increases every summer, designing new automated farms destined to subtract water resources from farmers and citizens is not easy.
For this reason too, many companies are investing to make this type of infrastructure more sustainable: in Portugal, the first part of the mega-project SINES Data Campus is being built, a private project worth 8.5 billion euros that wants to make the country a reference point in the region for the sector. The promise is to do it using only renewable energy and exploiting seawater for cooling the machinery. In China they are working on data centers submerged underwater, to solve the problem at its root.
It’s not the only case, obviously: the opening of new data centers brings with it environmental and energy problems, especially in terms of energy and water consumption. For this reason too, Norway (and Scandinavia in general) is moving quickly, exploiting a large availability of renewable energy and rather low average temperatures (even in this case climate change risks ruining the party, as demonstrated by the many days when temperatures exceeded 30 degrees in Scandinavia between July and August).
The new Stargate Norway is part of OpenAI’s project, will open in the north of the country and will be equipped with more than 100,000 Nvidia GPUs. The project is OpenAI’s but its “national” nature is often emphasized in official communications, confirming the need for nations to invest in the future while remembering how much AIs also have military and national security applications. The French government, which last February organized an “AI Summit” to celebrate the Macronist vision of the sector, wants to streamline bureaucracy to build new processing centers, and quickly. And the Italian government too, in a recent bill, has specified that “in choosing suppliers of artificial intelligence systems and models, solutions that guarantee the localization and processing of strategic data at data centers located on national territory may be privileged.” It’s not an obligation, obviously, but a very welcome preference.
In all this, however, the numbers continue to speak clearly. Microsoft’s spectacular growth in recent years, combined with that of a company like Amazon in the previous decade, can be traced to a series of factors. One in particular: the cloud. Both companies, in fact, have a division (respectively Azure and AWS) that deals with digital infrastructure, and therefore servers and racks, and have had unprecedented success. In both cases, as we said, these “clouds” are actually made of… data centers, the traditional ones, which have many concrete and profitable applications, beyond AIs.
Perhaps it would be enough to start with these and then, at most, specialize a part of them for AIs, in case this demand proves to be continuous and prolonged in the future. Andy Lawrence, head of research at Uptime Institute, a company that inspects and evaluates data centers, explained to the Financial Times that betting everything on the densest centers, to be filled with the most expensive chips (but which in two years could be surpassed, who knows), might not be the best solution.
It’s no coincidence that the main push for new data centers comes especially from Jensen Huang, head of Nvidia, the company that designs and sells these very expensive chips, a company that recently broke every record, reaching a stock market value of four trillion dollars. More than Meta, Apple, Alphabet. More than everyone, while it lasts.
