Digital economy

Data centres: the market has tripled in size over the past ten years, driven by AI

According to Frost & Sullivan, the race is being driven primarily by hyperscalers – the major cloud providers – who are stepping up their investment

by Gianni Rusconi

30 October 2026

Aggiungi Il Sole 24 Ore
ai preferiti su Google

4' min read

Translated by AI

Versione italiana

Key points

The big match
Gnanajothi: ‘The key word is affordability’
Managing energy consumption and communicating with the grid
The electricity grid: Europe’s bottleneck

4' min read

Translated by AI

Versione italiana

The global data centre market is already worth over 500 billion dollars and could exceed 1,450 billion by 2035, with an average annual growth rate of over 11 per cent over the period in question. This is according to estimates by Frost & Sullivan, based on the various components of the infrastructure, namely buildings and computing capacity, servers and accelerators, networks and storage systems, power supply, uninterruptible power supplies (UPS), generators, cooling systems and management software. This growth is being driven primarily by hyperscalers – the major cloud operators who are spearheading the advance of artificial intelligence.

The big match

By 2025, the eight largest providers are expected to have invested a total of over 316 billion dollars in data centres, representing an increase of 185 per cent compared with 2023. CapEx spending by the top four – Amazon, Alphabet (Google), Meta and Microsoft – rose from around 86 billion in 2023 to over 280 billion at the end of last year, and Frost & Sullivan estimates that this could more than double by the end of 2026, approaching and likely exceeding the 600 billion mark.

Within the mix of hyperscale investments – that is, capital allocated to the construction and expansion of data centres on a very large scale – AI already accounts for 77 per cent, and experts consider this to be a particularly significant figure.

Geographically speaking, spending remains concentrated in the Americas, which accounted for 46.3 per cent of the total in 2025, with the Asia-Pacific region generating 29 per cent of investment and the EMEA region (Europe, the Middle East and Africa) 24 per cent. According to forecasts by Frost & Sullivan, it is the APAC region that shows the strongest growth prospects for the coming decade (up to the end of 2035), with a compound annual growth rate of 13 per cent for the hyperscale segment and 19 per cent for colocation segment (the EMEA figure stands at 13 per cent), i.e. the service whereby an operator provides other companies with space, racks, power and connectivity.

Gnanajothi: ‘The key word is affordability’

The scale of demand, however, does not automatically translate into ready-to-use capacity. With the increase in workloads linked to the widespread use of artificial intelligence, designing a data centre means synergistically combining computing power and energy availability. ‘Demand is changing,’ confirms Gautham Gnanajothi, Senior Vice-President of the Data Centre Vertical and Global Head of Best Practices Analytics, ‘and it is no longer just a question of how much processing capacity I can install, but how much I can power and cool, and how I can operate the entire system in an economically sustainable way. Cost-effectiveness is the key word.”

AI-related workloads are driving the power requirements of systems beyond traditional levels and putting pressure on the entire data centre infrastructure (including chips, of course), effectively necessitating its redesign.

Frost & Sullivan has identified this evolution as comprising three stages. The first is the upgrading of existing server farms, with the addition of AI capabilities where the infrastructure permits. The second is the construction of modular facilities specifically designed for high-density workloads and equipped for liquid cooling. The third is the ‘native AI campus’, where energy supply becomes an integral part of the design in terms of on-site generation, storage systems and more dynamic management of the relationship with the grid.

Gnanajothi describes this new phase as the ‘industrialisation of the data centre’, emphasising that the priority today is to shift part of the construction work from traditional building sites to repeatable and standardised production processes, because the speed at which computing capacity is brought to market is essential.

Managing energy consumption and communicating with the grid

“This is why,” adds the expert, “modular, prefabricated and factory-integrated approaches are becoming increasingly widespread, for both electrical and mechanical infrastructure, and monitoring and predictive maintenance systems – which are needed to coordinate increasingly complex installations – are playing an ever-greater role.” In short, the scalability of AI data centres depends on the creation of infrastructure capable – at least in the long term – of modulating energy consumption and interacting with the grid, intelligently balancing computational power, energy and deployment times.

The electricity grid: Europe’s bottleneck

Whilst the trend described by Frost & Sullivan is closely linked to the availability of electricity, a study by Roland Berger has calculated that global data centre capacity could almost quadruple, rising from 88 GW in 2025 to an estimated 341 GW by 2035. The use of the conditional tense is necessary because the timeframes for bringing facilities online still pose an obstacle, at least in Europe, where grid connection can take up to seven years, compared with one to three years in the United States. This delay adds further uncertainty to investments, as capital must be committed based on indicators (primarily computing demand and energy costs) that may no longer be in line with market conditions by the time the new facilities become operational. The financial stake, moreover, is already substantial: the five largest hyperscalers (Oracle joins the four already mentioned) considered in the study are forecasting total investments of $755 billion by the end of 2026, and in 2025 they issued over $120 billion in debt – approximately four times the historical average cited by Roland Berger. For Europe, the challenge is therefore to attract new facilities whilst simultaneously accelerating the roll-out of networks and connections. Without an upgrade to the energy infrastructure, the Old Continent risks falling behind just as demand is growing, and the projections for 2030 – when Europe’s share of global data centre capacity could fall to 10 per cent (from the current 13 per cent) – serve as a sort of wake-up call that should not be underestimated.