Pragmatism and the AI boom

There has been much discussion recently as to whether the current AI boom is actually an AI bubble. With such uncertainty and volatility prevalent in the market, the data centre industry must carefully navigate a clear but challenging opportunity. 

The builders and operators of data infrastructure, and especially data centres, face a range of challenges as they must navigate the rising demand of a technological wave that will be a critical part of Europe’s competitive and sovereign future but one that is also far from predictable. 

Impetus and investment

The current AI boom has seen investment in data infrastructure soar. Data centres designed for AI workloads are projected to require $5.2 trillion in capital expenditure, compared to $1.5 trillion for those powering traditional IT applications. Almost two thirds (62%) of strategy leaders noted an overburdened legacy operating model cannot support current and future strategic objectives and plans, reports Gartner. 

However, even as providers scramble to meet demand, the Moody’s Ratings report of 2024 has warned of the dangers of overcapacity. As a result, hyperscalers are constantly revising capacity plans to cater to uncertain needs. AI infrastructure investment is already greater than other comparable scale outs in the past, increasing risk. The State of European Data Centres 2026 report (SoEDC 2026) found that data centre construction and installation Investments for hyperscale owned facilities would average around €7 billion per year to 2031. The ratings agency warns that existing finance vehicles may need to be adapted to be more suited for AI investments. 

AI fluctuations

Another issue to highlight in this context is that AI technology and adoption has not gone entirely smoothly or predictably. There have been significant instances of overinvestment and deployment that have had to be rolled back as they have not produced the value expected. Examples such as Johnson and Johnson, and various rehirings after AI failures have seen much coverage.

Furthermore, the emergence of DeepSeek, and developments by the Ant Group have shown that upstarts and disruptors can significantly affect market and general perceptions, with implications for share prices, valuations, and roadmaps.

Rhyming history

It is said that while history may not repeat itself, on occasion it rhymes. The data infrastructure industry has navigated previous disruptive waves of technology, with crashes and disruption, and learned lessons from them. 

From the rise of the internet and the client-server model to the development of as-a-service offerings, server virtualisation and the birth of cloud computing, a surge in demand must be met with a pragmatic, systematic response that looks over the lifecycle and lifespan of the required investments, anticipating and designing-in flexibility to accommodate expansion or contraction.

Already, the emergence of neocloud providers, offering bespoke AI and HPC-optimised solutions, such as Coreweave, Global AI, and Nebius, have shown that the industry is agile and adaptable. Forrester and JLL both cite necloud development as a key trend for the near future. More specialist platform and service providers are likely to emerge, in the same manner as specific cloud platforms to support industry verticals and sectors emerged in the recent past. The SoEDC 2026 report confirms that neocloud providers reinforce the AI market momentum. “Their focus on ultra-high-density compute, rapid deployment capability and large power tranches aligns with the needs of AI developers, global model providers and emerging cloud-adjacent platforms,” says the report. 

Pragmatism and experience

From long experience, data infrastructure builders and operators must carefully consider demand and build accordingly. Taking into account the volatility in technological development and operation, they must also work carefully not to stymy development by being too slow or too sparse with capacity. 

Phased and modular builds, with extensive use of prefabricated infrastructure based on reference and pre-certified designs, in conjunction with key partners, from the silicon foundries to server manufacturers and all supporting equipment, ensures that demand can be met in a way that can scale with minimum risk, both up and down. These considerations must be made for ever larger facilities with expected lifespans of up to 20 years.

Power issues

A key issue in this wave of technology is power – AI is power intensive. 

The Moody’s report highlights that AI data centres are increasingly being built as massive campus-style AI factories with 1-5 GW of power capacity. Data centre rack density has doubled since 2016, with new AI installations exceeding 200 kW per rack.

According to the SoEDC 2026 report, “The expansion of IT power supply across Europe highlights the ongoing structural shift toward larger and more power-dense facilities. In 2024, total colocation IT power reached 7.6 GW, with scale sites to grow at an expected 27% CAGR toward 2031.”

Data centres are increasingly under pressure not only to employ low or no carbon power sources to maintain emissions commitments, but also to generate more themselves to relieve stress on grids in constrained areas.

The ability to achieve this generative capacity varies with region and geography. In areas with abundant renewables available for long periods, such as storage augmented wind and solar farms, this is can be a practical approach. In areas of greater population density and more temperate climates, it may not. Southern Europe shows a strong forward trajectory, driven by Portugal, Spain and Italy, which together move from 682 MW in 2024 towards around 5.9 GW by 2031, representing a 36% CAGR.

In other cases, new sources of power are being explored including the likes of small modular nuclear reactors but also hydro, geothermal, wind and wave generation. 

Other considerations

Other key limiting factors include potential lack of resources in the supporting eco system, such as people, skills, and contractors. Environmental impacts must also be considered, especially where not just power but water sources and usage will also figure. 

Many geographies now also face issues with permitting and regulations for these potentially large facilities. This compounds the issue of a lack of standards when it comes to major components in use, such as graphic processing units (GPU), meaning efficiency can be hard to measure consistently at granular levels for reporting and transparency. The industry is going through a transition period of three to five years with many expected iterations before new standards can be set, with questions for financing of the transition. However, this change also leaves opportunities for environmental transition too, due to the pace of change. 

Digital sovereignty

A final point of note in this challenging environment is self-sufficiency and digital sovereignty. 

Increasing geopolitical tensions and economic reordering around the world are pushing many towards greater self-reliance with sovereignty concerns, especially with a technology as potentially powerful as AI. Most blocs regard AI as a competitive advantage and the more that can be done locally, the stronger a footing can be made. The European Union has stated that its approach to AI must be to promote excellence and trust, by boosting research and industrial capacity while ensuring safety and fundamental rights.

The pragmatic approach to scale out for demand ensures that capacity is available where it is needed, and while interoperability is key, it also provides the ability to maintain data and processing power where it needs to be. Europe’s competitiveness and ability to innovate will rely heavily on the type of infrastructure that is built out to support AI. 

While the EU approach has been to regulate while encouraging innovation in AI, it is strongly argued that Europe needs its own native industry for security and competitiveness. The independent European Central Bank has warned that Europe is jeopardising its own future if it misses the boat on artificial intelligence, warning that it must quickly remove obstacles that prevent the diffusion of this new technology.

Conclusion

Despite the challenges of a new, powerful, and disruptive technology, the builders and operators of data infrastructure have gained insights and experiences from previous waves of adoption and development that can inform a pragmatic approach to meeting demand, notwithstanding volatility and unpredictability. 

The current range of tools, technologies, and resources at their disposal means there is greater scope than ever to derisk the scale out needed, while being able to do so sustainably, ensuring competitiveness, digital security and sovereignty. 

Warnings have already come from various quarters of the need for an indigenous AI capability for Europe to thrive in tomorrow’s world, and a solid, commensurate data infrastructure that is AI-optimised will be key to achieving this.