Just a few years ago, competition among states and regions was straightforward: who could attract more investment in data centers. Today, in the world’s most mature markets, the question is no longer whether new data centers should be built, but how they can be developed without placing pressure on power grids, natural resources, and local communities.

An analysis published by Uptime Institute in June 2026—US data center critics pivot from moratoria to regulations, by Rose Weinschenk—shows that the United States is entering a new phase. While many communities have sought to block data center development through moratoria in recent years, the focus is now shifting toward specific regulations that determine who bears the cost of infrastructure and how transparently the impact of these investments must be reported.

The same trend is also visible in Europe.

Why Are These Regulations Emerging?

Artificial intelligence has completely transformed the scale of data centers. New campuses developed for AI workloads consume hundreds of megawatts, equivalent to the energy requirements of entire cities. In many regions, power infrastructure was not designed to accommodate demand on this scale.

As a result, authorities must address a number of legitimate questions:

• Who finances the expansion of the power grid?
• How are residential consumers protected?
• What is the impact on water consumption?
• How are emissions and energy consumption reported?
• What economic benefits remain within local communities?

It is no coincidence that these are precisely the issues dominating the legislative agenda in both the United States and Europe today.

The United States: From Moratoria to Rules

The Uptime Institute analysis reaches a clear conclusion: moratoria are losing ground, while regulations are gaining momentum.

Most U.S. states have recognized that data centers represent strategic investments, generating jobs and significant tax revenue. As a result, few administrations still want to block their development entirely.

Instead, they are seeking to establish clear rules.

Oregon and Wisconsin: Developers Must Pay for Infrastructure

Oregon has adopted one of the firmest approaches to power infrastructure.

New projects must bear the full cost of expanding the power grid, preventing those costs from being passed on to the public.

A similar decision was made in Wisconsin, where the regulator required two data center projects to fully fund the cost of the new power infrastructure. The decision is regarded as a precedent for other states.

Texas and Virginia: Protecting Consumers and Reviewing Tax Incentives

In Texas, the debate is focused on two main areas:

• limiting the impact on energy bills;
• reassessing the tax incentives granted to the industry.

Virginia, the world’s largest data center market, is moving in the same direction. The state is considering reducing certain tax incentives and introducing mechanisms to ensure that new investments contribute more to the development of local infrastructure.

Illinois, California, and New Jersey: Transparency

Other states are taking a different approach.

Illinois, California, and New Jersey are advancing legislative proposals that require operators to disclose information on:

• energy consumption;
• water consumption;
• carbon emissions;
• environmental impact.

The objective is to increase transparency and provide authorities with the data required for infrastructure planning.

Maine: A Rejected Moratorium

There are also examples that illustrate this paradigm shift.

In the U.S. state of Maine, the state legislature approved a moratorium on the development of certain large-scale data centers, but the governor vetoed it.

The argument was straightforward: the investment and jobs generated by these projects are too important to the local economy.

The example is relevant to the industry as a whole: rather than imposing moratoria, authorities are seeking to establish the conditions under which data centers can be developed and integrated into existing infrastructure.

Europe Has Already Been Through This Phase

While the debate is highly visible in the United States, Europe encountered the same challenges even earlier.

Ireland—Power Infrastructure Has Reached Its Limit

Ireland has become one of Europe’s leading hubs for hyperscale providers. Its success has been so significant that power grid operator EirGrid warned that the Dublin area could no longer sustain the same pace of development without posing risks to energy security.

Authorities restricted the approval of new grid connections and introduced additional criteria relating to energy efficiency and the integration of renewable energy sources.

New rules have been introduced since 2025: large data centers must secure 100% of their installed power capacity from their own generation sources and demonstrate that at least 80% of the energy they consume annually comes from new renewable energy capacity.

In March 2026, Pure Data Centres opened Europe’s first off-grid campus. It is not connected to the national electricity grid and operates through its own microgrid, initially powered by natural gas but designed to use biomethane and HVO fuel.

The Netherlands—From Enthusiasm to Strict Rules

The Netherlands has undergone a similar shift.

Following a temporary moratorium on hyperscale projects, authorities introduced strict criteria governing data center locations, energy consumption, land use, and the impact on local communities.

Development continues today, but under far more restrictive conditions than just a few years ago.

Meta’s Zeewolde project remains the best-known European example. Although local authorities had approved the development of the hyperscale campus, public opposition over energy consumption, the use of agricultural land, and limited economic benefits prompted Meta to abandon the investment.

What Do the United States and Europe Have in Common?

Although their approaches differ, the overall direction is the same.

It is no longer enough for a data center simply to bring investment. Operators must demonstrate that they:

• do not pass infrastructure costs on to the public;
• use energy and water efficiently;
• report their environmental impact transparently;
• contribute to the development of local communities.

In practice, the debate has shifted from “whether to approve construction” to “who pays the costs of development.”

Mature markets are approaching saturation. Virginia, Dublin, Amsterdam, and Frankfurt have one thing in common: they have attracted massive investment for more than a decade and are now facing shortages of available land, difficulties in securing power, increasingly complex permitting procedures, and, above all, opposition from local communities.

Romania Has an Opportunity It Should Not Miss

This situation brings Romania’s opportunity back into focus.

Compared with Europe’s major data center hubs, Romania is starting from a different position.

It still offers several important advantages:

• available capacity for infrastructure development;
• competitive costs;
• excellent connectivity to Central and Eastern Europe;
• a strategic location on the Black Sea;
• significant potential to expand renewable energy generation.

In addition, AI-driven demand is creating opportunities for the development of new campuses dedicated to HPC, AI inference, and regional cloud services.

As expanding capacity in Western European markets becomes increasingly difficult, Romania could become one of the natural destinations for the next wave of investment.

But This Opportunity Also Requires Clear Rules

Precisely because Romania is at a different stage of development, now is the right time to define a predictable legislative framework capable of reducing potential friction among investors, utility operators, and local communities.

A modern regulatory framework could provide both investors and local authorities with the clarity required for long-term development.

Such a framework could include:

• clear rules governing connections to the power grid;
• mechanisms through which developers contribute to the cost of new infrastructure;
• reporting standards for energy and water consumption;
• independent environmental impact assessments;
• incentives for the use of renewable energy and the recovery of waste heat;
• transparent mechanisms for dialogue with local communities.

The experience of the United States and Western Europe sends a very clear message: data centers are no longer viewed solely as technology investments, but as critical infrastructure that must be responsibly integrated into the energy and urban ecosystem.

Mature markets have reached a point at which development is constrained by infrastructure and public acceptance. Romania has not yet reached that point, and this represents a major competitive advantage. Romania should not copy the Irish or Dutch model, but instead take advantage of the fact that it is still in a growth phase and can establish clear rules early on.

Climate Risk Becomes a Strategic Factor in the Data Center Industry—Including Europe

The global race for computing capacity is no longer being played out solely on the grounds of power availability, land supply, and permitting speed. An increasing number of digital infrastructure investors, developers, and operators are bringing a fourth decision-making factor to the table—one that until now was treated as a secondary risk: climate exposure.

This is the central message of the latest report published by First Street, a firm specializing in climate risk analysis, titled “18th Risk Assessment – Climate Risk in Global Data Center Markets”. The study analyzes 97 investable data center markets globally and reaches a conclusion that is hard to ignore: climate risk is no longer a “what-if” scenario, but a variable already integrated into the operational performance of current digital infrastructure.

According to the report:

• Approximately 54% of global data center capacity operates in markets exposed to chronic climate stress (extreme heat or drought).
• 79% faces significant acute risks—such as flooding, extreme winds, or wildfires.

Practically, the vast majority of the world’s digital infrastructure is already operating under some form of climate pressure, whether gradual or episodic.

Chronic vs. Acute Risks: The Financial Difference

• Chronic risks erode margins through higher energy and water costs, as well as reduced cooling efficiency.
• Acute risks introduce volatility through downtime, repair costs, and skyrocketing insurance premiums.

Both ultimately reflect on Net Operating Income (NOI) stability and asset financing conditions.

How Risk is Distributed by Region

The report’s data maps out a geographically differentiated landscape of risk:

Region	Chronic Exposure (Heat/Drought)	Acute Risks (Floods/Wind/Wildfires)
Asia-Pacific (APAC)	89% of capacity exposed	60% of capacity exposed
North & Latin America	50% of capacity exposed	86% of capacity exposed
Europe, Middle East & Africa (EMEA)	46% of capacity exposed	25% of capacity exposed

The report identifies specific hubs that sit simultaneously in the high-risk zone across both dimensions. Johor (Malaysia), Singapore, Marseille, and the US state of Virginia are among the markets with the highest combined exposure. Conversely, Nordic cities like Stockholm, Oslo, or Copenhagen remain at the bottom of the risk rankings, benefiting from cooler climates and access to clean energy.

The authors’ conclusion carries direct implications for investment strategies: some of the world’s largest and fastest-growing data center hubs are being built precisely in the areas with the highest climate risk, rather than the safest ones.

Europe: Caught Between AI Ambitions and Climate Goals

Beyond First Street’s data, the climate risk debate directly intersects with a broader discussion gaining momentum in Brussels: Can Europe achieve its technological sovereignty goals in AI without compromising its climate commitments?

Lex Coors, President of the European Data Centre Association (which represents giants like Microsoft, Google, and Amazon before the EU, alongside European players like Digital Realty/Interxion, EdgeConneX, and LCL Data Centers), recently stated that the European power grid is not yet ready to power the next-generation data centers required for AI. Furthermore, Small Modular Nuclear Reactors (SMRs) will not be available in time. In this context, he suggested that the discussion regarding a potential temporary return to gas-fired power plants should at least be opened.

The European Commission aims to triple the bloc’s data center capacity by 2032 as part of its “AI Continent Action Plan”. The issue, according to industry voices, is that the expansion of grids, green energy capacities, and advanced nuclear projects is failing to keep pace with this growth rate.

On the other hand:

• Commission officials have signaled that data centers are only welcome in Europe if they actively contribute to the energy transition—by supporting renewable capacities and waste heat recovery.
• Environmental organizations like Greenpeace EU warn that expanding gas capacity would lock Europe into an energy dependency it is already actively trying to eliminate.

The tension is real and lacks a simple solution for now: competitiveness, sovereignty, sustainability, and speed are four objectives that currently cannot be maximized simultaneously.

And in Romania?

While the discussion on the climate risk of digital infrastructure has already reached public and investment agendas in Western Europe, the conversation in Romania remains much quieter.

Romania was not included in First Street’s analysis—a sign that the local data center market does not yet possess the critical mass of capacity required to appear on the radar of global institutional investors. However, the closest analyzed market in the region, Polonia, has its chronic risk level (heat/drought) evaluated in the 25-50% band. This represents a moderate risk—well below top markets like Johor, Singapore, Marseille, or Virginia, but also above the lowest-risk Nordic markets. Effectively, the Central and Eastern European (CEE) region sits somewhere in the middle of the global ranking—neither in the danger zone nor in the maximum climate safety zone.

In Romania, there is currently no real regulatory or local market pressure forcing the integration of climate risk into data center site selection decisions. Compared to the markets analyzed by First Street—where climate has already become an underwriting criterion alongside power costs and connectivity—the topic remains largely peripheral in Romanian public discourse.

The Corporate Divergence

However, things look different for major market players. International operators and large digital infrastructure companies present locally broadly follow the global agendas of their parent organizations:

• Aiming for near-100% access to green energy.
• Reducing carbon footprints.
• Rigorous ESG reporting.

This is driven less by a locally perceived climate risk and more by global corporate commitments, financing requirements, and European regulations (such as the data center sustainability label currently being developed by the European Commission).

The Bottom Line for Romania: A discrepancy is visible. Physical climate pressure, in the strict sense defined by First Street (extreme heat, drought, floods, wind), is still low compared to markets like Johor, Marseille, or Virginia. However, compliance and sustainability pressure coming from investors and European regulations is already present and will continue to escalate.

In the medium term, as regional data center capacity expands—fueled by growing demand for cloud and AI services—climate risk will likely become highly relevant for the local market. This will not necessarily be triggered by immediate extreme weather events, but because the financing, insurance, and valuation of these assets will increasingly depend on climate standards imposed at the European level.

 

The Data Center industry appears to be “booming” globally, fueled by billions in investment aimed at satisfying the computing needs and cognitive capabilities of major Cloud and AI players. AI “factories” and data centers swallow tens of thousands of chips, placing immense pressure on manufacturers like Nvidia and turning hardware availability into a project slowdown factor. To secure this aspect, Amazon recently signed a contract with NVIDIA to acquire over 1 million processors by the end of 2027.

But is this the primary obstacle facing the industry? Alexandru Chiriță, CEO @ Electrica, offers a very compelling perspective on a potential AI crisis generated not by a lack of silicon, but by a lack of energy and grid access. This is a topic we will further explore at DataCenter Forum 2026.

“The world’s most capital-rich industry is being bottlenecked by the world’s most capital-starved one: the electric grid.”

The conventional framing of AI’s energy appetite treats it as a demand-side problem: build more power plants, string more cables. This framing is dangerously incomplete. What we are witnessing is a structural collision between the tempo of digital capital and the tempo of grid infrastructure. Hyperscalers can site, permit, and build a data center in 12–18 months. A new transmission line takes 7–12 years. A nuclear plant, 10–15 years. This temporal mismatch isn’t a bottleneck to be optimized—it’s a systemic risk that will reshape where AI gets built, which grids get reinforced, and which countries win or lose the next industrial cycle.

The current situation is well-known in the market; even if data centers are not the most energy-intensive sector, they remain the one with the highest growth in consumption.

Current Scale

• Global Consumption: Data center electricity consumption reached approximately 415 TWh in 2024 (~1.5% of global demand). This figure has been growing at 12% per year over the last five years—roughly triple the rate of global electricity demand growth.
• Near-term Trajectory: The IEA projects global consumption will nearly double to 945 TWh by 2030 (equivalent to Japan’s entire national demand). London-based Energy Intelligence estimates demand could reach 1,000 TWh as early as 2026.
• The AI Multiplier: Facilities once designed for 6–12 kW per rack are now being engineered for significantly higher loads. AI training clusters demand dense, concentrated power blocks that must be provisioned upfront—a pattern fundamentally different from the gradual, predictable load growth grids were designed to serve.

Pipeline Pressure

Over 8.9 GW across 105 European projects are targeting operation by the end of 2026. The constraint is not capital, not land, and not even chips—it is grid connection capacity.

Alexandru Chiriță argues that we are at the point “Where Infrastructure Meets Its Limits.” Countries like Ireland, Germany, or the Netherlands are already in this scenario, forced to slow the momentum of their data center industries:

• United States: Demand is projected to reach 260 TWh by 2026. Projects from Virginia to Texas are stalled as power grids approach their limits.
• Ireland: The most extreme case. Data centers already consume 21% of national electricity. A four-year grid connection moratorium (2021–2025) turned the country into a laboratory for what happens when digital demand outpaces physical infrastructure.
• Germany & Singapore: Both are experiencing connection delays or have imposed strict efficiency requirements and moratoriums to manage the load.

The Romanian Context: From Periphery to Strategic Asset

Romania stands at a remarkable inflection point. The same grid constraints stalling hyperscaler projects in Dublin or Frankfurt could become Romania’s competitive advantage—if the country plays its cards right.

Our Advantages Are Real:

• Energy Diversity: Clean sources (hydro, nuclear, wind, solar) already provide over 60% of Romania’s electricity. Nuclear alone contributes ~20%—a baseload asset that hyperscalers increasingly demand.
• Strategic Location: A digital bridge between Western Europe and the Black Sea, supported by an IT workforce of over 200,000 professionals.
• Grid Expansion: Transelectrica’s 10-year development plan (2024–2033) involves ~€1.9 billion in investment. New grid allocation rules for 2026 are specifically designed to accelerate renewable and large-load connections.
• Renewable Pipeline: Over 60 GW of approved renewable projects—a massive reservoir of “clean electrons.”

But Our Challenges Are Equally Real:

• Grid Readiness: Romania’s SAIDI (interruption index) remains around 240 minutes annually, far above the 40–150 minute range of Germany. Hyperscalers demand “five-nines” reliability (99.999%).
• Absorption Capacity: The 60 GW of approved renewables exceeds actual grid capacity. At least 4 GW of storage will be needed by 2030.
• Execution Risk: In 2025, Romania’s data center IT capacity was just 77 MW (compared to Germany’s 4,260 MW). Scaling requires institutional speed, not just capital.

The Strategic Question

Romania is not competing with Dublin for yesterday’s market. It is positioning for tomorrow’s: AI-native infrastructure that needs clean baseload power, land, and regulatory willingness. Saturated countries are now deflecting demand; Romania can absorb it.

The question for Romanian energy leaders isn’t whether data centers will come. It’s whether our grid, our regulatory framework, and our institutional speed can match the tempo of the opportunity arriving at our door.

Read the full text directly on LinkedIn: Link to Alexandru Chiriță’s Article

Europe is rapidly advancing the development of a continent-wide AI infrastructure, and the numbers clearly reflect the scale of this ambition. Following the launch of the AI Innovation Package in 2024, which expanded EuroHPC’s role to full support for AI model development, the EU has already established 13 AI Factories. In 2025, the Commission raised the stakes through InvestAI, a plan aimed at mobilizing EUR 200 billion (mostly private investment), of which up to EUR 20 billion represents the public contribution dedicated to gigafactories.

Together with the Apply AI strategy, Europe is approaching an ecosystem of at least 17 publicly or public-privately funded AI (giga)factories, on top of which private projects are emerging. NVIDIA alone has recently announced a roadmap for building 20 AI factories in Europe, including 5 gigafactories — bringing the company’s active projects on the continent to 22, with another 15 planned by 2030, according to the Centre for European Policy Studies (CEPS). (See the full list of Nvidia’s projects in Europe here.)

At the local level, at the end of November, the Romanian government approved the memorandum mandating the Ministry of Energy and the Ministry of Finance to coordinate the implementation of the Black Sea AI Gigafactory project — an AI and supercomputing hub worth up to EUR 5 billion. The competition for European funding is open, and Romanian authorities are already preparing the official proposal to be submitted in March 2026.

What better moment to revisit the premises of the AI Gigafactory program and assess Romania’s plans and chances?

From AI Factories to AI Gigafactories: Objectives, Funding, and Perspectives

In February 2025, we reported that the President of the European Commission, Ursula von der Leyen, launched InvestAI, the initiative that includes a dedicated EUR 20 billion fund for building AI Gigafactories (AIGFs). The program aims to develop at least five such “gigafactories” — massive compute infrastructures conceived as an extension of the AI Factories program, capable of training models with hundreds of trillions of parameters by integrating over 100,000 AI chips and highly efficient, automated data centers.

Through InvestAI, together with contributions from member states, the EU seeks to provide researchers, startups, and industry with access to large-scale computing capacity, accelerating the development of advanced AI models and strengthening Europe’s technological sovereignty.

The estimated cost of an AI Gigafactory is EUR 3–5 billion, with funding shared between the public and private sectors. More specifically, public contributions can cover up to 35% of CAPEX, financed by the European Commission and the member states, while the remaining investment and 100% of OPEX come from private companies and investment funds. At the European level, funding dedicated to AI Gigafactories is therefore distributed as follows: 17% from the European Commission, 17% from EU member states, and 66% from the private sector.

The AI Gigafactories program is the natural extension of the AI Factories initiative , funded through EuroHPC with nearly EUR 2 billion. AI Gigafactories will build on this model but at a far larger industrial scale, creating the infrastructure needed for training advanced AI models.

• Learn more about RO AI Factory, Romania’s first AI Factory: “Interview with Victor Vevera, ICI Bucharest: What resources and services will RO AI Factory provide, and when will it become operational?”

The Countries Entering the Race for AI Gigafactories. What Comes Next?

In 2024–2025, the European Commission launched a “call for expressions of interest” to assess market readiness. A total of 76 proposals were submitted, ( covering 60 sites across 16 member states, indicating total intended investments of more than EUR 230 billion. Given the scale and complexity of the submissions, the official call for AI Gigafactories (AIGFs) has recently been postponed to 2026, instead of late 2025 as originally planned.

Although the European Commission has not officially disclosed the member states that submitted expressions of interest by the 20 June 2025 deadline—citing confidentiality reasons—several countries have announced their participation publicly. Among them are Romania, Austria (Vienna), Czechia (Prague), Spain (Mora la Nova), as well as Germany and the Netherlands, according to Data Center Dynamics.

Respondents to the call include data center operators, telecom companies, energy providers, European and global technology partners, as well as investors. According to the Commission, collectively, respondents expect to procure at least three million GPUs.

The call was launched to assess interest and create an informal registry of candidates. Between December 2025 and early 2026, the European Commission will engage with respondents to refine the proposals: technical details, location, specifications, funding structure, and sustainability plans.

• On 4 December 2025, the European Commission signed a Memorandum of Understanding (MoU)  with the European Investment Bank and the European Investment Fund to support the development and financing of AI Gigafactories.
• What’s next? The official call for proposals will be launched in early 2026 (estimated submission deadline: March 2026).
• AI Gigafactories are expected to become operational by the end of 2028.

What Are Romania’s Plans for the Black Sea AI Gigafactory?

The Black Sea AI Gigafactory project proposes installing more than 100,000 AI accelerators in two stages: Phase I in Cernavodă and Phase II in Doicești, locations chosen for their energy advantages and digital infrastructure. Powered by up to 1,500 MW of zero-emission energy, primarily nuclear, the gigafactory would position Romania as a strategic supercomputing hub in Europe.

Cernavodă provides direct access to nuclear power, along with excellent fiber connectivity and submarine cable links, while Doicești offers the advantage of an “industrial site with potential for SMR co-location, hybrid cooling, and integration into the national high-speed communications network” (source). The project also includes a strong cybersecurity component, supported by the expertise built around the presence of the European Cybersecurity Competence Centre (ECCC) in Bucharest.

The memorandum mentioned at the beginning of this article—approved by the government on 27 November—places the Ministry of Energy and the Ministry of Finance in charge, with support from the Authority for Digitalization of Romania, to coordinate preparation, negotiations, and the involvement of all key public, private, academic, and research stakeholders to bring and implement the Black Sea AI Gigafactory in Romania.

What else does the memorandum reveal?

• The Ministry of Energy becomes the main coordinator of the project and assumes primary responsibility for data center policy, leading the development and oversight of strategies in this field.
• Romania could attract additional funding through European instruments such as the European Innovation Council Fund, TechEU Scale-up, the EIB Group’s “European Tech Champions Initiative,” or the InvestEU.
• The gigafactory will integrate advanced energy-efficiency technologies—from sustainable cooling and renewable energy via PPAs to heat reuse and grid-resilience measures—to meet national and EU climate objectives.
• Selection criteria also include responsible water usage and the adoption of circular-economy principles.
• Romania is running a major investment program in nuclear energy, including the refurbishment of Unit 1 at Cernavodă (extending its lifetime by another 30 years), the construction of Units 3 and 4, and the development of advanced nuclear technologies such as the NuScale SMR at Doicești.
• The gigafactory will boost competitiveness by providing advanced compute capacity to industry, startups, SMEs, and researchers; create jobs; develop new skills; and, through new cross-border fiber corridors and trusted compute services, strengthen regional connectivity—complementing EuroHPC infrastructure and supporting interoperable collaboration between countries.

We are encouraged that, regardless of the European Commission’s decision after the March 2026 stage, Romania does not plan to pause the project. On the contrary: in the event of a negative response, authorities are prepared to recalibrate the plan, attract investors, and move forward with implementation, with technical support from the World Bank Group and other international financial institutions. Romania is not stepping back from its ambition to build a regional AI hub — and that may be the best industry news as we enter 2026!

The global data center boom, fueled by the generous budgets of hyperscalers and major investors, is steadily pushing construction costs upward. Analyses show that developers are already facing the challenge of working with a “final price” that is impossible to fix from the start. Estimating data centers budgets is becoming increasingly difficult as material prices fluctuate, critical equipment has delivery times exceeding a year, new technologies entail additional costs, and every project stage involves financial risks.

In early November, Turner & Townsend published the Data Centre Construction Cost Index 2025-2026, providing a clear view of the market’s actual state. This is currently the only index exclusively dedicated to data center construction costs, making it an essential reference for investors, developers, and operators. Its analysis helps us understand the directions the market is heading and the implications for future projects.

Liquid Cooling vs. Traditional Data Centers: Understanding the Cost Gap

According to the Turner & Townsend report, costs are rising, and the transition to liquid cooling systems involves significantly higher expenses compared to traditional (air-cooled) data centers. In numbers, the reality looks like this: construction costs for traditional cloud data centers increased by 5.5% in 2025 compared to last year. However, this growth is moderate compared to the 9% jump reported in 2024, indicating that the market is beginning to stabilize. The average construction sector inflation of 4.2% is felt less acutely in the data center segment as local supply chains develop, easing price pressures.

The situation is different for data centers using liquid cooling systems designed for AI workloads. In the U.S., construction costs for these facilities are, on average, 7–10% higher than those of traditional data centers with the same IT capacity. These high-density centers are more complex to build, integrate significantly more expensive technical and cooling systems, and see rapidly increasing demand in markets such as the U.S., the U.K., Europe, and East Asia, as companies seek to support increasingly intensive AI workloads.

Density Pays Off in the Long Run

The same Turner & Townsend report shows that, precisely because of their density, AI data centers can be a more cost-effective choice in the long term. They often feature more flexible designs, which can reduce project costs. Additionally, higher density allows for a smaller building footprint, and “mega campuses” designed to run AI models across multiple interconnected buildings bring significant economies of scale.

On the other hand, traditional cloud data centers require complex measures to ensure service continuity (technical redundancy measures), which considerably increase construction costs and, in the long term, operational expenses as well.

Developers’ Insights

How do developers view the 2025 situation? Nearly half of respondents reported construction cost increases of 6–15%, while 21% say costs have risen more than 15%. Additionally, under inflationary pressure, the majority (60%) anticipate further increases of 5–15% in 2026. About 21% are even more pessimistic, expecting inflation to exceed 15% in the coming year.

Major European capitals are climbing the ranks in data center construction costs, approaching the levels of large U.S. cities. Paris and Amsterdam reach $10.8 per watt, comparable to Portland, while Madrid and Dublin surpass U.S. cities such as Atlanta, Phoenix, and Columbus.

Check out the full Turner & Townsend report here.

Future-Proof Data Centers: Hybrid Designs for AI and Cloud Workloads

According to McKinsey, while AI training currently drives the size and scale of data centers, future facilities will be hybrid, combining training workloads, inference tasks, and cloud operations. These centers could surpass in size even the facilities considered large just two years ago.

Today, the time between requesting services and starting construction of a data center can range from 12 to 36 months, depending on type, design, size, and location. Optimizing the construction process can significantly shorten these timelines: for example, a U.S. architecture firm completed the design of a 929 m² data center in Colorado in just 30 days, using an Agentic AI platform – the first project of its kind entirely designed by artificial intelligence.

Such innovations could reduce construction time by 10–20% and deliver similar capital savings, potentially cutting global projected expenditures of $1.7 trillion by 2030 by up to $250 billion, according to McKinsey analysts.

They also recommend several key directions that could transform how data centers are designed and built, including:

• Designs should allow for phased expansion, modularization, and off-site assembly.
• Prefabricated and modular solutions currently account for 40–60% of data center components, with some projects using up to 80–85%. These solutions accelerate construction, reduce on-site labor, and improve quality.
• Generative scheduling tools allow simulation of thousands of scenarios to optimize resources and work sequences, reducing delivery time by up to 20%.

Discover more in McKinsey’s full analysis.

In short, the data centers of tomorrow will get more expensive unless builders embrace innovative technologies and scalable, modular designs. These approaches help developers cut costs, accelerate delivery, and handle complexity more effectively, ensuring the infrastructure is ready for the rising demands of AI and cloud workloads.

How much power does the infrastructure behind our digital world really consume? From alarmist reports to optimistic forecasts, the energy use of data centers has become a true “gray zone.” Numbers, rumors, analyses, and “myths” abound: some claim that AI will double or even triple global electricity demand by 2030, while others argue that green technologies will balance it all out. The truth? It’s more nuanced than it seems—and lies somewhere in between. In the following lines, we’ll explore it through some of the most relevant recent reports and studies.

The Four-Scenario Perspective of the International Energy Agency (IEA)

According to an IEA report ,  in 2024, data centers consumed approximately 415 TWh, representing 1.5% of global electricity use. Over the past five years, their energy consumption has grown by 12% annually. The increasing adoption of artificial intelligence is expected to double consumption by 2030, reaching around 945 TWh, equivalent to 3% of global electricity use. Between 2024 and 2030, the energy consumption of data centers will grow by 15% per year, more than four times faster than the average growth across other sectors.

AI servers will be the main driver of this growth, accounting for half of the total increase, while traditional servers will contribute 20%. The U.S. and China will generate nearly 80% of the additional demand, with increases of +240 TWh (+130%) and +175 TWh (+170%), respectively, followed by Europe (+45 TWh, +70%) and Japan (+15 TWh, +80%).

However, this represents only the Base Case scenario of the IEA report — the reference or most likely scenario considered by analysts. The agency also outlines three alternative scenarios, reflecting different possible industry evolutions:

• The Lift-Off Case: accelerated AI adoption, more resilient supply chains, and greater flexibility in location and operations. Result: rapid expansion of data centers. Energy consumption: >1,700 TWh in 2035, +45% compared to the Base Case. Global share: 4.4% of total electricity consumption.
• The High Efficiency Case: significant progress in energy efficiency at the software, hardware, and infrastructure levels. Result: the same digital/AI demand met with reduced energy use. Energy consumption: approximately 970 TWh in 2035, savings of >15% versus the Base Case. Global share: 2.6% of total electricity consumption.
• The Headwinds Case: slower AI adoption, local bottlenecks, and strained supply chains leading to delayed data center expansion. Result: stable IT capacity after 2030, limited growth, and increasingly efficient equipment. Energy consumption: about 700 TWh in 2035, <2% of global electricity consumption.

An important note: A March 2025 report by 4E TCP (a program under the IEA), authored by Radu Dudău (President of Energy Policy Group) and Vlad Coroamă (founder of the Roegen Center for Sustainability in Switzerland), shows that in 2023 data centers consumed 300–380 TWh globally (excluding cryptocurrency activity), and by 2030 their consumption is estimated at 600–800 TWh, representing 1.8–2.4% of the global electricity demand. According to the authors, global consumption will not reach 900 TWh in five years due to AI, as this scenario would require investments of USD 9,000 billion in CAPEX alone, plus operational costs. The most realistic scenario involves investments between USD 2,000 and 3,000 billion over the next five years.

The same report also includes a table summarizing most studies estimating the global energy consumption of data centers, as well as a table with studies on the global energy consumption of AI. Read the full report here.

Deloitte and the Impact of GenAI on Energy Consumption

In 2025, data centers account for approximately 2% of global electricity use, equivalent to 536 TWh, according to Deloitte. However, the energy consumption of GenAI data centers is increasing much faster, which could lead to a doubling of global consumption by 2030, reaching around 1,065 TWh. Estimates vary depending on processing and AI efficiency: if optimizations fail to materialize, consumption could exceed 1,300 TWh. Critical consumption for essential components (GPU, CPU, storage, cooling, networking) could reach 96 GW globally by 2026, with AI operations potentially consuming over 40% of that total.

• The annual energy consumption of AI data centers will reach 90 TWh in 2026, roughly ten times more than in 2022. In the first quarter of 2024, global net additional demand for AI reached ~2 GW, marking a 25% increase compared to the previous quarter.
• According to Deloitte analysts, if just 5% of daily global searches used GenAI, the required energy would be 3.12 GWh per day and 1.14 TWh per year, equivalent to the electricity used by approximately 108,450 U.S. households.
• Regionally, data centers will represent 6% of total U.S. electricity consumption in 2026 (≈260 TWh) — the same share as in China. In the United Kingdom, data center energy demand could increase sixfold over a decade, driven mainly by AI adoption.

Access the full Deloitte analysis here.

McKinsey’s Outlook with Europe in Focus

Demand for data centers in Europe is set to surge in the coming years, rising from 10 GW in 2024 to 35 GW by 2030, while the industry’s energy consumption will nearly triple, from 62 TWh to over 150 TWh. According to McKinsey, at this pace, data centers will account for 5% of Europe’s total electricity consumption, up from just 2% today, and will generate 15–25% of the region’s new net electricity demand. The average annual growth rate of electricity use will reach 13% CAGR, and most of the required energy will come from renewable sources, in line with the commitments made by major industry players. Find the full McKinsey report here.

Goldman Sachs Research: Global Data Center Power Demand to Increase by 165% by 2030

Goldman Sachs analysts estimate that global data center power demand will grow by 50% by 2027, and by up to 165% by the end of the decade, compared to 2023. More specifically, demand is expected to rise from approximately 55 GW in 2025 to 84 GW in 2027, reaching a total online capacity of around 122 GW by 2030. The share of AI in the total energy consumption of data centers will reach 27% in 2027, while cloud workloads will account for 50%, and traditional workloads for 23%. The power density of infrastructure will also increase from 162 kW/m² in 2025 to 176 kW/m² in 2027, driven by higher AI-related demand.

In Europe, over the next decade, the data center pipeline—estimated at around 170 GW—could generate a 10–15% increase in the region’s total energy consumption, equivalent to about one-third of its current total. Read the full Goldman Sachs Research analysis here.

How Is Data Center Energy Consumption Evolving at the National Level?

• United States: Data centers used 176 TWh in 2023 (4.4% of total national electricity use), up from 58 TWh in 2014. Energy demand has tripled over the past decade and could reach between 325 and 580 TWh by 2028, representing 6.7–12% of the country’s total electricity consumption. (Source: The U.S. Department of Energy)
• United Kingdom: Data centers use about 2.5% of the country’s electricity, equivalent to 5 TWh per year, and by 2030 their consumption could reach 22 TWh/year. (National Energy System Operator)
• Ireland: Data centers account for roughly one-fifth (20%) of annual energy consumption, and in the Dublin area, they represented almost half (48%) of the city’s total electricity use in 2023. (Commission for Regulation of Utilities)
• Germany: The annual electricity consumption of data centers is about 20 TWh, projected to reach 38 TWh by 2037. Major network operators forecast that by 2045, it could rise further to 88 TWh.
• Netherlands: Data centers consume 3.7 billion kWh, or 3.3% of the country’s electricity, equivalent to 13.32 PJ or 0.44% of the Netherlands’ total energy consumption (3,024 PJ). 90% of this energy is green. (Dutch Data Center Association)
• France: Data center electricity consumption rose from 3 TWh in 2019 to 10 TWh in 2022, representing 2.2% of the country’s total electricity use. It is estimated that by 2050, consumption will increase by 74%, according to Aurora Energy Research.
• Spain: According to consultancy firm DNV, data centers in Spain consumed over 6 TWh of energy in 2024, and this consumption is expected to increase to approximately 12 TWh by 2030 and reach 26 TWh by 2050, representing a growth of more than 300% compared to the current level.
• Meanwhile, in Romania, data centers consumed approximately 1.5 million MWh in 2022 out of a total of around 50.518 million MWh used nationwide. In other words, this represents 2.3–2.5% of the national electricity consumption, up from 2.1% in 2019, according to estimates by Tema Energy.

Aside from these figures, it is also worth mentioning the Energy Efficiency Directive  (revised in 2023), which estimates that in the EU, data center consumption will rise from 76.8 TWh in 2018 to 98.5 TWh in 2030 (+28%), representing approximately 3.2% of the Union’s total electricity demand.

Although it seems we cannot agree on a single set of figures or a single consumption scenario, it is becoming increasingly clear that as digitalization and AI technologies accelerate, data centers are playing a growing role in the national and global energy mix. Efficient energy management and investments in green energy are becoming essential for a sustainable future.

From September 7 to 9, Platform Global 2025 took place in Antibes, France—a landmark summit dedicated to leaders building and financing digital infrastructure worldwide. The event brought together C-level executives from the data center, cloud, edge, and networking sectors, as well as investors, technology leaders, policymakers, regulatory authorities, real estate experts, energy transition specialists, consultants, and analysts, providing a comprehensive forum to explore market trends and growth opportunities.

The program featured over 100 speakers, conferences, and networking sessions covering topics such as AI-driven development, land and energy strategies, long-term energy solutions, mergers and acquisitions, sustainable investments, global geopolitical risks, quantum computing, emerging markets, demand and leasing, cloud sovereignty, how inference is reshaping the edge landscape, and the latest lessons learned by hyperscalers in liquid cooling.

In short, the challenges and outlook for the data center industry can be summarized across four “fronts”: energy, policy (regulation), AI, and sustainability. Energy constraints in the FLAP-D region are driving projects to areas with large available energy blocks. On the policy side, increasingly strict regulations and local issues already emerging in some markets (restrictions and opposition from authorities or local communities) are putting pressure on developers. Large-scale AI deployment is primarily occurring in North America and China, while sustainability remains important, however, outside of Europe, has not yet received the same level of attention.

Below is a summary of some of the most relevant ideas and conclusions for the data center industry from Platform Global 2025.

Investments & Market Growth

According to Nicola Hayes, Director at Platform Markets, global investments in data centers have increased by 53% this year. Tomas Peshkatari (Global Infrastructure Partners / BlackRock) explained that this demand will lead to a doubling of capacity over the next five years, with an annual growth rate of 22%. In addition, contracts have extended to 10–15 years, with investment returns of 8–9%. Against this backdrop, Wes Cummins from Applied Digital highlighted that the U.S. remains ahead of Europe, where finding land and energy for new projects is much more challenging, while China is rapidly building and scaling infrastructure. “The amount of capital we are spending now is enormous—it’s shocking even for me,” said Wes Cummins. Smaller companies like Nebius and Lambada Labs are also growing rapidly, adapting to increasing demand. Charles Antoine Beynet (DataOne) added that while in the U.S. you can deploy 100 MW in a year, in countries like France, the UK, or Germany, the process takes much longer due to limited access to electricity and strict regulations.

Energy Challenges and Solutions

On the energy front, Nathan Luckey (Macquarie Group) explained that energy in Europe is greener but more prone to intermittency, which can lead to blackouts, as recently seen in Spain. Pablo Ruiz-Escribano (Schneider Electric) added that, although energy is available, access takes too long. Building new transmission lines can take up to 10 years, according to Neil Cresswell (Virtus), who noted that, for example, data center power in the UK doubles every five years—from 100 MW to 250 MW, to 500 MW. Richard Bienfait from Stack Emea highlighted that energy costs are much higher in Europe than in the U.S., raising the question, “Why build here?” Sean James (Microsoft) explained that data centers tend to develop in clusters in certain regions, requiring huge amounts of energy, and energy quality can be affected if a large load is suddenly shut down.

On the solutions side, Ash Evans from Google noted that lower-cost locations will emerge as AI monetization evolves, and that the tech giant prefers to control all MEP systems (mechanical, electrical, and plumbing), even though leasing remains an option. Solutions discussed include generators to relieve grid load, batteries (e.g., Tesla Megapacks – xAI DC), and specialized software to optimize energy usage. For example, at Google, on average, only 66% of contracted power is actually used, according to Ash Evans.

Data centers can accelerate energy access when standard grid connections are unavailable using several strategies. According to McKinsey, these include:

• Selecting new locations with shorter connection times (e.g., emerging cities, provinces such as Aragon);
• Leveraging existing infrastructure (e.g., abandoned industrial sites, former coal plants, industrial parks);
• Building on-site generation capacity and microgrids (e.g., gas plants, small modular reactors);
• Accelerating the development of energy capacity together with suppliers (e.g., PPAs, joint grid investments, restarting or extending the life of nuclear/gas/coal plants).

McKinsey’s analysis also shows that data center operators are increasingly exploring behind-the-meter (BTM) options to secure energy directly at the source, including:

• Solar and wind: Highly variable due to intermittency, but aligned with sustainability goals; requires storage capacity and large land areas. A feasible long-term solution (after 2030) if storage costs decrease.
• Hydropower: Seasonal variability (reduced output in dry season), limited construction opportunities near dams, and additional connectivity costs.
• Natural gas: Controlled variability (can be turned off when not needed), proximity to the source is essential to minimize transport investment; long-term sustainability remains debatable, though cheaper Carbon Capture and Storage (CCS) technologies could offset this.
• Nuclear (SMR): High reliability, zero carbon emissions, most promising BTM option, but realistic only after 2030.

Conclusion: BTM feasibility before 2030 is limited, depending on cost reductions in storage and CCS; in the long term, nuclear power is the most likely dominant solution.

Currently, the cheapest on-site microgrids from a Levelized Cost of Energy (LCOE) perspective are based on gas generators. However, in most cases, grid power remains cheaper than microgrids, with differences depending on technology and country. Additionally, partnerships between energy companies and data center operators are becoming more frequent.

According to McKinsey, data center operators can leverage energy company opportunities to generate an additional 2–4% revenue. Practically, data centers become active participants in grid management, not just consumers, contributing to stability and energy efficiency. This can be achieved through several mechanisms:

• Peak Shaving & Load Levelling: Adjusting energy consumption to reduce demand peaks that strain the grid, maximize solar energy capture, and efficiently use capacity during low-demand periods;
• Frequency Regulation: Data centers can rapidly increase or decrease supply to maintain grid frequency (50/60 Hz) during unforeseen events;
• Voltage Control: Managing reactive power to maintain near-constant voltage amid increasing volatility;
• Price Arbitrage: Buying and selling energy based on hourly or minute-by-minute price variations, optimizing costs and avoiding additional peak-hour charges.

Locations, Regulations & Markets

In addition to immediate access to energy and fast construction permits, flexible data center design—allowing adaptation to future demand—is equally important. Eric Boonstra (Kevlinx Data Centers) emphasized that location choice is primarily dictated by clients. For example, cities like Brussels are currently under-supplied and will soon attract new data center investments, while established hubs such as Frankfurt and Amsterdam face energy-related challenges. Milan, although seemingly saturated, has the potential to surpass Amsterdam if the current growth trend continues.

Amine Kandil (N+One Datacenters) sees major future developments in North Africa, particularly Morocco, while Umberto Sordino (EAE) highlighted strong growth in the Nordic countries, Greece, India, and the Philippines. Dan Thomas (GreenScale) noted that the Nordics and Portugal offer cheap energy and the newest networks, although energy access times remain long, they are expected to improve. Oliver Schiebel (hscale) stressed the importance of having multiple high-quality sites for clients to choose from, and Robert Bath (FoundDigital DS) added that 2N redundant power path designs will become increasingly common.

• According to McKinsey, the geographic strategy of data centers in Europe is shifting due to rising AI demand and energy shortages in Tier 1 and Tier 2 cities. Development is moving toward Tier 3 cities, new provinces, and areas around Tier-1 hubs.
• New locations must meet criteria such as energy cost and type, connectivity, latency, proximity to Availability Zones (AZs), and the level of support from the local ecosystem.
• Typical time to access energy: over 5 years in Tier-1 cities, 3–5 years in secondary cities, and 2–3 years in emerging locations.

Design, Flexibility & Cooling Technology

According to experts at Platform Global 2025, although data centers are often developed based on demand, a significant portion is built speculatively due to the importance of time-to-market, making financial risks and design assumptions crucial. Safi Farooqui (Brookfield Asset Management) highlighted the importance of hybrid cooling systems and the construction of high-density data centers. Pablo Ruiz-Escribano (Schneider Electric) noted that the industry has changed completely over the last 30 years: data centers are now part of the grid, white space impacts grey space, and the talent shortage for design, construction, and operation remains a challenge.

Wes Cummins provided an example illustrating the growing importance of flexibility. The first Applied Digital data center used 100% liquid cooling and 25% air-based cooling, whereas now the company designs buildings with a flexible mix of up to 50% air-based cooling. Vincent Barro (Schneider Electric) added that DCIM systems are increasingly becoming true copilots, using AI to automate operations and optimize consumption. The Schneider representative also emphasized the growing importance of microgrid developments. Tom Kingham (CyrusOne) believes that in the near future, data centers will require 600V or even 800V systems, while Andy Hayes (Polar) spoke about the increasing demand for Neocloud and AI colocation, especially in the pharmaceutical industry.

AI Demand & Leasing

AI-focused data centers, such as Neocloud, are already being largely built in Europe. The demand for AI is enormous and real—not just hype—and adoption rates are increasing rapidly, says Christina Mertens (Virtus). According to her, the way data center space is leased has also changed: from flexible contracts of a few years for a few racks or 1 MW in a large facility, leasing has evolved to entire floors or even entire sites being leased by a single client, with longer-term and more secure contracts. The more a data center is customized for a client, the stronger the relationship and commitment become.

• Global demand for data centers is expected to triple, reaching over 180 GW
• By 2030, AI/ML and HPC workloads will account for 71% of the total, while GenAI will grow from 14% in 2025 to 40%.
• According to McKinsey, although significant capacity increases are announced through 2030, supply may remain below estimated demand after 2027. In Europe, the gap could reach 10 GW.
• This gap will put pressure on energy grids in key regions but will also create opportunities for new hubs and the entry of new market players.
• McKinsey estimates that Europe will account for approximately 15% of global workload volume by 2030, much of it driven by AI inference services.

Platform Global 2025 showed that data centers are more than just infrastructure—they’re hubs of innovation, where energy, AI technologies, and smart development strategies will shape who leads the next digital wave.

A data center being offline for just a few hours can mean losses of hundreds of thousands or even millions of dollars. The “Annual Outage Analysis 2025” report by the Uptime Institute highlights a paradox in the industry: although the overall frequency of outages and reported severity have decreased for the fourth consecutive year, their financial and reputational impact is becoming increasingly severe. In 2024, more than half of the surveyed organizations (54%) reported that their most recent significant outage exceeded $100,000, and one in five reported losses of over $1 million.

What have the past few years really looked like for data center operators in terms of operational disruptions? Beyond charts and statistics, every percentage point hides real stories—high-profile incidents and major financial losses that demonstrate just how fragile the balance between availability and downtime can be.

Energy, the “Achilles’ Heel” of the Data Center Industry

Even though only 9% of incidents reported in 2024 were classified as serious or severe—the lowest level ever recorded by Uptime—power remains the “Achilles’ heel” of data centers, accounting for more than half (54%) of major-impact outages. The numbers become even more telling when placed alongside real-world cases—let’s look at a few concrete examples.

In October 2023, a failure in the electrical distribution system at a Microsoft data center in the Netherlands caused an outage of nearly two hours , after the switchover from the public grid to backup generators partially failed. The incident impacted key Azure services—from App Service and SQL DB to storage and virtual machines—and about 1% of racks lost power. Full recovery took until the evening, with some storage accounts affected for several hours, directly impacting customers and critical services that depended on them. Microsoft did not disclose details about the financial impact of this outage.

Read the article “Power cuts, a major challenge for data centers” to explore some of the solutions and preventive measures recommended for operators.

Cooling, Network, and IT – the Next Major Risk Factors

The Uptime Institute report also shows that, after power, cooling systems (13%), networks (12%), and IT systems (11%) follow as leading causes of outages, confirming that critical infrastructure remains vulnerable precisely at the points where it should be strongest.

We already know that heatwaves are not a data center operator’s best friend. Back in July 2022, Google’s and Oracle’s London data centers  were hit by a record-breaking heatwave, with temperatures soaring above 40 °C, which caused failures in cooling systems. Oracle’s first announcement about the incident stated that “unreasonable temperatures” had affected its cloud and network equipment at its South London data center, causing outages throughout the day and impacting customers. Google, in turn, partially shut down cloud services for several hours as a protective measure to prevent equipment damage and prolonged downtime, affecting a small number of users and causing temporary unavailability for services such as WordPress web hosting in Europe.

A more unusual incident was recently shared by Rick Bentley, founder of Cloudastructure and Hydro Hash, which operates a hydro-powered crypto-mining data center. This one occurred in Montana, USA, where the data center “froze solid overnight” . The problem, in this case, was the rapid temperature drop from -6 °C to -34 °C in less than 24 hours. Bentley emphasized that, although the team believed it was prepared, the combination of extreme cold with a power outage made the incident unavoidable.

Complex IT Infrastructures Mean More Frequent Outages

As mentioned earlier, in 2024 nearly a quarter of major-impact outages were caused by IT and network issues—a trend explained by the increasing complexity of infrastructures and the risks associated with misconfigurations. Uptime Institute data confirms this: the most common causes of IT-related outages are network and connectivity problems (30%), IT systems and software (23%), power outages (18%), third-party IT services such as public cloud or SaaS (8%), and cooling issues (7%).

A representative case is the incident on July 20, 2025, involving Alaska Airlines. This highlights that the damage is not only financial but also reputational. The U.S. airline suffered a critical hardware failure in its data centers, which led to the grounding of all flights for about three hours, between 8:00 p.m. and 11:00 p.m. PT. The issue disrupted core flight operations and also affected its subsidiary, Horizon Air. As a result, on July 21, FlightAware data showed that 7% of flights (66) were canceled, while another 12% (110) experienced delays, leading to crowded airports and confusion among passengers. The hardware failure was reportedly caused by a third-party component, with the company stating it was working with the vendor to resolve the issue.

Operational Outages Caused by Human Error on the Rise

In 2025, outages caused by human error increased by 10 percentage points compared to 2024, with the most common cause being failure to follow procedures—possibly amplified by the industry’s rapid growth and staff shortages. Investments in employee training and real-time operational support can help mitigate these risks. According to Uptime Institute, over the past three years, the main causes of major human errors have been, in addition to not following procedures (58%), staff following incorrect processes (45%), understaffing (18%), insufficient preventive maintenance (16%), and omissions in data center design (14%).

A Final Note

As data center infrastructure becomes increasingly complex and interconnected, operational risks are diversifying and becoming more costly. Even infrastructure designed to be robust can be vulnerable to extreme conditions or configuration errors, underlining the importance of an integrated prevention strategy.

To reduce the risk of outages in data centers, several complementary measures are essential: redundant power systems (generators and UPSs) to ensure uninterrupted hardware operation; regular maintenance and testing, supported by monitoring and predictive analytics; failover to mirror sites for rapid traffic redirection; disaster recovery plans with checklists and regular drills; and staff training to minimize human error.

Is Romania joining the AI race? In June, the Romanian Authority for Digitalization announced the country’s bid to host one of the most ambitious European projects in the field of artificial intelligence: Black Sea AI Gigafactory. The project could attract an investment of approximately $5 billion and has the potential to transform Romania into a strategic hub for high-performance computing in Europe.

The Black Sea AI Gigafactory project involves the installation of over 100,000 AI accelerators across two locations: Cernavodă (Phase I) and Doicești (Phase II), which were mainly selected for their energy advantages. The infrastructure will be powered by a sustainable energy mix of up to 1,500 MW, based primarily on low-carbon nuclear energy.

However, this is just one of the opportunities generated by the EU’s AI Factory / AI Gigafactory initiative funded by the European Commission through the European High Performance Computing Joint Undertaking (EuroHPC JU) — the body responsible for developing and coordinating Europe’s high-performance computing infrastructure.

Next, we explore the other opportunities, components, and key elements included in the EU’s strategic approach to AI development.

A “CERN for AI”: What Is the AI Factory / AI Gigafactory Initiative?

AI is becoming a strategic priority for Europe, and through the AI Factory / AI Gigafactory initiative, the European Commission is taking decisive steps toward strengthening the EU’s position on the global tech map. This so-called initiative is, in fact, a broad set of measures with complex objectives: reducing dependence on American and Chinese infrastructures and ensuring free access to computing power for startups, SMEs, and researchers. Instead of a Europe that consumes technology, a new Europe is emerging—one that creates, controls, and democratizes access to high-performance computing (HPC) resources, while fostering a competitive AI ecosystem across the continent.

Through the AI Continent Action Plan , the European Commission is accelerating the development of a strong AI infrastructure in Europe, based on three main pillars: AI Factories, AI Gigafactories, and the Cloud and AI Development Act.

On one hand, AI Factories aim to train and optimize AI models, backed by a €10 billion budget for the 2021–2027 period. AI Gigafactories, four times more powerful than AI Factories, are designed for the development of complex AI models and benefit from €20 billion in funding through InvestAI. In parallel, the Cloud and AI Development Act seeks to boost research in sustainable infrastructure and attract investments, aiming to triple the EU’s data center capacity over the next 5–7 years.

Timeline and context:

January 2024 – AI Innovation Package: The EU adds AI Factories to its list of strategic priorities and launches investments via Horizon Europe and Digital Europe.

February 2025 – InvestAI, a €200 billion program for AI investments, is launched, with €20 billion allocated specifically for AI Gigafactories. Public-private partnerships are strongly encouraged.

April 2025 – AI Continent Action Plan: The EU sets out its roadmap for a network of AI Factories and energy-efficient AI Gigafactories, integrated into major data centers and inspired by the CERN model.

 

What Are the Short- and Medium-Term Objectives?

Between 2025 and 2026/2027, the European Union has set concrete steps to strengthen its AI infrastructure through the AI Factory / AI Gigafactory initiative:

• At least 13 AI Factories will become operational.
• Several AI Antennas—regional access points—will be launched to allow users to remotely connect to high-power AI resources.
• At least 9 next-generation AI supercomputers will be acquired and installed, distributed across various EU regions.
• Up to 5 AI Gigafactories will be created, each equipped with over 100,000 AI accelerators and designed for maximum performance: energy efficiency, high-speed networks, secure supply chains, and AI-driven automation.

The AI Factory Projects Have Been Selected. Which Countries Will Host Them?

So far, EuroHPC JU has selected 13 AI Factory projects in two phases, laying the foundation for the future of artificial intelligence in Europe.

In the first phase, in December 2024, the first seven consortia were designated, bringing together 15 EU member states (including Romania) and 2 participating countries. The AI Factories will be built in Finland, Germany, Greece, Italy, Luxembourg, Spain, and Sweden. The estimated investment is €1.5 billion (EU funding and national contributions).

Here are the selected projects:

1. BSC AI Factory (Spain, Barcelona): An initiative led by Spain, Portugal, Turkey, and Romania. It develops AI infrastructure for industry, government, SMEs, and startups, with a focus on health, energy, and agriculture. The project includes an upgrade to the MareNostrum 5 supercomputer and an experimental platform for new technologies.
2. IT4LIA (Italy, Bologna): AI infrastructure based on the LEONARDO supercomputer, supporting the agri-food, cybersecurity, and manufacturing sectors. Italy partners with Austria and Slovenia.
3. LUMI AI Factory (Finland, Kajaani): Coordinated with five other Nordic and Eastern European countries, this project enables rapid AI solution development.
4. L-AI Factory (Luxembourg, Bissen): Powered by the MeluXina-AI supercomputer, it supports sectors such as finance, space, and the green economy. It offers fast and customized support for companies, with a strong focus on startups and SMEs.
5. MIMER (Sweden, Linköping): A mid-range AI supercomputer with cloud access and scalable storage for sensitive data. It focuses on medicine, materials, autonomous systems, and gaming, and develops AI models for structural biology and personalized medicine.
6. HammerHAI (Germany, Stuttgart): Offers a scalable and secure AI platform for research and industry, with support for machine learning and hybrid HPC/AI. It helps companies access pre-trained AI models.
7. Pharos (Greece, Athens): Uses the DAEDALUS supercomputer to address needs in healthcare, culture, and sustainability, offering end-to-end support for users (from dataset provision to AI model training).

In March 2025, six additional AI Factory projects were selected to be developed in Austria, Bulgaria, France, Germany, Poland, and Slovenia, as follows:

1. AI:AT (Austria) – Hosted at TU Wien in Vienna, this AI Factory will provide companies with access to datasets, advanced AI models, and scalable computing infrastructure.
2. BRAIN++ (Bulgaria) – Located in Sofia Tech Park, it will include the Discoverer++ supercomputer and a dedicated AI hub. The project targets support for startups, the development of large language models (LLMs) for the Bulgarian language, robotics, and AI-based space observation.
3. AI2F (France) – Will leverage France’s existing supercomputing infrastructure, including Alice Recoque, an Exascale EuroHPC supercomputer set to go live in 2026. It aims to support the use of AI in healthcare, energy, agriculture, education, and more.
4. JAIF (Germany) – Built around JUPITER, Europe’s first exascale supercomputer, located at Forschungszentrum Jülich. It will provide integrated access to AI resources and host an experimental platform for developing and testing AI models. The project will support key sectors such as health, energy, education, and culture, and will collaborate closely with AI2F.
5. PIAST (Poland) – Backed by universities in Poznań, Toruń, and the Wielkopolska region, it will use the national supercomputer and the Piast quantum computer. It targets areas such as healthcare, cybersecurity, robotics, space, sustainability, and the public sector.
6. SLAIF (Slovenia) – To be installed in a new data center near the Mariborski otok hydropower plant. Its goal is to foster AI innovation in both business and the public sector, offering training, technical support, and knowledge transfer across the region.

Good to Know:

• May 2, 2025, was the deadline for submitting applications in the third round of AI Factory establishment.
• In April 2025, EuroHPC JU launched a call for proposals for AI Factory Antennas , enabling EU member states to deploy national AI Factory services without investing in their own supercomputing infrastructure.

AI Gigafactories: Next Steps

Another important date is June 20 — the European Commission closed the call for expressions of interest regarding the establishment of AI Gigafactories. A total of 76 proposals were received from 16 member states, covering 60 different locations.

The purpose of the call was to gather early insights from industry leaders, private and public investors, as well as member states interested in shaping the future AI infrastructure in Europe. According to a press release , although these proposals are not considered official applications, they will help the European Commission and member states create a shortlist of potential candidates for developing world-class AI Gigafactory facilities.

Thus, the ground is being prepared for the official project call, scheduled to launch by the end of 2025, with the actual implementation of the infrastructures expected to be completed most likely between 2026 and 2027.

Black Sea AI Gigafactory: What Makes Romania’s Proposal Unique?

Romania’s proposal, developed with the support of international experts from the World Bank, was submitted to the European Commission via EuroHPC JU in the form of a letter of intent to host the Black Sea AI Gigafactory.

The project is backed by a national consortium bringing together public, private, and academic sectors — including companies from the energy industry, consumer goods, advanced technologies, innovative startups, and research institutes. Authorities have announced plans to further expand and strengthen it.

Practically, Romania has expressed the desire to develop a “state-of-the-art AI infrastructure with a hybrid architecture, capable of serving both complex AI training and inference processes within a robust, secure, and sustainable operational framework.” As mentioned at the beginning of the article, this would involve installing over 100,000 AI accelerators in Cernavodă and Doicești, powered by 1,500 MW of nuclear energy.

• Unique competitive advantages of the Romanian project: direct power supply from nuclear sources (Cernavodă), digital infrastructure connected to major European nodes via fiber optics and submarine cables, an industrial site with SMR co-location potential (Doicești), hybrid cooling systems, and integration into the national high-speed communications network.
• Impact: strengthening Europe’s AI capacity, stimulating innovation in Central and Eastern Europe, supporting Ukraine’s digitalization, technologically integrating the Republic of Moldova, and expanding AI services to Serbia and Turkey.

The development of AI Factories and AI Gigafactories in the EU marks a crucial step towards consolidating European technological sovereignty, boosting long-term innovation and competitiveness in artificial intelligence. This evolution will also drive the data center sector, which is essential for supporting digital infrastructure. We hope Romania seizes this opportunity and plays its cards wisely through projects that highlight its strengths.