Green datacentres - Powering the AI era
UnsplashUnsplashUnsplash
· 14 min read
“AI will catalyse an energy transformation.”
David Cahn’s 2024 prediction at Sequoia is already playing out in 2025.
In this new AI race, people no longer talk in dollars - they talk in gigawatts. One gigawatt of compute is now worth around $40 billion, and could reach $50–60 billion with the next generation of chips. This article isn’t about the chips themselves, but about how to power them.
Tech companies have understood that this is now an energy game. The biggest players are securing their own power supply and going vertical, integrating energy production directly into their infrastructure strategy.
The AI boom is creating a new generation of data centres. Packed with dense GPU clusters and advanced cooling, they deliver massive compute and real-time storage, but at the cost of soaring energy demand. Efficiency and continuous high-power supply are now the real competitive edge.
AI compute grows. Who builds it. Who powers it. Who keeps it alive. That’s today’s story.
There has been an immense growth in the need for data centres that has led to an effective buildout of an equal magnitude. Frankfurt, London, Amsterdam, Paris, and Dublin (FLAPD in data centre terms) are driving the growth in Europe. Supply shortages still persist across the continent, especially in core markets like Frankfurt. The demand is so tremendous that preleasing new facilities is very common, and even with increased supply, the vacancy rates (unused capacity) fell further in 2025, to an all-time low of 7.4% from 10% in 2024. Lastly, the net absorption (total new data centre space leased or occupied − Total space vacated) was +300MW, slightly down from +487MW in 2024, hinting at structural bottlenecks leading to supply delays (CBRE Report).
In Europe, the data centre capacity is expected to grow to 35GW by 2030 from 10GW of the current installed capacity (McKinsey). On the same lines, the power consumption is also expected to grow from 62 TWh to more than 150 TWh by 2030. The AI boom has prompted many businesses to experiment with AI internally, which is further pushing the demand.
AI alone is not powering the growth in demand for compute capacity. The resilience of SaaS and other cloud deployments has led to strong spending trends over the years, enabling the demand for data centres to sustain.
What’s more interesting is the fact that the infrastructure for non-AI business applications accounts for a large majority of high-density workload deployment.
Data in Europe legally belongs to the individual who creates it, requiring strict controls over storage, processing, and cross-border transfers under GDPR, unlike in the U.S. or China (see here), which drives a resilient demand for colocation. In 2023, colocation data centres alone contributed €30 billion to European GDP, and this figure will reach €83.8 billion by 2030. A trend towards building more small-scale data centres to meet the needs in each country and away from centralisation is anticipated.
Colocation providers spread the capital cost of infrastructure across multiple tenants and benefit from large-scale operations. At the same time, colocation enables very low latencies and provides an option for redundancies in case one of the data centres fails.
Data centres pose unique challenges to electricity grids compared to other types of energy users, and already represent a few % of demand (nearly 20% in Ireland). They represent large, localised loads which typically operate continuously and can ramp up their operations faster than most large energy users. Typically, they maintain a consistently high load factor (ICIS and IEA reports 1 and 2).
Demand for uninterrupted, clean, and reliable power is now the biggest bottleneck for data centres in Europe. Power constraints are forcing aggressive pre-leasing and extending construction timelines to 2027 and beyond. Data centre capacity growth slowed to 7.2% in 2025, down from 20%, primarily due to power challenges, with Amsterdam adding no new capacity. In Italy, power connection requests rose to 42 GW in 2025, from 30 GW in 2024. Across Europe, operators are seeking sites farther from city centres. In Ireland, Dublin imposed a de facto moratorium on new data centres until 2028.
Hybrid storage arrangements that combine renewables and batteries in a single contract are gaining traction, particularly among energy-intensive industries and data centres seeking 24/7 energy matching (Wood Mackenzie). Additionally, representing over 70% of the regional market, corporates continued to dominate the renewable PPAs. More specifically, the technology and data sectors were the primary drivers of offtake activity in 2024. A tailored combination of PPAs and on-site generation and storage is necessary for an uninterrupted energy supply and workload processing.
This points to an increased demand for micro-grids and self-production capacities in addition to BESS solutions becoming an integral part of the data centre ecosystem.
Data centres globally consume 560 bn litres of water annually, mainly for cooling processing units, with projections reaching 1,200 bn litres by 2030 (source). Traditional air-cooling systems emit waste heat at 25–35°C, whereas liquid cooling can reach 50–60°C. Adopting closed-loop or direct-to-chip liquid cooling can reduce water use and improve energy efficiency by at least 15% compared with air cooling.
The thermal byproduct, often vented into the atmosphere, presents a major reuse opportunity. According to the IEA, waste heat from data centres could supply up to 10% of Europe’s space heating demand by 2030. Some initiatives are already underway: Microsoft’s Finnish data centres supply waste heat to Espoo and Kauniainen, and Meta’s Odense facility warms thousands of homes.
On the supply chain inputs, more than 60% of the embodied carbon in data centres comes from the equipment. LCAs show that backup generators are a large contributor to embodied carbon emissions, representing roughly 40% of the embodied carbon related to equipment in the data centres. HVAC equipment is also a key driver for embodied carbon in data centres.
On the output side, industry press releases show that even with improving technology, the depreciation schedules of GPUs and other accelerators have not improved, primarily due to rising workloads. Additionally, even though Big Tech accounting considers about 5 years as useful life for GPUs, in practice, they only last for about 1-3 years with current utilisation rates (which are on an upwards trend themselves). This has been greatly increasing the e-waste volumes.
Source: Data Centre Uptime Institute
Power and cooling remain major operational disruptors. Outages are costly: 54% exceed $100,000, and 20% surpass $1,000,000 in direct, opportunity, and reputation costs. Factors such as inflation, SLA penalties, labour, and hardware replacement have driven these costs higher in recent years. As reliance on digital services grows, the impact of outages is becoming broader and more consequential.
Outsourcing power to the grid, now that is so critical, does not seem to be a sustainable option.
Cooling, which is the second largest disruptor, is a crisis looming over the data centres that are being upgraded. Over the last three years, the rack density (amount of power consumed by all the IT equipment in a server rack) has seen a significant upward trend. Lower-density racks are falling out of favour, and the average rack density is moving higher each year. This leads to higher heat densities and consequently requires more efficient cooling systems. While the average rack densities hover around 8-10 kW for data centres, AI data centres use 150kW per rack as standard, and estimates expect it to rise to close to 500 kW per rack by 2027.
We expect to see higher cooling-related outages if developments are not made in the data centre operations.
Source: Data Centre Uptime Institute
Power Usage Effectiveness (PUE) measures a facility’s energy efficiency as total facility power divided by IT equipment power. While innovative designs are improving efficiency and shaping expectations for the next five years, their impact on average PUE is limited due to the large number of aging legacy facilities worldwide. Data centre designs have not yet reached physical efficiency limits, nor have they been standardised across the industry.
Advanced solutions, architectures, and services, especially in cooling, heat recovery, and software-based energy efficiency optimisation, become necessary to make data centre operations sustainable.
Private capital and REITs are pouring money into infrastructure, and tailored business models are popping up within four major segments - advanced cooling techniques, waste heat recovery & reuse, energy use optimisation, and data centre design & operations.
Brookfield has pledged up to $10 billion for data centre build-out in Sweden, while Microsoft plans to invest nearly $40 billion in new data centres across Norway. These moves align with the global strategies of private equity firms, which are increasingly building their theses around growth driven by hyperscalers (CBRE).
The new generation of compute is creating a strong pull effect - driving the emergence of new technologies designed to meet this massive computing demand and efficiency challenges.
This category works directly with the hardware layer of the data centres, the chips, and the racks. This can be done in two ways: software and physical methods. Software methods involve work scheduling, autoscaling, parallelisation, or other techniques to reduce workloads, while physical methods involve cooling via immersion techniques, microfluid channel architectures, or even air cooling.
Common business models in this segment are:
• Hardware sales + maintenance: The startup sells cooling systems and charges a recurring maintenance or operating fee. Example: Ionic Wind
• Performance contract: The startup guarantees PUE or energy savings, and the customer pays for measured outcomes. Example: Etalytics
• Optimisation SaaS: The startup charges a subscription fee for monitoring, predictive maintenance, and heat-routing optimisers. Example: Wattdesign
This category has an overlap with the cooling technologies, with the core differentiator being capturing the heat and then repurposing it for various other applications.
Common business models in this segment are:
• Offtake agreements: The startup captures low-grade heat, upgrades the temperature with heat pumps, and sells heat under long-term offtake contracts to district heating companies. Example: Nohewa
• Embedded heat reuse: The startup provides water-cooled servers or radiator-style units that host compute and deliver heat to buildings. Example: DeepGreen
This category involves startups that provide services for enabling clean energy, usage optimisation, and storage and trading services.
Common business models in this segment are:
• SaaS: Software that monitors, predicts, and optimises data-centre energy use for a recurring subscription fee [Zendo]
• Digital Twins: Provide digital models and sensors used to reduce cooling energy, sold as licensed analytics software [EkkoSense]
• Carbon-aware workload routing: Software that dynamically shifts computing to lower-carbon or cheaper power sources, earning SaaS or savings-based fees [Helio]
This category comprises the operators of data centres and providers of hardware for easy integration.
Common business models in this segment are:
• Colocation/wholesale hosting: Lease racks, cages, or full halls on €/kW/month contracts. Example: atNorth, Green Mountain
• Build–own–operate (BOO)/hyperscale operator: Raise project capital, build large AI/hyperscale campuses, and sell long-term hosting/power. Example: Evroc, EcoDataCenter
• Build-to-suit: Design and deliver bespoke facilities for a single customer on multi-year contracts. Example: Latos
• Modular/containerised deployments: Prefab modules or mobile enclosures sold, leased, or operated close to power sources. Examples: Chainergy, Modul
Data centre technologies are a high-risk, high-reward play that is currently receiving a lot of interest through greater focus on AI. But these technologies are capex-heavy, and their consumers are asset-light industries with changing behaviours. Additionally, the product life cycles in this industry are growing at an immense pace, which further increases the risk of tech obsolescence.
The AI bubble - Many acknowledge that we’re in an AI bubble, at least when it comes to stock market valuations and superstar developer pay packages. However, the big difference with the cloud era is that back then, not many people truly believed the cloud would be a game changer - whereas today, almost everyone sees AI as a historic technology shift. The AI bubble may burst in the short term, but the underlying demand will remain over the long term.
Power or heat itself is a commodity — the real money lies in high-value services and chips. The real value now comes from the technical expertise required to deliver reliable, efficient onsite heat, power, and cooling, in a bundled energy-as-a-service scheme.
Tenant concentration & vertical integration by hyperscalers - If hyperscalers build their own facilities or renegotiate existing terms, demand for third-party enablers could decline.
Overheating of demand & supply-chain risk - Extended lead times for chillers, transformers, and PCBs are driving up build costs and delaying revenue realisation.
Overestimation of demand - A recent report from JPMorgan Report assessed that to deliver a modest 10% return on the AI buildout, $ 650 billion in annual revenues in perpetuity is necessary from AI firms. Thats equivalent to a $ 35/month from every iPhone user or a $ 180/month from every Netflix subscriber in perpetuity! This could drive an overcapacity or a dark data centre situation in case revenue projections do not materialise in the future.
Blended finance drives Europe’s green/hyperscale data centres. Equity funds tech and roll-out, while debt, loans, EIB support, guarantees, and grants cover capital-heavy sites. Sustainable debt in the form of green bonds and securitisations is growing as operators seek low-cost, long-term ESG-linked capital. Market momentum for green bonds stayed strong in 2024–25. Strategic corporate capital and M&A provide both financing and exit paths.
Hardware:
Typically, VC equity + occasional convertible notes for pre-seed → Series A. Hardware scaling often uses mezzanine/growth debt later. Ex: Corintis raised a Series A led by BlueYard.
For later stages (Series B+), the common way is to raise equity + venture debt + strategic investors (corporates). While transitioning to production, supplier credit + bank facilities are common, and if revenue is predictable, venture debt or receivable financing becomes important.
Data centre developers/operators:
This segment is heavy on project financing. A combination of equity (infrastructure VCs/PE /strategic), senior bank debt, EIB or national promotional bank loans, possible green bond issuance or securitisation for stabilised cash flows comes into play. Grants and guarantees (InvestEU/EIB) also go on to reduce the effective cost of debt. Evroc illustrates this through its recent fundraise that included VC equity, debt, and grants.
Asset-light/software-based:
Equity is the major source of capital for asset-light models with a lot of corporate VC participation, especially in the growth stages.
Some notable examples:
• Submer 🇪🇸 [€50M - Growth - Oct 2024] led by M&G’s Catalyst impact strategy (~$55.5M / €50M); participation from Planet First Partners, Norrsken, Mundi Ventures
• Evroc 🇸🇪 [€50,6M - Growth - Mar 2025] led by Blisce/with Giant & EQT Ventures participation. The company explicitly plans to combine further venture capital + project debt + grants to reach project finance needs
• Corintis 🇨🇭 [€24M - Series A - Sep 2025] led by BlueYard and others. It came with strategic board appointments (Intel CEO Lip-Bu Tan), indicating likely strategic partnerships and potential for customer-led co-investment or later strategic exit
• etalytics 🇩🇪 [€16M - Series A - 2025] led by M12 (Microsoft’s VC) and others, highlighting the trend of corporate VC leading rounds in energy optimisation for data centres
• FläktGroup 🇩🇪 [€1,5B - Exit - May 2025] Samsung acquired FläktGroup from Triton PE, demonstrating strategic industrial buyers paying premiums for proven cooling/HVAC tech for data centres.
Data centres have become the digital backbone of modern society, powering cloud computing, artificial intelligence, and the vast array of services that drive both business and everyday life. As the world becomes increasingly data-driven, the scale and sophistication of these facilities continue to expand at an extraordinary pace. However, this exponential growth also amplifies their environmental and energy footprints, creating an urgent need to balance technological progress with ecological responsibility. Sustainable data centres, those that integrate renewable energy, efficient cooling technologies, circular material use, and heat recovery systems, represent the path forward. They are not only essential to meeting rising computational demands but also to ensuring that digital transformation aligns with global climate goals. We at Vinculum believe that the shift toward greener, smarter, and more resilient infrastructure will determine how effectively the digital economy can grow without compromising environmental integrity. In essence, the future of data innovation depends on how quickly and effectively the world embraces sustainability at the very core of its data centres.
This article is also published on Substack. illuminem Voices is a democratic space presenting the thoughts and opinions of leading Sustainability & Energy writers, their opinions do not necessarily represent those of illuminem.
See how the companies in your sector perform on sustainability. On illuminem’s Data Hub™, access emissions data, ESG performance, and climate commitments for thousands of industrial players across the globe.
Philipp Petry
AI · Power Grid
Rahul Rajeevkumar Urs
Solar · Renewables
illuminem briefings
AI · Power Grid
The Guardian
Power Grid · Net Zero
Euractiv
Energy · Power & Utilities
Euractiv
Power Grid · Power & Utilities