We need a whole stack approach to introduce novel technologies for sustainable AI infrastructure
By Dr Ilian Iliev, CEO of EMV Capital PLC
The race to dominate in AI is not constrained by capital or talent so much as by electricity. Data-centre energy use is ballooning so quickly that the result is a looming collision between digital ambition and physical limits.
As billions are poured into new sites, the energy systems and environmental costs are becoming impossible to ignore. Data centres have become indispensable to modern economies, yet their surging appetite for energy and water risks straining utilities and igniting political and regulatory backlash. As investment spills from the US and rich economies into emerging markets, where systems and environmental regulation are often weaker, the stresses will only intensify.
The world is drifting towards a utility crisis. Avoiding that outcome demands innovations that cut the energy requirements of AI and data-centre infrastructure, while aligning the expansion of digital services with sustainability targets. Encouragingly, the UK is emerging as a hub for precisely such scalable solutions, particularly from university spin-outs in disciplines such as quantum photonics, advanced cooling and modularity, offering globally scalable solutions.
The Hidden Costs of Digital Abundance
Data centres are the hubs that store the world’s digital information and its computing power, such as cloud services, video streaming, AI training and AI delivery. Their growth mirrors society’s surging data consumption. But the facilities housing these workloads are ravenous and thermally intense. High-temperature processors require constant cooling, drawing vast quantities of power and water.
The International Energy Agency estimates global data-centre electricity use at around 415TWh for 2024, some 1.5% of global electricity, and expects usage to more than double by 2030 as AI spreads through the economy.[1] Crucially, most additional demand is likely to be met by fossil-fuel generation in the near term, threatening climate commitments and heaping pressure on consumer bills.
Grid strains are already visible. In Virginia’s “Data Centre Alley”, more than 200 sites now draw as much electricity as the entire city of Boston;[2] a recent protocol failure forced 60 centres onto diesel backup, nearly triggering a regional blackout. In Santa Clara, California, two hyperscale projects sit idle because utilities cannot yet deliver sufficient power.[3]
Water use is similarly a challenge. Cornell researchers estimate that by 2030, US AI data centres could consume 1,125 million cubic metres of water annually; equivalent to the household consumption of ten million Americans.[4] And as AI models expand, so does their footprint: one estimate suggests that if GPT-5 handled all of OpenAI’s daily queries, it would consume the daily electricity of 1.5 million US homes.[5]
With AI infrastructure spreading into emerging economies, the risks will multiply.
Boosting renewable capacity, nuclear investment and micro-grids will help, but supply-side reforms alone cannot solve the problem. A sustainable AI economy requires a whole-stack approach: whereby energy demand is tackled head on, even as energy supply is reconfigured to meet the needs.
A Whole-Stack Investment Strategy
Through our EIS and Martlet funds, we at EMV Capital are backing university spinouts and other startups across the technology stack, from software and chip design to novel materials, telecoms, power electronics and energy storage. We also examine how operators will adapt, including moves toward modular, decentralised, or off-grid architectures, and the inevitable tightening of regulation.
Our portfolio includes companies developing technologies that reduce power consumption, boost computational efficiency, and support lower-carbon or resilient data-centre operations.
Among them:
Data-centre efficiency
- OctaiPipe uses collaborative AI and federated learning to optimise energy use across complex systems, enabling data centres to cut waste, maintain reliability and reduce cooling energy consumption by up to 30%.
- Porotech is redefining photonics with GaN microLED light sources offering up to 68% lower power use, 100x higher reliability and lower-cost scalable integration than conventional laser-based components; facilitating ultra-efficient, high-speed data movement.
- Cambridge GaN Devices (CGD) produces next-generation gallium-nitride power electronics achieving conversion efficiencies above 99% and energy savings of up to 50%, providing compact, stable power systems for modern computing infrastructure.
The future of computing
- Nu Quantum is developing quantum-photonics interconnects to link multiple quantum systems within data centres, enabling scalable and energy-efficient modular architectures capable of supporting the next wave of AI workloads.
- Paragraf harnesses graphene’s exceptional electrical and thermal properties to create transistors and sensors that could deliver orders-of-magnitude improvements in computational speed while significantly cutting data-centre energy use.
Distributed infrastructure
As AI spreads beyond major hubs, demand for resilient off-grid and remote solutions will grow.
- Sofant, based in Edinburgh, provides phased-array satellite-communications antennas that can deliver broadband connectivity to off-grid or backup data-centre locations.
Forces Shaping the Next Wave of AI Infrastructure
Several trends will determine how AI infrastructure evolves:
Regulation
Performance per watt is becoming a competitive necessity. Policy pressure will soon reinforce this: expect carbon-reporting mandates and energy-efficiency standards in the EU and US within three to five years, alongside surcharges for water use.
From mega-factories to modular sites
As AI adoption globalises, operators are likely to increasingly adopt modular, distributed and off-grid systems over the monolithic data-centre “mega-factory”. This shift will spur demand for integrated renewable generation, energy storage and satellite communication.
Cooling
Cooling demands will rise as facilities move into hotter geographies. Proximity to water bodies may ease thermal stress but will heighten environmental scrutiny.
Security and resilience
Physical and cyber-security will become paramount. Demand will grow for sovereign AI capabilities, secure supply chains, encryption, robotic security systems and drone-defence technologies. Geopolitical tensions and export controls will only sharpen this trend.
Regional risks
Rapid AI deployment across Asia-Pacific, Africa and Latin America will expose fragile grids and water systems to severe stress; potentially triggering political instability and supply-chain disruptions.
In Summary
The AI revolution must be powered responsibly. Without immediate collaboration between innovators, investors and policymakers, the world will stumble into an avoidable energy crisis, followed by heavy-handed regulation. Sovereign security concerns will further complicate the landscape.
But the UK is well placed to contribute solutions. With four of the world’s top ten universities and numerous research centres of excellence, its spin-outs – often supported initially by EIS investors and later by corporate and institutional venture funds – are producing novel and breakthrough technologies capable of addressing global challenges while generating significant commercial value.
A whole-stack approach to innovation and technology deployment, applied with urgency, can ensure that AI’s growth remains both economically transformative, and sustainable.
[1] International Energy Agency
[2] Reuters
[3] Bloomberg