AI is fuelling hunger for ever-more computing power – which means global demand for data centres is booming. For context, one ChatGPT prompt requires ten times the compute of one Google search. To power AI computing, data centres need newer generations of central processing units (CPUs) that are generally larger, more energy-intensive, and generate more heat. As such, the growing number of data centres requires an inordinate amount of energy. This poses a dual challenge: first, how can we meet the high energy needs of AI-focused data centres? And second, how does this align with net zero targets?
Environmental concerns
AI’s increased energy consumption is raising some notable environmental concerns. Microsoft, for example, recently adopted a ‘power first’ strategy, prioritising the availability and reliability of energy supply when selecting data centre sites. However, the most reliable and available energy sources today are of fossil origin. As such, Microsoft has reported a sharp increase in their greenhouse gas emissions, which can primarily be attributed to the growth in their data centre operations. We are seeing similar trends amongst other big tech companies, which are starting to bear resemblance to cement or steel plants in terms of the energy used to power their data centres.
Nonetheless, environmental concerns could become starker. The European AI market is projected to reach €60 billion by the end of 2025, a 40% increase from 2024, and to maintain an annual growth rate of 15.9% until 2030. If data centres continue prioritising power-first strategies to meet surging AI demand, emissions in the sector could escalate at an unsustainable pace.
Stretched grid capacity
In addition to the immediate environmental concern of increasing emissions, there is the technical issue of electricity grids that data centres rely on being stretched to capacity. For example, data centres now represent a fifth of all Irish electricity consumption, and in the UK, the National Grid anticipates a six-fold increase in commercial data centre demand by 2034.
In its current state, the UK grid simply cannot handle the additional load forecast by the rise in demand for AI without significant infrastructure upgrades. In North America, many regions are experiencing brownouts and rolling blackouts due to the inability to meet demand during peak periods – this could be the direction the UK is heading. The UK’s grid was built for predictable loads and primarily uses outdated infrastructure, initially designed around pockets of highly intensive industrial demand. As the grid decarbonises, power generation becomes increasingly decentralised and reliant on renewables, leading to a more fragmented energy landscape, and putting the UK grid under increasing strain.
As such, the issue is two-fold: How can the UK grid meet the power demands of AI-focussed data centres, and how can it be met sustainably? Putting more ‘copper in the ground’ and increasing grid capacity might help meet short-term power demand – but it certainly doesn’t solve the sustainability challenge. The good news is there are few alternative options for data centres that will promote a greener growth of AI technology.
Energy efficiency
First, data centres need to become more energy efficient. This can happen internally, through enhanced computational processes, and externally, by optimising their relationship with the grid, reducing capacity constraints, and lowering their carbon footprint.
Internally, this could look like more efficient chip designs, better AI cooling algorithms, and advanced, less energy-intensive cooling systems. Externally, efficiency gains can be unlocked through ‘demand response’, which involves optimising data centre energy consumption on a day-ahead basis, using forecasts for both renewable energy sources and a given data centre’s load.
An innovative step further, pioneered by Google, includes ‘load shifting’, whereby data centre operations are shifted daily or even hourly around the world to leverage excess renewable energy production. Load shifting allows data centres to align computing workloads with times and locations of abundant clean electricity. Recent data shows that by ‘chasing the sun’ from Europe to the US West Coast and back in this way, 34% of computing emissions can be cut compared to relying on servers in either location alone. As such, data centre operators could see massive efficiency gains through these internal computational and external demand-related activities.
Generating electricity
In addition to efficiency measures, AI data centre operators can complement grid-sourced energy by generating their own energy or engaging in power purchase agreements (PPAs). Data centres have untapped potential in using behind-the-meter generation assets, such as solar panels, wind turbines, or storage assets like batteries and backup generators, both of which support grid stability.
However, due to the 24/7 power requirements of data centres and the need for constant operation, the sustainability of behind-the-meter generation options is not always guaranteed. Wind turbines and solar PV panels do not provide constant power, so many data centres prefer using natural gas generators for backup power to ensure high availability. More sustainable behind-the-meter options can include hydrogen fuel cells or small modular (nuclear) reactors, providing the all-important 24/7 power but at lower carbon emissions.
The second route is via PPAs. Corporate PPAs allow data centre operators to procure electricity directly from an energy producer, providing an alternative to grid power. These agreements offer protection against price volatility in the wholesale energy markets, with long-term contracts for electricity at agreed prices from corporate counterparties. They are already popular among tech firms, with Amazon, Meta, Google, and Microsoft being the four biggest corporate buyers of renewables globally in 2022. Tech giants’ involvement in PPAs has significantly accelerated the adoption of renewable energy in data centre operations. Through this arrangement, data centre realtors can also ensure a reliable revenue stream for renewable projects, facilitating their development.
Safeguarding the AI future
AI data centres are becoming increasingly incentivised to adopt sustainable energy generation methods. Particularly in energy-constrained European markets, renewable energy is critical in granting permits for new developments. As such, AI data centres powered by renewable energy are not only more sustainable, but also more likely to attract investment from corporations with strict environmental policies.
The UK Government has bold plans to boost the country’s AI sector and safeguard its technological future. In light of this, integrating clean energy and efficiency measures into AI-driven data centre is more than an operational need: it is a strategic imperative for sustainable growth. With AI driving data centre needs to unprecedented levels, investments in solutions like data centre energy efficiency, behind-the-meter generation, and strategic PPAs are key to building a resilient, low-carbon energy ecosystem and safeguarding our AI future.
