There is an ever-shifting demand for datacentre capacity, which requires stakeholders to consider everything from power supply and cooling, through government policy and planning, to processing needs, connectivity, labour availability and location.
“There’s more change and uncertainty in how the market is going to develop right now than there’s ever been,” says Andrew Jay, head of EMEA datacentre solutions at commercial realtor CBRE. “Right now, right here, it’s really difficult to predict exactly where the new datacentres in the UK are going to get built.”
Facilities may need to be sited in a specific area. Technologies evolve, making this or that scenario more or less feasible and costly. In addition, power availability in sought-after locations remains a UK and global problem – one which CBRE research suggests will require greater collaboration among stakeholders, from datacentre operators, investors and property developers to utility companies.
Operators might look at new ways to reduce grid congestion or lift reliability or consistency of supply, as well as decreasing costs. According to Jay, who also sits on the UK datacentre trade association administered through TechUK, this should include grid-enhancing technologies and micro-grids – but these may take years more to build out and involve changes at government policy level.
“There’s a mismatch between the time needed to change up power availability in the UK and the time needed to build a new datacentre,” Jay notes.
“What we’re saying to the government – very politely – is that, rather than land for AI [artificial intelligence] zones, we need to first look at UK power infrastructure, discussing plans and the art of the possible with the new National Energy System Operator (NESO) and the National Grid.”
Rising power demand
NESO’s 2023 analysis – the latest available – projects total annual UK electricity demand of 570TWh (terawatt-hours) if net-zero targets are not met by 2050 and 726TWh if they are met by 2046. That’s up from 286TWh in 2022.
Competition for resources has been part-driven by the perceived gold rush to deploy and support AI, estimated to require multiples of the power needed in traditional enterprise IT. Globally, many are agitating for datacentres serving AI and large language models (LLMs) to come on stream as soon as possible, with the mantra “develop capacity or be left behind”.
CBRE’s February figures suggest that 937MW (megawatts) of new European supply could come online this year. While 57% of this is expected in the main hubs – London, Dublin, Frankfurt, Amsterdam and Paris – double-digit supply expansion is expected in five of 10 secondary markets tracked by CBRE. Seven of those 10, including Milan and Madrid, will exceed 100MW as 2025 closes, up from just four on 31 December 2022.
Ironically, micro-grids and renewables mean it has become “quite a complicated exercise” to work out exactly what power is coming on when and where, versus the past when those commissioning datacentre builds just needed to know when the power station would be switched on.
Jay notes that electricity generation landscapes have changed “massively” in the past five years. At the same time, there’s little chance that regulatory frameworks will be ready to support small modular reactors (SMRs) to fill supply gaps in the next five or more years.
This all suggests building collaborative efforts in capacity planning, to mitigate risks where possible, might be appropriate. Beyond the electricity requirements, planners need to consider the land required to build these datacentres.
With so many moving targets, it makes sense for operators to take their future into their own hands as much as possible.
One thing to do is continue to reduce risk by maximising energy efficiencies by deploying energy-saving technologies and energy-efficient designs; reducing the overall need for equipment on site; and replacing inefficient setups, including for cooling.
Artificial intelligence is one area driving demand for new datacentre builds. The second is hyperscale cloud. Typically, customers are looking for mega facilities hosting one or two customers, says Séamus Dunne, managing director for the UK and Ireland at Digital Realty.
The government has been focusing on the need for AI development zones up and down the country, yet hyperscalers often seek to be somewhere like west London, where availability zones need to be within 100 miles (161km) or so of each other. They need density of connectivity options such as fibre-optic cables and subsea cable points.
“West London has built up over a couple of decades. That’s almost impossible to replicate,” says Dunne. “But across Europe, thousands of enterprises need and want to use such facilities.”
While what the government is doing is not wrong, it’s not all about AI Séamus Dunne, Digital Realty
Indeed, Digital Realty’s main customers are in enterprise IT, comprising everything from banks and financial services to pharma, medical, retail and manufacturing businesses. When building a datacentre, it expects interest from hundreds of customers doing various things in aid of economic growth and productivity, he points out.
“While what the government is doing is not wrong, it’s not all about AI,” Dunne adds. He believes there needs to be more focus on catering for the datacentre capacity requirements of enterprise customers.
Dunne believes the government should be wary of focusing too heavily on mega datacentres that are built mainly for AI training. “Inference is where you use it [AI], typically nearer population centres,” he adds.
While recent planning reforms have been positive, they’re insufficient on their own to address the complexity of buyer-group requirements and the consequent lack of datacentre capacity, he says.
Renewables and efficient cooling advances can help, but are simply “tweaking at the edges” as power densities rise. However, Dunne believes that more attention should be paid to plans for different user needs – for instance, developing the London, Oxford and Cambridge triangle further – which, according to Dunne, are essential to support ongoing research.
Among the many barriers to building out datacentre capacity is finance, as David Bloom, chairman of datacentre operator Kao Data and CEO and founder of investor Goldacre, explains. “This sector has a delightful set of barriers to entry,” he says. “Because, whisper it quietly, the purpose of investment is to make money.”
Redefine collaboration on capacity planning
He urges industry and government to develop joined-up thinking on buildings, construction and overall infrastructure. Supporting technological advances with datacentre capacity must be planned for as part of a larger national picture.
Of course, individual organisations are free to “set their own agendas” as well … but these, too, should be based on first principles. This should help to expose and manage the risks of any given pathway.
“For example, what do we mean when we say ‘AI opportunities’? What are they? What are the first principles? What are we trying to achieve? Sometimes I feel that isn’t clear,” says Bloom.
According to Bloom, only a small share of the global datacentre installed base today processes AI workloads. Large investments without a strong use case may not pay off. Humans have a tendency to assume the future will resemble the past – but extrapolated upward curves based on past performance are not necessarily predictive.
Oversupply can also curb datacentre capacity plans. For instance, in February, investor TD Cowen reported Microsoft’s cancellation of several US datacentre leases in the hundreds of megawatts and pulled back on converting statements of qualification into signed leases.
Bloom says that even if AI requirements accelerate, judging future capacity needs sensibly should begin with analysing carefully what you want to achieve, then working out how to support that goal.
“Start with a question you do know the answer to,” he says. “To reach that future, what do you want, what do you actually need, what should you do, and who is going to do it?”
Bloom says a lack of understanding does hold back datacentre developments. A datacentre professional can end up in front of local councillors who, through no fault of their own, have spent 20-odd years signing off housing, and maybe offices or industrial. He recommends that a set of very understandable and clear guidelines are needed for local councils, in addition to education outreach and help from the industry. “Not all developments should be approved, yet sometimes wrong decisions are made,” he adds.
It follows that anything that reduces uncertainties or a fuller understanding of risk can ease datacentre capacity planning.
DSIT’s classification of datacentres as critical to the UK means there is now a dedicated CNI data infrastructure team of senior government officials who are tasked with monitoring and anticipating potential threats. This team provides prioritised access to security agencies, including the National Cyber Security Centre. It is also responsible for coordinating access to emergency services should an incident occur.
Peter Judge, senior research analyst at Uptime Intelligence, says even eligibility for CNI status can prove double-edged. First, not all datacentres are doing critical work and therefore classed as CNI – for instance, something inessential that doesn’t support “continuity of life”, however defined. Operators can’t assume that if generators and the grid fails, they’ll be able to requisition 10 diesel tankers a week to keep going, says Judge.
“Are you serving frivolous videos, or supporting national services? Quid pro quo, the government will want more transparency into what’s happening in datacentres,” he says.
It might be harder to greenlight certain projects. Managing risk itself becomes more critical and potentially costly. Security, cyber security, confidentiality and resilience will loom larger.
“With great power comes great responsibility. They’ll be required to fall in line with government policy and also be transparent to local objections and explain the necessity of what they’re doing,” he says.
Colocation providers have specifically noted that revealing more about their activities might break the contract they have with their customers, Judge adds.
In short, UK datacentre capacity planning remains a complexity of siloed stakeholders, often with conflicting interests. Pros and cons exist for each variable in any given proposal, against an evolving compliance backdrop.
Bodies like NESO are grappling with the challenges. Meanwhile, operators might be best advised to act to reduce uncertainties where possible in their own capacity plans.