Digitalisation is already touching every area of our lives, but it is set to have ever greater impact thanks to the AI revolution. McKinsey research has estimated that generative AI alone has the potential to add up to $4.4 trillion in economic value to the global economy.
The processing power of AI applications is considerably greater than existing applications such as Google Search. Generative AI has two key phases: training and inference. The training phase requires substantial amounts of data, in excess of one trillion parameters, to teach the AI models how to process data and generate predictions. The inference phase, which is still evolving, requires low latency, or reduced time in relaying information from server to end user.
The consequent explosion in demand for computing power will act as a huge demand driver for data centres. “Most current cloud platforms are not equipped to handle the quantum computing requirements AI models need,” says Morgan Laughlin, global head of data centre investments at PGIM Real Estate. “This means Tier 1 data centre locations need to expand capacity to enable optimal communication among devices.”
For instance, self-driving cars need almost instantaneous communication signals from servers to devices to function efficiently, which requires the data centres to be nearby the vehicles. Hyperscalers are rapidly building many additional data centres across global markets to accommodate this additional previously unforeseen demand.
“AI readiness is becoming a mission-critical prerequisite for data centres given that generative AI is behind an estimated 20 percent annual increase in data centre demand through 2026,” Laughlin says. “Over the next four years, more than $100 billion is projected to go into data centre capital expenditures necessary to enable generative AI.”
More power
The problem – as Dalmar Sheikh, global head of data centre operations at Actis, points out – is that “data centres are notoriously energy-hungry”. In Denmark, for example, data centre energy use is projected to rise by six times by 2030 to account for almost 15 percent of the country’s electricity use.
All experts agree on this point. Adam Waltz, managing director in BlackRock’s diversified infrastructure team, believes data centre power needs will double by 2030, with “nearly 80 percent of all data centre power being consumed by AI compute by 2040”.
Meanwhile, Richard Marshall, head of infrastructure research at DWS, estimates the proportion of European power demand coming from data centres as “1 percent currently, increasing to 5 percent in the 2030s”.
This hunger for power is obviously an issue as the world struggles to achieve a sustainable energy transition. Fortunately, a majority of hyperscalers have made public commitments to power their operations with 100 percent renewable energy by 2025-30, and major corporates are investing in renewable energy projects including solar and wind farms in addition to geothermal solutions.
“It sometimes feels as if AI has come out of nowhere and thrown gasoline on the fire of infrastructure investing”
Michael Brand
Mercer Investments
For example, Amazon has more than 500 solar and wind projects globally, enough to power 7.2 million US homes each year. Laughlin notes: “This high-level corporate commitment makes digital infrastructure development a driving force behind continued renewable capacity expansion globally.”
Powering data centres through renewables, either co-located or through PPAs, can therefore create greener and more profitable data centre operations. But this is not straight forward. Chris Manser, head of infrastructure, international, at Swiss Life Asset Managers, observes: “Given the intermittent generation profiles of most renewable power technology, the challenges around procuring renewable power will grow increasingly larger and will require storage or other solutions.”
Transmission and distribution grids will also need to be strengthened to get the renewable power to the data centre hubs. As an example, the German transportation grid company Amprion is planning to build an 8GW corridor to bring green electricity produced in the North Sea to the data centres in the Frankfurt area.
Waltz thinks meeting this future power demand from AI will “include this combination of on-grid and off-grid, with a focus on carbon-free power”. He adds: “We also expect that data centre operators will look to behind-the-meter and distributed generation carbon-free power solutions.”
But AI can itself play a significant role in managing its impact. Sheikh says: “The other big factor to green data centres is energy efficiency. By implementing green building designs to data centres and using a variety of tools such as district cooling for cooling needs or real-time AI monitoring of energy efficiency, for example, data centres can continue to use fewer resources.”
Michael Brand, head of real assets at Mercer Investments, sums up this imperative: “It sometimes feels as if AI has come out of nowhere and thrown gasoline on the fire of infrastructure investing.
“The crucial question is where we will get the massive increase in power generation that is necessary. It has to also be said that this power generation has to be green, or otherwise the explosive growth of AI is going to be an impediment to the energy transition. It certainly seems so in the near term.”
Problem solving
The good news is that AI itself can be part of the solution to the very problems that it threatens to cause. This is true in the energy industry, for example.
By predicting failures and improving the timing of maintenance activities, AI can ensure peak energy production efficiency, minimising downtime and excess inventory. “AI is the cornerstone of innovation, increasingly helping to maximise revenues while optimising maintenance costs,” says Ralf Nowack, head of Actis’s long life infrastructure operations.
Smart meters and Internet of Things (IoT)solutions, combined with AI, promise to usher in proactive maintenance planning by leveraging data from sensors and drones to anticipate issues before they arise. AI-driven video analytics is redefining inspections, boosting efficiency and safety. And in data centres themselves, AI software can now allow for energy arbitrage, buying and selling electricity when the prices are best.
The use cases stretch across the infrastructure sector. Laughlin notes: “The IoT and connected devices have allowed us to gather a huge amount of data over the past decade on infrastructure performance and infrastructure use. AI tools are now at a point where we can really begin to analyse and utilise this data.”
Analysing traffic patterns to optimise road infrastructure to reduce congestion and the use of fossil fuels would seem an obvious use case, as is optimising real estate and infrastructure development project design and planning by incorporating historical data, geospatial information and predictive modelling.
As Laughlin says: “These examples are just the tip of the iceberg of AI use cases that will positively impact the infrastructure sector. One only has to use one’s imagination to consider the use cases.”