The explosion of generative AI and the associated increase in demand for computing power to implement it is already having an impact on the data centre industry. We are just at the start of the wide-spread adoption of generative AI so we can expect the impact to gather pace over the coming months and years.
As with any big technological advance, the rise of generative AI brings significant opportunities, but also risks. The biggest risks in the data centre industry are co-location facility providers becoming overloaded and challenges securing the space and power required for new developments. There is an urgent need for increased investment to mitigate this.
Demand outstripping supply
The headline from JLL’s recently published North America Data Centre Report is that record demand is meeting limited supply and this is largely driven by generative AI and the continued adoption of cloud services. We are seeing a comparable situation emerging in the UK and this looks set to continue as businesses across all sectors embrace generative AI.
The JLL report highlights a number of key issues.
Supply chain slowing down
The JLL report points to the supply chain slowing as an exacerbating factor in the imbalance of supply and demand, stating: Supply chain issues and the lack of available land with onsite and scalable power necessary to satisfy future requirements have lengthened data centre development timelines to three to five years, or more.
AI use is increasing
AI use is on the rise. Market research firm Gartner reports that synthetic data will completely overtake real data in AI models by 2030. The Business and Insights Conditions Survey (BICS) from the Office for National Statistics, which was published in June 2023, showed that approximately one in six UK businesses are already implementing at least one AI application. That figure is sure to rise as more organisations trial generative AI and weave it into different areas of their operations.
Demand for higher density and larger data centre requirements
The JLL report notes that the increasing adoption of AI has caused demand for higher density and larger data centre requirements. Just one example highlighted in the report is that Investment Bank TD Cowen recently reported that 2.1GW of US data centre leases were signed in the second quarter of 2023.
Impact on data centre infrastructure
Outlining the impact on data centre infrastructure, the JLL report notes that AI adoption will need high power density server clusters, with densities of 50-100kW per rack. The report also flagged the fact that many colocation providers have reduced the upfront cost to deliver power to these high-density clusters by increasing the voltage delivered to the floor to 415 volts (in the UK we do this anyway). The report concludes that to meet sustainability goals, innovations will be needed to improve cooling and energy efficiency. We are looking at projects where they are talking about racks requiring 200kW per rack.
Impact on co-location facility providers
The growth of generative AI results in a risk that co-location facility providers will become overloaded.
When I look at the UK specifically, there are many fallow halls that clients have moved out of. In response, Mercury Power is currently designing platforms in these data centres to enable AI and Super Compute to be deployed.
However, there will become a time where power may restrict these deployments as Direct Liquid Cooling reduces PUE and increases the compute power. The physical space required is a fraction of what a traditional deployment has been. Where hundreds of racks were used now this is in 30 or 40 racks. Multiple cooling systems will need to be installed to make the gains in efficiency.
Impact on customers with basic data centre requirements
Not all customers have fully embraced generative AI, and many still have more basic requirements.
The JLL report highlights that major cloud service providers are growing at “breakneck speed” to support workloads. The report goes on to emphasise that these large requirements are causing challenges for users with smaller requirements, due to the difficulty finding space and power. I agree with the report’s findings that planning ahead is crucial for users with smaller requirements, so that they can find space before they need it.
With the shortages highlighted in the JLL report, which we are also starting to see in the UK along with more deployments being planned and commissioned, there is an opportunity for smaller regional players to offer services to meet the needs of customers looking for smaller or more basic data centre requirements.
It is likely that we will see increased investment from co-location providers to accommodate for the increased demand.
Inevitably everyone would like to ride the crest of this wave and be the most powerful, fastest to deploy and of course offer the most power dense facilities.
The race is on to push the boundaries on the infrastructure to maximise the opportunities that exist. At Mercury Power we are seeing investment from the global players all the way through to the UK wide Data Centres and regional players, which is encouraging. We are seeing large investment in the smaller data centres, and they are assessing how they can scale to meet the new demands. Looking at flexible M&E designs that can adapt to the ever-changing needs of the clients. This is through driving efficiencies as well as infrastructure investment.
Investment and product development is required from the Cooling OEM’s. They need to develop products that can enable our designs to be more efficient and offer more capacity. This needs collaboration between the AI OEM’s, Data Centre Operators, Data Centre Designers, M&E OEM’s and of course the clients using the data centres.