AI training facilities
AI training facilities represent the cutting edge of computational power, demanding
Tom Glover, Head of Data Centre Transactions at JLL massive resources. These facilities house specialised hardware like GPUs and TPUs, which consume vast amounts of energy and generate significant heat. For these facilities, key location characteristics include:
• Access to power: These facilities can require 200MW, 500MW, or even upwards of 1GW. This level of power consumption needs proximity to robust power grids and potentially even dedicated power generation infrastructure.
• Extensive land area: A minimum of 200 acres is typically required to accommodate the sheer scale of these facilities, including the physical footprint of the data halls, power infrastructure and cooling systems. This also allows for future expansion and redundancy. The clustering
phenomenon is leading to power delivery bottlenecks in major markets, making large land availability in less congested areas even more critical.
• Price sensitivity: While cost is always a factor, access to sufficient power and land trumps price sensitivity for these facilities. Clients are often willing to pay a premium, sometimes up to 20 % above the next highest value location, to secure the necessary resources.
Cloud / Inference Data Centres: Balancing scale and cost
Cloud and inference data centres operate on a slightly smaller scale. Their location requirements differ accordingly:
• Moderate land area: 10, 20, or 30 acres are generally sufficient, although more is always welcome. Flexibility is
www. intelligentdatacentres. com 65