The meteoric rise of generative AI, exemplified by ChatGPT’s success, has ushered in an era of unprecedented demand for data centers. However, this surge in demand comes at a cost, as data centers grapple with the need for extensive computing power and infrastructure enhancements to accommodate the relentless appetite for generative AI.
Generative AI, while transformative, is a capital-intensive operation that demands substantial computing power. The surge in demand for generative AI applications has significantly increased the cost of infrastructure, presenting a challenge for data center operators and, subsequently, customers.
Demand Outpaces supply: A struggle for space and costs
JLL, a real estate firm specializing in commercial property and investment management, has identified a substantial gap between demand and supply for colocation space in North America. This shortfall has resulted in a significant surge in costs for customers, with a notable increase of up to 30%.
The report notes that major cloud service providers are rapidly expanding to support AI requirements, driving the need for more computing power. This expansion has made it increasingly difficult to find suitable space and power for smaller requirements in various markets.
Leasing surge in 2023 and Meets AI development needs
The second quarter of 2023 witnessed a substantial increase in leasing activities driven by the demand for higher density data center facilities, primarily to support AI development. This surge underscores the growing importance of data centers in meeting the computational requirements of generative AI.
JLL’s report aligns with global projections concerning the impact of generative AI on data centers. Research firm Tirias Research forecasts that the demand for data center infrastructure to support generative AI could surpass $76 billion by 2028, more than double the current estimated cost of Amazon Web Services (AWS), a prominent player in the cloud infrastructure services market.
Data center leases at record levels
Earlier this year, a study by TD Cowen highlighted “record levels” of data center leases as hyperscale providers expanded their capacity to accommodate AI development. The industry’s response to a “tsunami of AI demand” strained some service providers, challenging their ability to keep pace with the requirements of hyperscale clients.
The need for infrastructure overhaul
Data centers are now tasked with improving efficiency and adapting infrastructure to accommodate high-power density clusters. This overhaul encompasses changes in cooling capabilities and energy consumption, addressing the growing demands of generative AI.
The surge in AI server usage has also raised concerns about water and energy consumption. Data centers are consuming substantial volumes of water to cool generative AI servers, contributing to environmental concerns. For instance, Microsoft reported a 35% increase in water consumption at its data centers in 2022 compared to the previous year.
Meeting the demands of high power density clusters
Some data centers are responding to the challenge by increasing cluster density to 50 to 100 KW per rack, a significant departure from current hyperscaler demands. Many colocation providers have boosted the voltage delivered to the floor to reduce the upfront power delivery costs for high-density clusters.
While generative AI-related demand puts pressure on data centers, the industry is witnessing positive implications, especially in Europe and the UK. The surge in demand for High-Performance Computing (HPC), AI, and GPU-powered technologies is driving the development of new Tier 2 data hubs.
This wave of innovation is prompting a comprehensive redesign and re-engineering of data center capacity to accommodate high-density computing. Innovations include the deployment of high-density and pre-configured systems like Nvidia’s SuperPOD, increased adoption of liquid cooling, and the continued demand for ultra-energy-efficient colocation capacity powered by 100% renewable energy.
Balancing generative AI focus with Customer access
While the focus on generative AI development is reshaping data center infrastructure, concerns exist about long-term access for customers with operational needs beyond AI development. The industry must strike a balance to ensure that all customers can access colocation services without constraints.
The surge in generative AI’s demand has reshaped the data center landscape, driving infrastructure enhancements, surging costs, and environmental considerations. As data centers adapt to the era of AI, they must navigate these challenges while remaining accessible to a broad range of customers with diverse needs.
