Uptime Institute emphasizes that liquid cooling remains niche in AI data centers, suited only for ultra-high-density setups. This insight helps engineers evaluate cooling options as AI compute demands intensify without overinvesting in specialized systems.

• Liquid cooling targets racks over 100 kW, while most AI workloads stay under 50 kW with air cooling
• Adoption rate in hyperscale facilities is below 5% for AI-specific deployments
• Implementation costs for liquid cooling range from $500,000 to $1 million per rack row
• Air cooling handles most setups under 50 kW effectively

Roland Berger’s report highlights how energy storage systems address power volatility in AI data centers, which is essential as AI workloads cause rapid demand spikes that strain grids and increase costs.

• AI training can cause power demand to fluctuate by 20-50% in seconds, risking outages
• Battery systems provide 50-200 MW of backup power for 15-60 minutes during peaks
• Roland Berger projects storage integration could reduce energy costs by 10-25% annually
• California and Texas lead with over 5 GW of data center storage deployments planned by 2025
• Lithium-ion batteries dominate, with 80% market share, followed by flow batteries for longer durations
• Report recommends hybrid setups combining storage with renewables for 99.99% uptime

Meta is accelerating training programs for fiber optic technicians to address a critical shortage that’s delaying data center expansions. This initiative highlights the growing need for specialized skills in high-speed networking infrastructure amid surging AI demands.

• Program targets training 500 technicians over the next 12 months through partnerships with community colleges in California and Virginia
• Shortage of fiber splicers has delayed Meta’s data center projects by up to six months in key regions
• Training focuses on fusion splicing, OTDR testing, and high-density fiber management for 400G networks
• Meta allocates $10 million for the initiative, including equipment and certification costs
• Effort aims to support deployment of 100,000 kilometers of new fiber optic cable in 2024
• Similar bottlenecks affect competitors like Google and Amazon in Nevada and Oregon data center hubs

In this Datacenter Dynamics interview, host James Radings speaks with Rajat Bagghdad, senior mechanical engineer at Arcadis, about current challenges in data center cooling systems. Data center engineers should watch to gain insights into deploying liquid cooling at scale and addressing operational hurdles in high-density environments.

• Liquid cooling technology is well-known, but the main challenge lies in its industrial-scale deployment for hyperscalers across multiple data centers.
• Hyperscalers need vendors and OEMs to supply the necessary liquid cooling infrastructure to meet deployment demands.
• Operators trained in air cooling must quickly learn liquid cooling while maintaining high uptime requirements.
• Design challenges include understanding interactions between servers and cooling systems in high-density AI factories where loads fluctuate rapidly.
• GPU workloads can spike every couple of seconds or microseconds, but cooling systems are not reactive enough to handle such variations.
• Engineers focus on stabilizing systems to provide continuous flow of cooling or chilled water, as chillers are not suited for rapid changes.

A recent report analyzes the explosive growth of data centers in Virginia and Texas, detailing their outsized effects on regional power grids. Understanding these dynamics helps engineers plan for scalable, reliable electricity infrastructure amid rising AI and cloud demands.

• Virginia data centers consumed 5,000 MW in 2023, representing 25% of Dominion Energy’s total load
• Texas saw data center capacity reach 2,500 MW by mid-2024, driven by expansions in Dallas and San Antonio
• Projections indicate Virginia’s data center power needs could hit 10,000 MW by 2030
• ERCOT forecasts 15 GW of additional load growth in Texas from data centers over the next decade
• Both states face transmission upgrades costing billions, including $5 billion in Virginia projects
• Report highlights policy responses, such as Texas incentives for on-site generation

Marshall County in Indiana has approved a permanent ban on data centers, effective immediately, potentially shifting development focus to more permissive regions. This move highlights escalating local regulatory hurdles for data center expansions amid concerns about infrastructure strain.

• Marshall County Commissioners approved the ban in a unanimous vote during their latest meeting
• Ban prohibits all new data center constructions and expansions within county limits
• Existing facilities remain operational but face restrictions on modifications or upgrades
• Decision driven by local worries over high water consumption, energy demands, and noise pollution
• Effective immediately, with no grace period for pending project applications
• Similar restrictions under discussion in adjacent counties like Kosciusko and Fulton