The AI data center dilemma: upgrade or start from scratch
In today’s data-driven landscape, demand for AI shows no signs of slowing down as it becomes entrenched in the daily tasks of businesses and consumers. A report from Goldman Sachs shows that AI-focused tech giants will spend more than $1 trillion in the coming years, with a significant portion of the capital focused on data centers, AI infrastructure and the power grid.
Increased AI investments lead to the need for more AI data centers, which can provide the additional computing capacity needed, but require significantly more power than traditional data centers. For example, in 2022, data centers consumed more than 4% of all electricity in the United States. Driven by AI’s power needs, that number is expected to more than double to 9% by 2030, according to the Electric Power Research Institute. Driving this energy consumption are higher power GPUs deployed in configurations that, at the rack level, will boost energy consumption from the current sub-50 kW range to a MW by 2030. In light of such predictions, data center operators are faced with a crucial decision to ensure long-term AI readiness: upgrade existing infrastructure or build new facilities from scratch.
Data center and wireline segment leader for OmniOn Power.
Each option offers unique benefits and challenges. Understanding these pros and cons is essential to making informed choices that fit business goals, budget constraints and future needs. The range of factors to consider when deciding whether to update existing technology or start over and build a new AI data center include:
– The infrastructure of the existing facility: When updating an existing data center, HVAC systems, energy infrastructure, energy distribution systems and structural load capacity should all be inspected to ensure they can meet the increased demands that will be placed on them upgrading to an AI data center.
– The ability to upgrade to higher voltages, such as a 48 volt (V) or even 800 V power architecture: Deploying or upgrading to a higher bus voltage architecture can improve thermal management and efficiency while delivering the higher power that needed for AI servers and IT equipment.
– The potential to leverage three-phase power, conduction cooling and liquid immersion cooling: both technologies can help address energy and thermal challenges associated with larger computing capabilities and more powerful networking equipment.
High capacity requirements for HVAC, energy infrastructure and energy distribution
A data center’s HVAC, energy infrastructure, and power distribution systems must meet high capacity requirements to support the accelerated computing needs of AI. Building a new data center from scratch could be a better option than the high cost of a complete overhaul if these systems are not up to the task.
HVAC systems
AI workloads are compute intensive and generate significant heat. Advanced HVAC systems are critical for maintaining optimal operating temperatures to prevent overheating, which can lead to data center hardware failures and reduced performance. Good airflow design, hot/cold aisle configurations and effective cooling strategies can ensure cooling is efficiently targeted where it is needed most. This is especially important for densely packed server environments typically found in AI data centers.
As AI applications become more complex and in demand, HVAC systems must be scalable to handle increased heat loads and flexible to adapt to changing needs. Efficient HVAC systems in data centers can help lower operational costs, improve sustainability and reduce a data center’s overall energy footprint.
Power infrastructure
AI workloads require a data center’s power conversion solutions to deliver consistent, reliable energy to ensure optimal operations. Fluctuations in power quality can negatively impact sensitive AI hardware, so power infrastructure must provide clean, stable power to avoid costly downtime. According to a report from PCMag, the financial losses associated with network disruptions are significant, as just one minute of global internet outage would cost the global economy $20 million.
As existing data center systems approach or reach their functional limits, operators will look for innovative ways to provide greater computing capacity and meet associated power needs. Whether upgrading or building new, backup power systems will be essential to ensure continued operation and data integrity. Backup systems and dual-redundant power sources can provide reliable, uninterrupted service during power outages and fluctuations, reducing the risk of downtime. Furthermore, well-designed energy infrastructures can accommodate the dynamic loads caused by AI demands.
Energy distribution systems
High-capacity data centers require energy distribution systems that efficiently manage and deliver energy without significant losses. High-voltage DC architecture can significantly improve energy efficiency by reducing power losses during transmission.
Direct utility systems that connect power directly to a data center’s power and network cabinets can also improve energy efficiency by minimizing the distance electricity must travel. Furthermore, to optimize the performance of IT equipment and energy systems, implementing real-time monitoring and control functionality enables the proactive management of energy consumption during the ebb and flow of demand.
Structural tax considerations
The overall architecture must take into account an increase in floor loads, in addition to roof loads, because higher power density results in more copper and liquid cooling: direct-to-chip, liquid-cooled rails and power supplies. The increased PSI (pounds per square inch) requires a reassessment of charging capabilities in existing data centers.
When deciding whether or not to upgrade an existing data center, the costs associated with refurbishing infrastructure and energy distribution systems can make building a new facility the most cost-effective strategy.
Upgrade to a 48V bus architecture
To improve efficiency and meet increased power needs due to AI and other technologies, operators are now using 48V network equipment and 48V bus architectures instead of traditional 12V IT equipment and DC buses. The transition to 48V architecture in data centers allows less power to be sent to network equipment, which in turn helps improve thermal management and efficiency, reduce energy costs and create higher density power delivery. Moving to a 48V architecture, which was recently just a consideration but has become a necessity, reduces the number of conversion phases required to power IT equipment, helping to reduce the overall cost of power distribution.
As AI increases computing demands, 48V systems can better support the high-density power needs of modern servers and networking equipment and enable easier scaling as new equipment is deployed. Depending on how easily and cost-effectively power systems and DC bus architectures can be upgraded to 48V versions, building a new data center could be the most cost-effective long-term solution. Even higher voltages are coming – both to the rack and within the rack
Addressing energy and thermal management challenges
Three-phase input power and liquid cooling are innovative solutions designed to address the energy and thermal management challenges of small data centers. If an existing data center cannot support the addition of these solutions, building a new data center may be the best option.
Three-phase input power systems
In data centers, three-phase power systems can help reduce the power required for the same amount of power, which can lead to lower energy losses. Three-phase systems can minimize the need for larger transformers and distribution solutions, saving space and simplifying power conversion. Three-phase power systems can distribute electrical loads evenly across three conductors, reducing the risk of overheating and allowing more stable and reliable operation of the equipment.
Liquid immersion cooling
Liquid immersion cooling is particularly effective in dealing with the high thermal loads generated by densely packed servers in data centers. Liquid immersion can reduce the need for traditional cooling infrastructure, allowing for more efficient use of available data center space. By reducing reliance on air cooling systems, liquid immersion cooling can reduce energy consumption and operating costs.
Analyzing specific needs to make the best choice
Whether an operator decides to upgrade an existing data center or build an entirely new data center, key factors to consider include budget, strategic goals, the state of existing infrastructure and the ability to scale. Therefore, data center operators must thoroughly analyze their circumstances to determine the best path forward. In some cases, a hybrid approach, using upgrades and new builds to maximize efficiency and capabilities, can deliver the best possible ROI.
We have offered the best IT infrastructure management service.
This article was produced as part of Ny BreakingPro’s Expert Insights channel, where we profile the best and brightest minds in today’s technology industry. The views expressed here are those of the author and are not necessarily those of Ny BreakingPro or Future plc. If you are interested in contributing, you can read more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro