Strategies for scaling data centers to meet AI demands
- By Pankaj Sharma
- 09 Dec 2024
- 5 min read
Traditional data centers are designed for general computing tasks, but now they are being burdened with supporting AI workloads that require significantly more computational power. Graphics Processing Units (GPUs), which are essential for AI computations, consume about four times more power than standard Central Processing Units (CPUs). This escalation leads to higher power densities within data centers, and it is pushing the limits of existing infrastructure.
In a typical data center, a rack might handle around 30 kilowatts of power. However, with high-density AI workloads, we are now looking at 100 kilowatts and more per rack. When you surpass the 50-kilowatt threshold, thermal management becomes a critical concern. Without proper infrastructure, businesses face the risk of insufficient computational power, leading to slower AI processing, reduced efficiency, and ultimately, a loss of competitive edge. This calls for the adoption of advanced cooling technologies to maintain optimal operating conditions.
Unlike air cooling, liquid cooling can more effectively transfer heat away from sensitive components. There are two primary types of liquid cooling being considered:
The industry is still evaluating which method will become the standard, but we strongly believe that Direct-to-Chip technology will ultimately win the day. Our recent investment in DTC technology through our acquisition of Motivair underscores our position that market needs and regulatory conditions, such as incoming PFAS restrictions, will push the market toward DTC.
Either way, what is clear is that liquid cooling will play a key role as chips continue to evolve and generate more heat.
To handle increased computational loads, we will need to scale data center infrastructure, but we must do so sustainably. Currently, data centers account for approximately 1.5-2% of the world’s electricity consumption. With the growth trajectory we are on, this could potentially double by 2030 if we do not move quickly to adopt more energy-efficient practices. This involves several strategies:
Optimizing Power Usage Effectiveness (PUE)
By improving how efficiently a data center uses energy, operators can reduce overall consumption. This includes upgrading to more efficient power supplies and implementing intelligent power management systems.
Implementing Renewable Energy Sources
Wherever possible, data centers should utilize renewable energy, such as solar, wind, or hydroelectric power. This not only reduces the carbon footprint but also can provide more stable energy costs over time.
Enhancing Building Design
The physical structure of data centers can significantly impact energy efficiency. Utilizing designs that promote natural cooling and reduce the need for artificial climate control can lead to substantial energy savings.
Not all businesses are positioned to build new, energy-efficient data centers. In those cases, retrofitting existing facilities can be a cost-effective way to increase capacity and efficiency.
Upgrading Cooling Systems
Incorporating advanced cooling solutions into existing infrastructures can enhance performance without the need for complete overhauls.
Implementing AI and Automation
Utilizing AI for predictive maintenance and operational optimization can reduce downtime and improve energy efficiency. For example, AI can predict potential equipment failures and recommend proactive measures to prevent them.
Power Management Enhancements
Updating power distribution units and integrating uninterruptible power supplies can improve power quality and reduce energy losses.
The geographical location of data centers is becoming increasingly important, especially as data centers grow larger. Factors such as land availability, climate, and access to renewable energy sources influence where new data centers are established.
With its strong demand growth, the U.S. remains the largest region for expansion. But regions with abundant renewable energy resources—like Northern Europe with its hydroelectric power or countries like Spain and Australia with significant wind or solar potential—are attractive for new data center developments. Additionally, cooler climates can naturally assist in reducing cooling requirements, further enhancing energy efficiency.
As we look beyond AI, emerging technologies like quantum computing and the continued rise of digital assets will place even greater demands on data center infrastructure. Future-proofing means we must anticipate and prepare for what comes next.
Scalability
Infrastructure must be designed with scalability in mind, allowing for expansion without significant overhauls.
Modularity
Utilizing modular data center designs can facilitate easier upgrades and expansions, making it simpler to incorporate new technologies as they emerge.
Collaboration with Legislators and Industry Leaders
Working closely with policymakers and other industry stakeholders is essential to develop standards and regulations that promote sustainability and innovation.
There are challenges in scaling data center infrastructure to meet the demands of AI, but they are not insurmountable. By embracing advanced cooling solutions, prioritizing energy efficiency, retrofitting existing facilities, and strategically considering location and renewable energy sources, we can future-proof our data centers.
No one will solve this alone; this is a collective responsibility. Industry leaders, policymakers, and technology innovators must work together to ensure that as we push the boundaries of what’s possible with AI—and beyond—we are doing so sustainably and responsibly. Afterall, the goal is to support technological advancement without compromising the health of our planet.
Latest in AI and Technology
When buildings begin to think
10 ways to harness energy and water efficiencies with direct liquid cooling
Why is AI such a disruptive force for data center design compared to traditional workloads?
When air isn’t enough: The liquid cooling revolution in data centers
Need help?
Product Selector
Quickly and easily find the right products and accessories for your applications.
Get a Quote
Start your sales inquiry online and an expert will connect with you.
Where to buy?
Easily find the nearest Schneider Electric distributor in your location.
Help Center
Find support resources for all your needs, in one place.