How AI is radically changing data center design?
Since the launch of ChatGPT at the end of 2022, the general public has discovered artificial intelligence (AI). This technology has quickly become a social phenomenon and has profoundly transformed the way we work, create, and communicate. But behind every AI tool (chatbot, image or video creator, automation tool, etc.) is a data center capable of processing large amounts of data. Long overlooked, these data centers are now undergoing an unprecedented transformation. To meet the growing needs of AI, particularly in terms of computing power, the very design of these infrastructures must be reinvented.
AI: a new tool that is revolutionizing everything in its path
Today, AI is everywhere. In business, it automates repetitive tasks, improves customer service, and optimizes logistics flows. In the healthcare sector, it has enabled the modeling of Covid-19 vaccines in record time. In our private lives, it personalizes our information flows and facilitates many daily tasks (writing emails, creating recipes, summarizing documents, editing images, etc.).
The arrival of generative AI in 2022 marked another milestone. Unlike traditional algorithms, which simply classify data, generative AI is capable of creating original content (text, images, plans, videos, computer code, etc.). This capacity for reasoning and creation requires large amounts of data and significant computing power. Faced with these demands, traditional data centers are reaching their limits, and their architecture must therefore adapt.
Data center design: how are data centers adapting to AI?
Data centers that were originally created to host websites and email are not technically suited to AI workloads. Designers and operators must therefore rethink these infrastructures as a whole. A data center adapted to the needs of these new technologies must have high-performance equipment and appropriate power supply and cooling systems. Depending on usage, a redundancy system may also be considered to ensure service continuity, particularly for real-time inference applications.
AI data center architecture: high-performance design and hardware
For artificial intelligence, servers are equipped with high-performance hardware such as:
GPU (Graphics Processing Unit): This is a processor with hundreds or even thousands of cores. It is capable of processing large data sets simultaneously, making it perfect for deep learning, simulation, modeling, and video special effects applications.
TPU (Tensor Processing Unit): This is a chip created by Google that was designed for neural networks and AI inference. It can perform even faster and more complex calculations than GPUs. However, their use is more limited than GPUs because they only work with TensorFlow-based applications.
AI-enabled switches: Switches are key components in any network infrastructure. They are used to connect servers to each other. With the advent of AI, new switch models have also been developed to handle massive data flows and ensure optimal connectivity without bottlenecks.
These chips consume and generate a lot of energy, not to mention that they operate 24/7. This high power consumption and heat generation require data centers to rethink their design, particularly in terms of power availability and cooling. Infrastructures must invest in power and cooling systems that are powerful enough to accommodate this cutting-edge equipment. In fact, it is often necessary to reinforce floors so that they can support server racks that are becoming increasingly heavy.
Cooling and power consumption
Cooling and energy consumption management are among the major challenges facing data centers, particularly in the age of AI. In a conventional data center, a server rack (a cabinet) consumes between 5 and 15 kilowatts (kW). For AI, this density can exceed 100 kW per rack. At this level, air cooling (traditional air conditioning) becomes completely ineffective. In addition to traditional CRAC-CRAH air conditioning, free cooling, and adiabatic cooling, new techniques such as liquid cooling are therefore becoming necessary. This solution cools the ambient temperature using cold plates installed next to the equipment. Liquid cooling can also be achieved by immersing the components directly in a non-conductive liquid.
Meanwhile, the issue of energy consumption remains central. To power and cool servers, no less than 415 TWh were consumed worldwide in 2024. These figures are only set to increase in the coming years. AI could even double the energy demand of data centers by 2030. The sector is therefore adapting to remain sustainable and limit its carbon footprint. Many solutions are being proposed, such as the use of free cooling (using outside air), the recovery of waste heat to heat neighboring neighborhoods, and the use of biofuels and renewable energies, which make it possible to maintain a PUE (Power Usage Effectiveness) close to 1, a sign of high energy efficiency.
UltraEdge: our edge data centers for all your data
UltraEdge is the leading independent and neutral provider of edge colocation in France. With more than 250 edge data centers across the country, we support businesses in their transformation. This network enables local companies to outsource their IT services without moving away from their operational bases. By hosting your data in our edge data centers, you benefit from:
Low latency : by physically bringing data storage and processing closer to the end user, UltraEdge guarantees optimal responsiveness and efficiency. This is a real asset for disruptive technologies such as 5G, the Internet of Things (IoT), and especially artificial intelligence, which require near-instantaneous information processing.
Neutrality that boosts your performance : as an independent operator, UltraEdge offers a neutral colocation model for edge data centers. This gives your company complete flexibility in choosing your digital services and operators.
Optimal data security : our edge data centers are located in France and therefore comply with French data protection standards and regulations.
Our data centers are equipped with chips capable of supporting all your artificial intelligence-based applications. They are also designed to support ever-larger server racks and to efficiently cool server rooms. Cloud computing, artificial intelligence-based tools, business applications, digital services... whatever your project, UltraEdge offers you the resources you need to develop your business.