How to design server farms for the AI age? Architectural firms are bracing for complex challenges

As artificial intelligence (AI) advances and language models grow increasingly sophisticated, the demand for large-scale server farms will escalate. The explosion in computational requirements for cutting-edge language models has transformed the landscape of data centre architecture.

"Key architectural challenges include managing heat generated by high-density AI workloads and ensuring energy efficiency,” said Sanjeev Bhandari, Founder & CEO of AirBrick Infra, adding that traditional cooling systems are nearing their limits, prompting the adoption of advanced liquid cooling technologies, such as direct-to-chip cooling, to handle escalating heat densities.

CareEdge Ratings, in its recent report, highlighted India’s data centre capacity is expected to double in three years from 0.9 GW (Gigawatt) in 2023 to around 2 GW in 2026. Further, the additional capacity has created substantial investment prospects due to the estimated capex requirement of Rs 50,000 crore in the next three years.

Thermal management is a significant challenge. The dense configuration of servers generates intense heat, which can jeopardise the equipment’s performance and lifespan. Another big issue is managing the colossal energy consumption associated with these server farms.

ALSO READ | Skill-first hiring emerges as a game-changer but challenges remain

By 2030, Arm Holdings CEO Rene Haas said that the world’s data centres are on course to use more electricity than India, the world’s most populous country. Finding ways to head off that projected tripling of energy use is paramount if artificial intelligence is going to achieve its promise, he said.

Industry experts say the size and energy demands of server farms have significant ecological footprints. They suggested that urban planners and utility providers must be carefully coordinate to ensure that server farms are seamlessly embedded into their environments.

Tackling challenges 

As language models continue to advance, the challenges faced by architectural firms in designing and maintaining server farms will only grow, experts say. Hence, the industry is rising to these challenges with innovative solutions and forward-thinking designs, ensuring that the infrastructure supporting the next generation of AI remains robust and efficient.

“In terms of power, the need for uninterrupted supply is leading to advancements in energy storage solutions, including more efficient batteries and flywheel energy storage systems. Microgrids, which integrate renewable energy sources like solar and wind with battery storage, are being developed to provide reliable, off-grid power for data centres,” said Keith Menon, architect and co-founder of Spiro Spero.

ALSO READ | Global Fintech Fest 2024: India's fintech boom marred by talent shortage

Additionally, he said the push for sustainability is fostering innovations in renewable energy integration and energy harvesting from waste heat. These developments aim to address the increasing power demands of AI while minimising environmental impact and ensuring continuous operation.

High-speed data transfer and low-latency connections are essential for the performance of language models and other computational tasks. Network infrastructure must be robust and capable of handling enormous volumes of data traffic.

“Architects are designing data centres with high-density fibre optic cabling and implementing advanced networking technologies to minimise latency. Multi-path networking and software-defined networking (SDN) are also being used to enhance connectivity and manage data flow more efficiently,” said an Indore-based architect Prateik Patidar.