Singapore’s next data-centre growth wave lies in AI inference: Digital Realty

[SINGAPORE] As Singapore grapples with tight land and power supply, Digital Realty said the next phase of growth for its data centres will lie in artificial intelligence (AI) inference – the process of running AI models for real-world applications.

Serene Nah, managing director and Asia-Pacific head at the data centre operator, told The Business Times that Singapore is increasingly positioning itself as a regional hub for AI inference.

Unlike larger facilities that are focused on model training, AI inference data centres require less computing power and space. However, such infrastructure is sensitive to latency, meaning that it needs to be located nearer to end-users to minimise delays when accessing services.

Lower latency allows users to access AI tools, such as ChatGPT, almost instantly – which is especially important for business hubs including Singapore and Tokyo, Nah noted.

“You actually need to be closer to where users are,” she said.

Digital Realty is listed on the New York Stock Exchange, and has over 300 data centres across more than 55 metropolitan areas globally. It is also the sponsor of Singapore Exchange-listed Digital Core Real Estate Investment Trust .

The heavy computing needed to train AI models can be offloaded to data centres in adjacent cities such as Johor and Batam, where land and power are more readily available, Nah said.

Yet, Singapore and Japan remain the top markets in the Asia-Pacific for data centre demands, she noted, with vacancy rates for Digital Realty’s facilities in both markets being “below single digits”.

To tap rising demand, the data centre operator in February unveiled its first regional innovation labs in Singapore and Tokyo. The labs showcase the latest development in AI infrastructure, including high-density racks designed to support AI workloads.

The data sovereignty advantage

As neighbouring countries such as Indonesia and Malaysia tighten regulations on where sensitive data must reside – complementing Singapore’s already established data sovereignty rules – some market observers worry that regional data flows could fracture into smaller and less efficient silos.

However, Nah argued that strict data sovereignty laws will bring about a net-positive effect to the region. “With that clarity, it becomes a good guideline for companies, and also for the country, to really start driving policies and also... guard rails around that.”

With clearer regulations, companies will be more mindful of where their data resides – serving as a catalyst for workload “bifurcation”, where sensitive data will reside in local data centres, while non-sensitive backup data will be routed to neighbouring facilities.

“(Data sovereignty) is going to become a… bigger theme, especially with AI because you have AI training and you also have private enterprise training their own workloads,” Nah said.

But while this could change how data is stored, it is unlikely to lead to smaller facilities. “Smaller data centres tend to be less economical and less productive to operate,” she noted.

Instead, she believes that the future of data centres lies in being multi-tenanted – many companies located within the same facilities.

But she added that managing multi-tenanted infrastructure requires a fundamental shift in how the data centre industry is viewed, as more niche skill sets will be required to handle a more complex data centre environment.

“I think it’s important to start seeing data centres as not just a real estate play, but (also) a technology layer and critical infrastructure layer.”