AI boom exposes data centre gap across Southeast Asian economies

Companies across Southeast Asia are rapidly expanding the use of artificial intelligence (AI) for everyday business operations, but limited computing infrastructure is emerging as one of the region's biggest barriers to sustained growth.

Businesses are increasingly using AI to recommend products, respond to customer enquiries and detect fraudulent financial transactions.

A survey conducted in February by McKinsey, in partnership with Singapore's Economic Development Board and Tech in Asia, found that nearly half of the 330 companies surveyed across six regional markets had progressed beyond pilot projects. The respondents reported allocating between 11% and 40% of their technology budgets to AI initiatives.

Despite the growing investment, the computing power needed to support AI remains in short supply. Indonesia's Directorate of Digital Infrastructure recorded around 185 data centres with a combined capacity of 274 megawatts at the end of last year, far below the government's target of more than 2,000 megawatts by 2029. The country's data centre association estimates that only about 30% of existing capacity is capable of supporting AI workloads.

Singapore, the region's leading data centre hub, is also facing capacity constraints, prompting operators to look beyond the city-state for opportunities to expand their infrastructure.

Industry experts note that AI services are no longer sold simply through access to machines or monthly subscriptions. Instead, usage is increasingly measured in "tokens" – the units representing the fragments of text that AI models process and generate. Simple customer service responses may consume only a few hundred tokens, while more complex tasks, such as summarising lengthy contracts, require several thousand.

Speaking at a forum in Shanghai, Liu Jian of the Energy Research Institute, which operates under China's state planning authority, said that unlike electricity, where the value of a kilowatt-hour is fixed, the value of a token varies depending on the AI model used to generate the output.