S’pore announces plans to tighten data centre rules at launch of AI test-bed facility by Singtel, Nvidia

SINGAPORE - Singapore is drawing up regulations for the data centres and cloud service providers (CSP) powering its digital economy, while it welcomes a new test facility for cutting-edge compute backed by the world’s top chipmaker Nvidia.

The proposed Digital Infrastructure Act (DIA) will require that such providers meet power efficiency standards, and have in place cybersecurity and service disruption countermeasures. They will also be required to log incident reports.

To be tabled later this year, the legislation will impact both existing and future facilities, said Senior Minister of State for Digital Development and Information Tan Kiat How at the launch of Singtel and Nvidia’s new Centre of Excellence (CoE) for Applied AI on Feb 24.

“It would be irresponsible for the Government to take a hands-off approach and leave it entirely to commercial arrangements between the CSPs, data centre operators and their enterprise customers, when there are spillover implications to the broader economy, society, national security and Singapore’s international reputation,” he said.

Existing operators will have time to comply as they refresh or retrofit equipment.

Singapore hosts more than

1.4 gigawatts of capacity across more than 70 data centres

, but these land-based and energy-guzzling set-ups have come under focus as the island’s demands for compute – or processing power – soar alongside its artificial intelligence (AI) ambitions.

The new energy efficiency requirements proposed – set in consultation with the industry – will be “ambitious but practical and consistent with international benchmarks”, said Mr Tan, who urged businesses to support the Bill when it gets tabled.

Singapore has in place regulatory frameworks under its Digital Connectivity Blueprint for subsea cables, fibre networks, compute capacity and emerging technologies, but relies on advisory standards for data centres and cloud service providers.

“The DIA will be a major step to ensure that Singapore’s digital infrastructure is sustainable, resilient and cybersecure, and importantly able to meet future demands,” Mr Tan said.

There is room for Singapore to deploy more advanced and efficient chips, said Nvidia’s senior vice-president of solutions architecture and engineering Marc Hamilton. Singapore’s current compute capacity features significant conventional computer processing chips, he said.

“For Singapore, a country that doesn’t have a lot of land, using a chip that is 50 times more energy efficient than its predecessor really democratises data centre space.”

For its tie-up with Singtel, Nvidia is bringing more than just hardware to the table, Mr Hamilton added.

It will offer users the chip giant’s full AI stack, including its Nemotron family of large language models, as well as orchestration and optimisation software.

Users will also have access to Nvidia’s ecosystem, which includes over 40,000 AI start-ups, some of which are based here.

Singtel’s Digital InfraCo chief executive Bill Chang, who oversees the telco’s digital infrastructure arm, said the new centre will be housed in the Punggol Digital District, which has been earmarked as a living laboratory for autonomous tech.

By June, the centre will be able to offer enterprises and public agencies a “micro-grid” to trial AI pilots on Nvidia’s latest chips while keeping sensitive data within Singapore’s borders, Mr Chang said.

Government agencies, banks, healthcare providers, transport operators and researchers are among prospective users, with some already signed up, he added.

Mr Chang is also gunning for the new facility, which will cost about $5 million annually to run, to churn out local applied-AI professionals.

It will also be used to validate new data centre designs that can handle much higher rack-level power densities and cooling requirements as more powerful AI chips, including new Nvidia chips in the pipeline, enter the market in the following years.

This is as AI workloads rise with the use of machines at the edge. Robotics and humanoids, for example, will burn through compute power for language, computer vision and spatial recognition at much higher rates compared with agentic or generative AI.

“It’s almost three times the GPU consumption,” Mr Chang said, referring to graphics processing units.

Singtel is also preparing for the advent of edge AI – the deployment of AI algorithms and AI models directly on devices – which requires low latency, fast and reliable networks.

This means, Mr Chang said, being able to tell customers, down to milliseconds, how fast a given AI task will be completed given its endpoints into the system.

“The power of networks and AI coming together and to innovate the ecosystems around that – I do not believe any Centre of Excellence has done it,” he said.