How Singapore's Focus on Data Infrastructure Powers Its AI Expansion: Insights Into Data Centers & Models - ZDNet

How Singapore’s Focus on Data Infrastructure Powers Its AI Expansion: Insights Into Data Centers & Models - ZDNet

Yuichiro Chino/Getty Images

With the adoption of artificial intelligence (AI) fast accelerating , Singapore says it has taken steps to ensure data centers operating in the country are energy efficient and government data used to train models are adequately secured.

The government will establish the necessary computing power and grow the datacenter market in a “sustainable manner” that is in line with its international climate commitments, said the Ministry of Communications and Information (MCI) in a statement.

Also: Every AI project begins as a data project, but it’s a long, winding road

The ministry was replying to parliamentary questions on how the government was balancing the growing demand for computing power with the country’s AI developments and sustainability targets, as well as ensuring that the necessary infrastructure remains environmentally friendly.

Acknowledging that AI compute power from data centers is a critical enabler of Singapore’s national AI strategy , the ministry said one key strategy is to improve the energy efficiency of data centers and drive the development of efficient cooling solutions. It pointed to liquid cooling as one example on which AI computing infrastructure often relies, adding that it is more energy efficient than air cooling for intensive AI workloads.

There also are measures, such as grants, to support operators that work to reduce greenhouse gas emissions from their data centers.

Newsletters

ZDNET Tech Today

ZDNET’s Tech Today newsletter is a daily briefing of the newest, most talked about stories, five days a week.

See all

In addition, Singapore is developing sustainability standards that will pave the way for data centers to operate at higher temperatures and, hence, use less energy for cooling, MCI said.

Singapore released operating standards last June that it said would optimize energy efficiency in data centers located in tropical climates. Developed by the Infocomm Media Development Authority, the recommended standards offer a roadmap to increase data-center operating temperatures to 26 degrees Celsius and above. The standards can potentially yield energy savings of between 2% and 5% for every one degree Celsius increase, the government agency said, citing research from the University of Toronto .

A data center set up to operate in such climates began operating last month, offering a facility for researchers and industry players to develop energy-efficient cooling technologies. Touted as the world’s first data center testbed for tropical environments, the new site is hosted by the National University of Singapore’s College of Design and Engineering, on its Kent Ridge campus. The university is leading the initiative jointly with a fellow local institution, Nanyang Technological University.

Also: Tropical datacenter now up and running in Singapore

Singapore also is reviewing its Green Mark certification scheme for data centers to update the energy-efficiency criteria, MCI said.

“Beyond sustainable AI compute infrastructure, it is important to invest in the development of green computing methods,” the ministry noted. These include coding and algorithm optimization, optimization for both software and hardware, and developing standards for low-data, low-energy AI models.

“The government will continue to deepen international and domestic partnerships with the research community and industry partners on this front,” it added.

Data used for AI models to be appropriately protected

Measures also are in place to manage sensitive information used to train AI models , MCI said in response to questions on how the government ensures confidential data is protected in prompts powered by large language models (LLMs).

“We adopt a risk-managed approach for LLMs [that is] consistent with the existing public-sector framework for handling classified information when using technologies, such as internet-based applications and commercial clouds,” it said.

It noted that highly sensitive applications and data are not accessible online.

“Where use cases involve sensitive data, open source models may be finetuned for use but must be deployed on government servers and computers,” MCI said.

Also: New research initiative aims to build large language AI model for Southeast Asia

Where less sensitive data is involved, the AI models may be owned and managed by commercial and private companies, MCI said. The government’s contracts with these companies are governed by service agreements that include clauses on data handling and security, it noted. These include the non-retention of data and limitations on the use of data to train other products or models.

The government also has implemented technical measures to monitor sensitive data and provide visual cues to remind users of data security policies. In addition, there are governance measures to enforce compliance, MCI said, adding that it would continue to reassess the effectiveness of such measures as technologies evolve.

Singapore last month launched a research initiative to build an LLM that can better meet the demographics of Southeast Asian nations. Dubbed the National Multimodal LLM Programme, the initiative will build on current efforts from AI Singapore’s Southeast Asian Languages in One Network (SEA-LION ), which is an open-source LLM that the government agency said is designed to be smaller, more flexible, and faster compared to LLMs on the market today. SEA-LION currently runs on two base models: a 3-billion parameter model, and a 7-billion parameter model.