World Bank expert says poor countries should focus on “small AI”

AI is expensive. Processors are expensive, data centers are expensive, energy and water are expensive, data acquisition is expensive. Giants like the US and China can bear these costs. But can other smaller regions — like Southeast Asia, home to the world’s largest group of unconnected people outside of sub-Saharan Africa — keep up?

Still, experts at the Fortune Innovation Forum in Kuala Lumpur, Malaysia, last week were optimistic that smaller countries could invest in AI that works for them, even as they pointed out many of the limitations that still impede greater investment.

“There is an opportunity to really leverage what has come to be known as ‘small AI,’ which is much more targeted, potentially suitable for offline use, and doesn’t necessarily compete with some of the big innovations we’re seeing emerge in larger countries,” said Mahesh Uttamchandani, regional director of digital practices for East Asia, South Asia and the Pacific at the World Bank.

Jon Omund Revhaug, head of Telenor in Asia, agreed that there was “huge opportunity” for smaller countries to invest in sovereign AI.

Countries like Singapore, Malaysia and Thailand are trying to build their own AI industries, whether by encouraging the development of new models more aligned with local conditions, investing in infrastructure like energy and data centers, or creating regulations to maintain data sovereignty.

Still, there is much to be done.

“We need more data centers. We need to build more in Southeast Asia,” said Lionel Yeo, CEO of Southeast Asia at ST Telemedia Global Data Centers.

He admitted that a growing sector of data centers also needs electrical power to operate. “How do we guarantee energy supply from origin to final destination?” he asked. “We need to look at collaboration across the supply chain,” he suggested, and work with “regulators to resolve grid and transmission and distribution issues.”

Water is another limitation. Singapore briefly paused data center construction in 2019 due to concerns about excessive water consumption. The Malaysian state of Johor also warns that water could remain limited until mid-2027, even as it tries to attract new investment in data centers and other infrastructure for AI.

Still, water “opens up an opportunity for cross-border collaboration,” Uttamchandani said. “Not every country will justify having its own data center,” he argued, so resources like water and energy could perhaps be shared between countries.

Talent is another problem. “There are not enough people with the skills needed to set up (servers and data centers). They are not in the right places around the world,” said Wendy Tan White, CEO of Intrinsic.

And some of this work cannot be automated. “One of the biggest problems when setting up data centers is handling cables. At the moment, this is still only done by humans. There is no other way to do it,” he said.

Still, “Asia has an opportunity,” White said. “Right now, (it’s) partly the manufacturing hub, but it’s facing population declines and dealing with geopolitical issues. I think the region can really take a forward stance on regulation and policy.”

Asian governments are starting to act to encourage more investment. Uttamchandani highlighted a recent ruling in the Philippines that eliminated the requirement for legislative approval for new entrants into the telecommunications market. “There is a lot of legacy legislation and regulation that can act as barriers,” he said.

But at some level, supply simply won’t be able to meet demand — leading to a degree of “self-regulation,” Yeo argued. “Everyone is racing to build data centers to support AI, but the infrastructure, talent and energy won’t keep up.”

“Companies will have to find a way to live with the infrastructure and become more efficient to make AI work,” he said.