Southeast Asia Has 243 Million GenAI Users and Almost No Paying Market

On 21 October 2026, Google, Temasek and Bain and Company reported that 243 million people in the six largest Southeast Asian economies use generative artificial intelligence, equal to 58 per cent of the working population. The figure was reached in under three years. Only 1 to 2 per cent of the population in those markets pays for generative AI.

That pair of numbers is the story. Southeast Asia has solved distribution for generative AI. It has not solved payment. The next phase of the region's AI economy will be decided by pricing, workflow and trust, and by whether companies can turn a free habit into a service that a consumer or a business will pay to keep.

The report, e-Conomy SEA 2026, counts a generative AI user as someone who used a standalone platform in the previous 30 days. Embedded AI inside search, shopping and messaging apps sits outside that count. The definition matters. It means 243 million is a measure of deliberate trial and repeat use of a dedicated tool, and the paying share still looks small against that strict base. The report also measures depth in tokens, the units a model processes when it reads a prompt and produces a reply. On that measure, Southeast Asian use per person stands at about 10 per cent of United States levels.

The consumer numbers underneath that gap are strong. Half of generative AI users in the six markets use it weekly. Almost a fifth use it daily. 69 per cent of consumers use AI to compare product options. 93 per cent say they trust AI recommendations. 75 per cent say they are open to an AI agent finding the best price and making a purchase on their behalf.

Those figures describe willingness. They do not describe a transaction. In the shopping journey covered by the report, most users still verify a recommendation before they act. Trust is rising, with 62 per cent of consumers saying they trust AI tools more than a year ago, but trust in a suggestion is a weaker commitment than handing over payment, a password or a business process. The paying test is where the region's AI story gets specific.

Free trial scaled faster than paid value

Generative AI reached 243 million users in Southeast Asia in under three years. The report contrasts that pace with the internet, which took more than a decade to reach similar penetration. Free and freemium tiers, minimal onboarding and local language expansion removed the usual friction.

The sector totals show the same pattern at market level. Southeast Asia's digital economy is projected to reach US$368 billion in gross merchandise value in 2026, up 17 per cent year on year, and to pass US$1 trillion by 2035. Ecommerce accounts for US$226 billion, up 18 per cent. Online travel is projected at US$59 billion, up 10 per cent. Online media revenue is projected at US$44 billion, up 22 per cent. Mobility and food delivery are projected at US$40 billion, up 15 per cent. Digital financial services are measured separately, at US$1.65 trillion in gross transaction value, up 16 per cent, with digital lending balances of US$116 billion, up 22 per cent.

Gross merchandise value and gross transaction value measure value traded through platforms. They are company revenue in no sense, and they are not profit. A large traded value can coexist with thin paid AI revenue when AI sits inside a free consumer journey, improves discovery and conversion, and is monetised indirectly through ads, commissions or seller tools. The report cites order volume uplifts as large as 75 per cent in cases where AI supports discovery, conversational assistance and personalised recommendations. Those cases show what AI can do inside a transaction. They do not show that users will pay a separate subscription for the tool that helped them.

Country detail sharpens the point. Singapore leads on generative AI penetration at 70 per cent of the working population, about 3.2 million people. Its digital economy gross merchandise value is put at US$31 billion in 2026, up 8 per cent, with a projection of US$59 billion by 2035. Malaysia follows at 68 per cent, Vietnam at 65 per cent, Thailand at 63 per cent, Indonesia at 54 per cent and the Philippines at 53 per cent. Penetration is high across the group. Payment is low across the group.

Enterprise use shows where paid demand should come from first. Consumers account for about 60 per cent of regional AI token use. Enterprise workloads form the bulk of use only in Singapore and Malaysia, the two most mature markets in the report. In Indonesia, the Philippines and Vietnam, consumer applications account for about 80 per cent of AI use. Paid AI revenue in those markets must therefore come from workflows that are still largely personal, mobile and free, or it must wait for enterprise buying to catch up.

The awkward part is the price, and the workflow behind it

The complication is plain. A user who will not pay for generative AI today may still be a valuable user for an advertising or commerce platform. That user is a weak foundation for a standalone AI business that needs subscription revenue, inference costs covered and a margin left over. The report's own projection makes the condition explicit. It says regional AI use could rise to 60 times current levels by 2030, but only if agentic adoption accelerates, inference becomes more efficient, infrastructure scales and unit economics become sustainable. A projection with four conditions attached is a research agenda and a pricing test, and it should not be quoted as a forecast on its own.

Capital is already moving, but it is moving unevenly. Digital economy deals totalled US$8 billion, in line with the three year run rate. AI focused companies captured 80 per cent of private deployment in the last six months, up from 30 per cent in the second half of 2024. Investment in digital infrastructure reached US$20 billion in the last 12 months, led by the US$11 billion acquisition of ST Telemedia Global Data Centres. Live data centre capacity in the region stands at about 4 gigawatts, and the report says meeting projected demand requires roughly a threefold expansion over five years.

Those two totals should not be added together as if they were the same flow. The US$8 billion figure describes digital economy deals. The US$20 billion figure describes infrastructure investment over a different window and includes a large acquisition. Announced hyperscaler plans, which stretch over several years, sit in a third category. Money is still being committed to capacity, power and sites before the paid demand that would fill that capacity has been proven at scale. Who will pay to use that capacity, at what price, and for which workflow? Labour data point the same way. The International Monetary Fund, in its October 2026 Regional Economic Outlook, found demand fell by nearly 10 per cent in occupations more exposed to AI, yet only about 6 per cent of firms ask for AI skills in job postings.

Why that may not matter much

1. Reach is a real asset, even when it is unpaid. 243 million users, weekly use by half of them and daily use by almost a fifth, give a platform a distribution channel that a new paid product would take years to build. Conversion becomes the work. A company that already sits inside search, shopping, travel booking or payments does not need to find the user again. It needs to find the moment when the user will pay for speed, accuracy, control or a completed task.
2. Investors are funding the application layer as well as the concrete. AI focused companies took 80 per cent of private deployment in the last six months. Public listing proceeds rebounded to more than US$8.5 billion in the last 12 months, led by Singapore and Malaysia, across digital infrastructure, digital applications and healthcare. Data centre capacity additions exceeded 52 per cent a year, with low vacancy in third party operated facilities. That pattern is consistent with a bet that depth will follow reach, and it gives builders time to test paid tiers, ads and agent executed transactions.
3. Trust and automation habits are already in place for a limited first step. 72 per cent of AI users already use everyday automation such as price alerts or automatic payments. 62 per cent say they trust AI more than a year ago. Ecommerce is the natural testing ground because the task is bounded, the price is visible and the result can be reversed. The report's openness figure, 75 per cent willing to let an agent buy at the best price, is permission to test that bounded case. It is proof of nothing about high value or irreversible decisions.

The real test is narrower than the headline projection. Will the paid share move from 1 to 2 per cent to a level that covers inference and support costs? Will enterprise token share rise in Indonesia, the Philippines and Vietnam, where consumer use still accounts for about 80 per cent? Will job postings show AI skill demand spreading beyond the 6 per cent of firms that already ask for it? Nobody can answer those questions from a free user count. Each one can be checked in the next edition of the same reports.

What to watch

First, the paid share and the token gap. If paid use stays at 1 to 2 per cent while token use per person stays near 10 per cent of United States levels, the business model story has not changed, however large the user count grows.

Second, the enterprise split by country. Singapore and Malaysia already show enterprise workloads as the bulk of use. Movement in Indonesia, the Philippines and Vietnam, away from about 80 per cent consumer use, would be the clearest sign that companies are buying AI as a workflow tool, and that employees are using more than personal accounts at work.

Third, capacity against occupancy. Live capacity of about 4 gigawatts must roughly triple in five years on the report's projection. Low vacancy today supports that build. Vacancy, power availability and the price of inference will show whether the US$20 billion infrastructure wave is being filled by paid enterprise demand, by consumer traffic that is still largely free, or by capacity waiting for a market that has not arrived yet.

Southeast Asia did not fail at AI adoption. It adopted generative AI faster than it adopted the mobile internet. The failure to avoid now is to mistake that adoption for a market. A market needs a price, a buyer and a repeat purchase. Southeast Asia has the users, the habit and the infrastructure plan. The paying customer is still the work to be done.