The Maze: A new report forecasts US$226 billion in ecommerce transactions across ten Southeast Asian countries in 2026, up 18% from 2025. Google, investment company Temasek and consultancy Bain & Company published the outlook on October 21. Their findings on AI shopping describe a separate question: how much purchasing customers will delegate.
Source: Google, Temasek and Bain. The report compares ten-country digital-economy estimates for 2025 and 2026 with projections for 2030 and 2035; ecommerce is the dark-blue series. All values are US$ billions; online media uses revenue.
The Details:
The forecast counts merchandise, not marketplace earnings. Gross merchandise value is the value of goods transacted through the covered channels, before distinguishing what sellers, platforms and other participants keep. It is not platform revenue or profit. The report includes marketplaces, brands selling directly to consumers, groceries and video commerce. It excludes informal commerce because reliable data are lacking. Its ten-country series puts 2025 ecommerce value at US$191 billion, compared with the US$226 billion estimate for 2026. That is a forecast for the full year, not a completed-year sales result. The coverage comprises Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam, Brunei, Cambodia, Laos and Myanmar. Timor-Leste is excluded because of data limitations.
The AI survey covers six markets and existing users. Google and research firm YouGov surveyed 9,006 AI users in July 2026 across Indonesia, Malaysia, the Philippines, Singapore, Thailand and Vietnam. Among that group, 75% said they would be comfortable letting an AI assistant find the best price and buy a product for them. The question assumed the assistant was generally reliable. This is conditional willingness among existing AI users in six countries. It is not evidence that three quarters of Southeast Asia's population already shops through autonomous agents. The report's companion site describes AI as widely used for product discovery and evaluation, while final transactions remain largely human-led.
Permission to buy still comes with conditions. Among surveyed AI users, 44% named strong privacy safeguards as something that would increase willingness to delegate. Clear information about capabilities scored 36%; the ability to reverse actions scored 30%. These are separate survey responses, not shares to add together. Country results also differ: willingness to let AI find a price and buy ranged from 85% in Thailand to 59% in Singapore. Those figures describe stated attitudes. The inspected material does not establish that the gap predicts completed purchases or provide grounds for calling it statistically significant.
MarketMaze Take: The opportunity is substantial, but the forecast and the survey answer different investment questions. A brand outside Southeast Asia can use the forecast to frame research into these ten markets. It should treat AI shopping intentions as a reason to test a purchase flow, rather than as a forecast of orders.
The first commercial test is the handover. The report places AI strongly in discovery and comparison while transactions remain human-led. That makes the point where a shopper approves a purchase a useful test of whether assistance becomes a sale. Privacy, capability explanations and reversibility belong in that test because respondents explicitly ask for them. Success would need actual completed orders, not agreement with a survey question.
A regional average cannot choose a national launch. Thailand's 85% and Singapore's 59% show why one regional enthusiasm figure is an incomplete rollout brief. Willingness may still anticipate future demand. But neither those responses nor the 18% merchandise-value forecast establish how much growth AI caused, what a particular seller will earn, or whether the same behavior exists elsewhere.
Sources: Google, Google, Temasek and Bain report, e-Conomy SEA
