Google projects US$368 billion in Southeast Asia digital-economy GMV for 2026

Southeast Asia’s digital economy is projected to reach US$368 billion in gross merchandise value (GMV) in 2026, up 17% year over year, according to Google’s announcement dated October 21, 2026. The forecast comes from the 11th annual e-Conomy SEA report, produced by Google, Temasek and Bain & Company.

For engineering teams, the announcement’s findings on AI shopping agents raise a separate question: what controls should be in place before an assistant can act on a customer’s behalf?

What the headline measures

Google labels the US$368 billion figure as GMV—not platform revenue or a valuation of technology companies. It is a projection for 2026, not a completed full-year result.

The report covers ten countries: Brunei, Cambodia, Indonesia, Laos, Malaysia, Myanmar, the Philippines, Singapore, Thailand and Vietnam. Google’s scope notes say Timor-Leste is excluded because of data availability and quality constraints.

Interest in agents is not deployment evidence

Google reports that 243 million people in Southeast Asia actively use generative AI. Its summary also says 75% of users express willingness to let AI agents autonomously compare prices and complete transactions on their behalf. Stated willingness is not evidence that those users already delegate purchases to agents.

The same summary reports expectations for privacy safeguards, transparent information about what AI can and cannot do, and the ability to reverse agentic transactions. Those findings make transaction control a useful design focus, rather than a reason to assume demand for unrestricted automation.

Design for delegated actions, not just conversation

For teams evaluating agentic commerce, a practical starting point is to define the boundary between recommending an item and committing a purchase. Before enabling purchases, establish explicit authorization, spending limits, an auditable action history and a process for handling mistakes.

Reversibility needs equally precise treatment. For each commerce and payment integration, determine whether an action can be stopped before submission, whether an order can be canceled, and whether a completed payment requires a separate refund request. Explain the applicable limits before the agent acts; do not present these different operations as a single guaranteed “undo.”

For capacity planning, measure cost, latency and failure rates across a completed purchase workflow. If the implementation uses multiple model calls, external API requests or retries, include those steps in the accounting. Regional adoption figures offer market context, but they do not specify the workload a particular service must support.