Google's e-Conomy SEA Report Pegs Region's Digital Economy at $368B, With Agents as the Next Bet

Google, Temasek and Bain & Company published the 11th edition of their annual e-Conomy SEA report, which puts Southeast Asia's digital economy at $368 billion in gross merchandise value for 2026, up 17% year on year. The headline framing is that generative AI is now a mainstream utility in the region — used by 243 million people, or 58% of the working population, according to the report — and is scaling four times faster than mobile internet did. The more consequential claim for anyone building in the region is the forecast that agentic commerce will drive a 60-fold increase in AI usage over the next five years, measured in tokens.

What the numbers say

The report, titled "Building the Intelligent Economy," breaks the digital economy into five sectors across ten countries: the SEA-6 (Indonesia, Malaysia, the Philippines, Singapore, Thailand and Vietnam) and the SEA-4 (Brunei, Cambodia, Laos and Myanmar). Timor-Leste is excluded, the authors say, because of data quality limits on digital economy and AI metrics.

E-commerce remains the anchor at $226 billion GMV, growing 18% year on year. Digital financial services are measured separately on gross transaction value at $1,650 billion, up 16%. Online travel comes in at $59 billion GMV (+10%), online media at $44 billion in revenue (+22%), and mobility and food delivery at $40 billion GMV (+15%). On that trajectory, the authors say the region is on track to exceed $1 trillion by 2035.

The AI-attributed figures are where more care is warranted. The report says personalized shopping experiences — multimodal discovery and in-app assistants — "organically" drive 10% higher conversions, with order volume uplifts as large as 75%. In online media, it cites a 10% to 20% uplift in quality engagement, and sizes commerce media at $4 billion. The blog post does not explain how these uplifts were isolated from other merchandising, pricing or promotional changes, what the baselines were, or whether "as large as 75%" describes a typical result or a best case. Product leaders reading those figures as a planning input should treat them as directional.

One genuinely interesting datapoint for travel platforms: the report says traditional search engines are still the top tool people use to plan trips, but generative AI has moved past social media into second place for organizing complex itineraries. That is a ranking shift, not a volume claim, and the post does not disclose the margin.

Tokens as the new unit of account

The report proposes tokens as the benchmark for measuring the AI economy — a reasonable framing, since token throughput is what actually drives inference cost, data center demand and energy draw. It then projects a 60-fold surge in regional AI usage over five years on the back of growing confidence in agentic commerce.

A 60x figure is a strong claim, and the blog post does not show the underlying model: no base-year token volume, no assumptions about per-task token consumption as reasoning models get more verbose, no split between consumer and enterprise workloads. Token counts are also an awkward growth metric, because they move with model architecture as much as with human demand. A shift toward longer chain-of-thought responses or agent loops that re-query the same context can multiply tokens without a proportional increase in economic value. Conversely, continued efficiency gains in serving could push cost per task down faster than tokens go up.

The agentic commerce bet, and its conditions

The most forward-looking section concerns consumer willingness to hand transactions to software. The report says 72% of users already rely on everyday automation such as price tracking, and 75% express willingness to let AI agents autonomously compare prices and complete transactions for them. It also reports that 62% trust AI tools more than they did a year ago, citing proven track records (56%), rising familiarity (55%) and output accuracy (53%) as the drivers.

Stated willingness in a survey is not the same as revealed behavior at checkout, and the post does not disclose sample size, country mix or question wording beyond one footnoted trust question. The gap between "I would let an agent buy things" and actually granting payment credentials to one is where most agentic commerce pilots have historically stalled.

The report is at least explicit about the preconditions. Users say they want robust privacy safeguards (44%), transparent information on what AI can and cannot do (36%), and the ability to reverse agentic transactions (30%). That last one is a hard engineering and commercial problem: reversibility implies a dispute and chargeback layer purpose-built for machine-initiated purchases, and neither card networks nor regional wallet operators have a settled standard for it. Merchants, acquirers and agent platforms also have unresolved questions about who bears liability when an agent buys the wrong thing.

Infrastructure and policy are the gating factors

To support the projected usage increase, the report names four conditions. Two are commercial: businesses moving from single-question AI use to multi-step task completion, and inference economics continuing to improve so smaller firms can adopt sustainably. Two are physical and regulatory.

On infrastructure, the authors call for roughly a 3x expansion in data center capacity over five years, with AI workloads taking a dominant share. That is the number most likely to collide with reality. Data center buildout in Southeast Asia is constrained by grid interconnection queues, water availability and land, and the report's own framing acknowledges this by pointing to Singapore's binding Power Usage Effectiveness rules under its Digital Infrastructure Bill and Malaysia's Corporate Renewable Energy Supply Scheme as the kinds of frameworks that should guide new facilities. Efficiency mandates and 3x capacity growth are not automatically compatible; the post does not model the tension.

On governance, the report points to the ASEAN Digital Economy Framework Agreement, due in November 2026, which it describes as the world's first region-wide digital trade pact. The text of the agreement, its enforcement mechanisms and how it treats cross-border data flows for AI training and inference are not detailed in the announcement.

How to read it

This is a sponsored industry report from a cloud and advertising vendor, a state investor and a consultancy, all of which benefit from a bullish regional AI narrative. That does not make the sector GMV figures wrong — the series has eleven years of history and the sector-level numbers are the most checkable part. It does mean the AI-attribution statistics and the 60x projection deserve more scrutiny than the growth rates, and the post provides no methodology appendix to support them.

What to watch

• Whether the 75% stated willingness to use purchasing agents translates into measurable agent-initiated transaction volume on regional marketplaces, and whether any platform discloses that figure separately.
• Whether a workable reversibility and dispute mechanism emerges for agent-completed purchases, given that 30% of surveyed users named it as a precondition.
• The actual text and enforcement provisions of the ASEAN DEFA when it lands in November 2026, particularly on cross-border data transfer for AI workloads.
• Whether data center capacity additions track anywhere near the ~3x five-year path the report calls for, and how grid and water constraints interact with efficiency mandates like Singapore's PUE rules.
• Whether future editions publish base-year token volumes, so the 60x usage projection can be checked against outcomes rather than restated.