Transparency Needed as Southeast Asia Pushes Ahead on AI

The speed at which Artificial Intelligence (AI) has proliferated in modern life has exceeded common understanding of its environmental impacts. Currently, the available evidence about the environmental impact of AI use is mixed. To attract investments into data centres and AI, regional governments should consider implementing policies to enhance transparency and encourage the use of green energy pathways.

The Southeast Asia Climate Outlook (SEACO) 2026 Survey Report found that 47.9 per cent of respondents across the 11 ASEAN member states viewed AI as good for the environment. Another 30.2 per cent saw it as bad and 21.9 per cent said they did not know enough to judge (see Figure 1). This uncertainty is not ignorance. It reflects an evidence gap in how AI companies, data centres and governments disclose the energy and water demands of AI.

Figure 1. Varied Views

Uncertainty has not stopped people from acting. Asked about lifestyle changes made for the climate, 17.5 per cent said they had reduced their use of AI tools such as ChatGPT. That is more than the share who cut meat consumption (13.4 per cent) or paid to offset their carbon footprint (11.9 per cent; see Figure 2).

Figure 2. Green Choices

This is the first time in six reports that SEACO posed AI-related questions. But what does the literature say about AI’s environmental footprint, and is cutting back on chatbots the right response?

The demand AI places on data centres is growing fast. In 2024, data centres used 415 terawatt-hours (Twh), roughly 1.5 per cent of the world’s electricity, according to the International Energy Agency (IEA). The IEA projects this will more than double to 945 Twh by 2030, driven largely by AI. This is slightly more than Japan’s total consumption today. Water is the less visible cost. One study estimates that global AI demand could account for 4.2–6.6 billion cubic metres of water in 2027, more than half of the UK’s current annual water consumption.

Estimates about the environmental footprint of individual AI queries are scarce, with few companies disclosing such data. Google is the only major AI company so far to publish its own query-level estimate — putting the median Gemini text prompt at 0.24 watt-hours and about five drops of water. This is equal to watching television for less than nine seconds. These figures are company-reported, cover text only and have not been peer reviewed.

There is no consensus on whether AI’s environmental costs outweigh its potential benefits. That uncertainty makes transparency even more crucial.  The environment-relevant question is not whether every chatbot prompt is wasteful, but whether the infrastructure supporting millions of such prompts daily is being built in ways that align with the region’s climate and resource constraints.

There is no consensus on whether AI’s environmental costs outweigh its potential benefits. That uncertainty makes transparency even more crucial.

With the lack of robust data, it is unsurprising that public views diverge. A 30-country Ipsos survey found Southeast Asians among the world’s most excited about AI, with Indonesia (80 per cent), Thailand (79 per cent) and Malaysia (77 per cent) topping the list, well above the 30-country average of 52 per cent.

The latest SEACO reveals further divides within ASEAN. Respondents in Laos (45.7 per cent), Cambodia (33.7 per cent) and Vietnam (33.1 per cent) were the most likely to deem AI to be “clearly good” for the environment (see Figure 3). Brunei was the outlier, with 35.4 per cent calling it clearly bad and just 1.0 per cent clearly good.

Figure 3. Clear Divide

Educational qualifications were associated with respondents’ answers to the question. Only 11.4 per cent of master’s and 11.8 per cent of doctoral degree holders said AI was clearly good for the environment, compared with 38.0 per cent of those with no formal education. Figure 4 shows that more than a third of degree holders saw AI as somewhat or clearly bad. The education gradient suggests that greater familiarity with AI is correlated with more caution than optimism.

Figure 4. Views on AI’s impact on the environment, by education levels

For individuals, reducing casual chatbot use may have symbolic value, but its direct energy savings are likely modest compared with larger infrastructure choices. The authors have calculated that a typical household air-conditioner (0.70 kW) running for one hour uses 700 watt-hours, the same as close to 3,000 Gemini text prompts. This is especially crucial in Southeast Asia, where the IEA expects air-conditioner units to grow sixfold to 300 million by 2040.

The impact is spatially concentrated, though, with Malaysia being the fastest-growing data centre market in Southeast Asia. In 2019, Singapore paused the growth of new data centres to review how to manage these land-, water- and energy-intensive facilities in line with her climate commitments. Operators moved to Johor and Batam, and Johor’s capacity grew from 10 MW in 2021 to over 900 MW by end-2025. In Malaysia, data centres could make up as much as 31 per cent of peak demand by 2035. Previously, developing countries such as Malaysia chased data centre foreign investments; now, these centres are becoming sources of risks, given power and water constraints as well as geopolitical exposure.

Policy Implications

With over one in five SEACO respondents unsure of AI’s environmental impact, and the more educated respondents expressing greater scepticism, public trust will hinge on transparency. The EU already requires data centres to report their energy and water use every year. ASEAN governments courting AI investment should do the same.

Respondents cutting back on AI use are responding to a legitimate concern but the issue cannot be addressed through individual behaviour alone. The more significant choices lie with governments, data centres and AI companies. For ASEAN economies seeking AI investment while meeting climate goals, the ASEAN Guide for Sustainable Data Centre Development recommends energy efficiency and clean-power procurement, transparent emissions reporting, siting and cooling choices that reflect local water stress, and circular use of materials and equipment. But its targets remain voluntary, and the real test is whether member states follow through on the indicated areas. The public uncertainty shown in the SEACO findings should not be dismissed but treated as a signal that the region’s AI transition needs clearer rules and accountability.

2026/294