AI in ASEAN power systems: momentum is building, but system-wide gains remain untapped

Yet adoption remains fragmented, uneven and largely pilot-based. AI is being applied to specific assets or functions rather than embedded across system planning, dispatch and market design. Without coordinated, system-wide deployment, ASEAN risks capturing only a fraction of AI’s potential economic and emissions benefits.

Under the widespread adoption scenario, AI could deliver up to $67 billion in cumulative cost savings and reduce nearly 400 million tonnes of CO2 by 2035. Current deployment levels fall far short of this threshold.

ASEAN’s expanding digital economy and growing data centre capacity provide a structural foundation for AI adoption. The region’s digital economy is now at around $300 billion and it is projected to reach $1 trillion by 2030.

Data centres are crucial for efficient digitalisation in the power sector, enabling big data analytics, AI, and smart grid operation. The region is actively prepared to adopt AI with large investment in data centre infrastructure, and ASEAN’s data centre market is projected to grow from $14 billion in 2024 to $30 billion by 2030. Major investments from global cloud service providers are also flowing into the region, such as the $1 billion in investments by Google into Thailand’s expanding cloud infrastructure.

Substantial development of digitalisation and data centres position ASEAN countries high in assessments of readiness to adopt AI in the power sector. In fact, several major power systems, including Indonesia, Viet Nam, Thailand, Malaysia and the Philippines, score above the global average on AI readiness indicators, suggesting institutional capacity to deploy AI in public services, including the power sector.

Utilities across these power systems have initiated pilot applications in forecasting, predictive maintenance and optimisation, and successfully gained positive results from AI adoption.

However, infrastructure quality remains uneven across the region, with significant gaps in connectivity, data governance, cybersecurity and interoperability standards. AI applications are concentrated in leading markets, while smaller or lower-income systems risk falling behind.

This uneven deployment creates the risk of a regional “digital divide,” in which countries with stronger digital infrastructure attract greater renewable investment and system efficiency gains, while others capture fewer benefits. Without deliberate regional coordination, ASEAN could develop pockets of digital sophistication rather than an integrated, intelligent power system.

Global evidence demonstrates that AI in power systems can generate substantial efficiency gains.

Deloitte estimates that by 2030, AI-enabled energy efficiency improvements could deliver more than 3,700 terawatt hours (TWh) of energy savings worldwide far exceeding the projected energy consumption of data centres while generating approximately $200 billion in cost savings and reducing 660 million tonnes of CO2 equivalent emissions.

Complementing this, the International Energy Agency (IEA) estimates that integration of AI applications in power plant operations and maintenance could yield up to $110 billion in annual global cost savings by 2035, primarily through enhanced asset performance, predictive maintenance, and reduced downtime.

ASEAN cost savings and emission reduction potential

The potential cost savings and emissions reductions are quantified under two AI adoption scenarios of Deloitte: Baseline and Widespread, which reflects levels of AI preparedness, access and exposure; and two policy pathways: Low-VRE and High-VRE scenarios from IEA’s projections on ASEAN’s power outlook. Details of analysis are stated in Methodology.

On an annual basis, as AI adoption rates increase from 2026 to 2035, from 17.5% to 55% under Baseline adoption scenario and 20% to 64% in Widespread adoption scenario, AI could enable ASEAN to realise increasing savings on power sector costs. Annual savings could rise from about $2.5–3.5 billion in 2026 to around $7–10.5 billion in 2035. These estimates correspond to the Baseline-Low VRE and Widespread-High VRE scenarios, respectively. Cumulatively, this translates into $45–67 billion in cost savings over the period 2026–2035.

In terms of emissions, the largest absolute reductions are observed under the Widespread AI adoption–Low VRE scenario. This outcome reflects the fact that the Low VRE scenario assumes lower penetration of VRE and consequently higher baseline emissions, which are approximately 25% higher than under High-VRE scenario. As a result, AI driven efficiency gains yield greater absolute emissions reductions in this scenario, while scenarios with higher VRE shares exhibit lower marginal emissions benefits from AI.

Overall, AI adoption could reduce ASEAN power-sector emissions by approximately 290 to nearly 400 million tonnes of CO2 by 2035, under the Baseline–High VRE and Widespread–Low VRE scenarios, respectively.

AI becomes more valuable as renewables penetration increases

Our model indicates that AI cost benefits are maximised under scenarios with high penetration of VRE, specifically under the High-VRE scenario, where VRE reaches a 42% share of generation by 2035. This is due to the cost structure of renewable based systems. As renewables entail higher upfront CAPEX and relatively higher O&M costs, cost savings from AI adoption are more significant.

In contrast, the Low-VRE pathway relies more heavily on fossil fuel generation, which is characterised by lower CAPEX and O&M costs but significantly higher fuel costs. Thus, cost savings are moderate.

AI adoption and VRE integration are strongly complementary and mutually reinforcing. High VRE penetration necessitates advanced digitalisation such as forecasting, system optimisation, and asset management which AI can enable more effectively. Conversely, higher VRE shares amplify the operational and planning challenges that AI is best suited to address, thereby realising greater economic benefits from AI adoption.

The cost to deploy AI might be minimal compared to the benefits

Beyond the scale of cost savings and emissions reductions, a key question for policymakers and utilities is whether AI deployment costs are justified by the returns. Although this question is case-specific, initial piloting projects show that deployment costs are modest relative to potential savings when applied strategically at scale.

The cost of deploying AI solutions varies by application and level of model sophistication. For example, deploying a large language model (LLM) typically requires relatively low upfront cost with a capital expenditure (CAPEX) of $50 thousand and operating expenditure (OPEX) $0.0048-0.015 per inference but more advanced analytical or system optimisation models may entail higher costs.

For generators, a standard AI asset management solution subscription including predictive maintenance might cost around $100 thousand but could yield $900 thousand in savings. A US AI service provider claimed that a large utility deployed machine learning models to analyse vibration patterns from transformers, catching bearing failures 3-4 weeks before breakdown and reducing emergency repairs by 60%. This system cost about $2 millions to implement but saved $8 millions annually by preventing outages.

For utilities, the cost to deploy AI might be more expensive especially when the model has to be developed from scratch. Training compute costs can be $50-100 million for frontier models not including R&D and data acquisition costs. However, even these higher costs are small relative to the cumulative cost savings and emissions benefits identified in the preceding analysis, especially when amortised over the lifetime and scale of utility operations.