BCG-Temasek Says AI Could Unlock $600 Billion in Climate and Sustainability Value by 2028

For most of the last two decades, sustainability and financial performance were treated as forces in tension. Boards approved green initiatives as a cost of doing business, a line item justified by reputation, regulation, or conscience rather than by returns. The working assumption was that every dollar spent on lowering emissions was a dollar that could have gone elsewhere. A report from Boston Consulting Group and Temasek, produced through Temasek's Ecosperity platform, challenges that assumption for a specific class of applications. Where artificial intelligence meets climate and sustainability challenges, the report argues, financial returns and environmental outcomes are not in conflict. They move together by design.

The mechanism is straightforward. Sustainability is fundamentally about resource efficiency, achieving the same or better outcomes with less energy, fewer materials, and less waste. AI's core capability is optimizing how resources are used. When AI reduces the energy a process consumes, it lowers both the operating cost and the emissions at the same time. The dollar saved and the ton of carbon avoided come from a single action. That shared mechanism is what allows the authors to group sectors as different as cement production, insurance underwriting, and classroom instruction inside one investment landscape.

 

AI's investment potential

 

Deploying current AI capabilities across climate and sustainability sectors could generate roughly 600 billion dollars in annual global value by 2028. The report is precise about what that number represents. It is a directional indicator of opportunity scale, not a forecast of realized returns. It assumes broad deployment of AI tools that are already commercially proven, adopted at rates consistent with current trajectories. It is also deliberately conservative, capturing only the slice of each sector's AI opportunity where efficiency gains produce both financial and measurable environmental or social outcomes. The broader AI opportunity in industrial systems, financial services, and health care is substantially larger and sits outside the report's scope.

To build the estimate, the analysts assessed more than 40 subsectors across three domains: climate and energy transition, natural capital and resource management, and social systems and livelihoods. They scored each subsector on market attractiveness and AI impact, then sorted them into priority, attractive, and opportunistic tiers. Five priority subsectors account for around 423 billion dollars of the total:

• Industrial equipment and systems efficiency: ~$300 billion
• Climate risk modeling (including insurance): ~$75 billion
• Grid, storage, and system flexibility management: ~$32 billion
• Inclusive education: ~$13 billion
• Materials discovery: ~$3 billion
  
  

AI solution providers capture only an estimated 15 to 25 percent of the value they help create. The majority flows to the businesses deploying the tools, including utilities, insurers, and manufacturers, through lower costs, better asset performance, and new revenue. For investors, this points to a counterintuitive conclusion: the most defensible returns may sit with established operators who control the data, distribution, and physical assets, rather than with the AI vendors themselves.

 

Industrial applications

 

Industrial activity, covering the extraction, processing, and manufacture of physical goods, accounts for roughly a quarter of global carbon emissions. These are continuous, energy-intensive processes that waste heat, scrap material, and expose workers to hazardous conditions. AI reduces those inefficiencies, and because the wasted energy is both a cost and an emission, eliminating it produces a financial saving and an environmental gain simultaneously.

Industrial equipment and systems efficiency is the largest priority pool in the report, at roughly 300 billion dollars in annual value. Predictive maintenance accounts for 79 billion, adaptive process optimisation for 59 billion, energy management systems for 56 billion, and quality control for a further 47 billion. The supporting cases are drawn from deployed systems rather than projections. One of the world's leading cement manufacturers used AI-driven sensor analytics to catch developing gear faults and lubrication contamination before a machine failed, avoiding a catastrophic gearbox failure and preventing over 500,000 dollars in costs. Japan's Tokuyama Cement deployed real-time kiln optimisation and cut thermal energy consumption by 3 per cent while reducing manual operator interventions by 70 per cent. A steel manufacturer using an AI energy platform reported 14 million dollars in annual energy savings and a 40MW reduction in monthly utility charges.

The environmental results follow from the same interventions. The report estimates that AI applied across these industries could cut Scope 1 and 2 emissions by around 0.6 gigatons a year, roughly equivalent to the total annual emissions of Germany. In steel alone, AI-driven safety systems could prevent approximately 20,000 workplace injuries per year. The report counts only applications that demonstrably reduce energy, emissions, material waste, or workforce harm. Scheduling optimisation that improves throughput without lowering resource intensity is treated as an operational improvement and excluded.

 

AI in climate-risk modelling

 

Climate risk modelling is the report's clearest example of AI generating value that did not previously exist. The economic cost of climate hazards is now concrete. Between 2000 and 2025, the frequency of billion-dollar natural disasters more than doubled, and insured losses from natural hazards exceeded 100 billion dollars in both 2023 and 2024. Traditional catastrophe models, built on historical data and coarse geographic zones, increasingly fail to capture a climate that is shifting beneath them.

AI improves the resolution of risk assessment by integrating satellite imagery, sensor networks, and atmospheric models to produce asset-level, forward-looking estimates that older methods cannot generate. The report values this opportunity at roughly 75 billion dollars annually across three applications. Operational hazard intelligence, worth around 30 billion, converts weather data into site-specific triggers; one platform predicted the location and number of utility customers affected by a major hurricane within 3 percent accuracy a day before landfall, allowing utilities to pre-stage repair crews. Asset-level risk analytics, another 30 billion, lets insurers underwrite individual properties rather than blanket zones, and one US insurer cut its combined ratio, the core measure of underwriting profitability, by 4.4 percent in the first year of deploying an AI risk platform. Portfolio climate stress testing, the earliest-stage application, makes up the remaining 15 billion.

Better risk pricing is usually treated as a defensive measure that helps insurers avoid losses. The report identifies a second effect. More accurate, asset-specific scoring could expand insurance coverage to an estimated 15 to 20 million additional policies in regions and hazard types previously deemed uninsurable. AI does not only protect the insurer's margin. It extends protection to communities that older models had excluded.

 

Grid and storage applications

 

The electricity system is under greater strain than at any point in its history, pressed by electrification, urbanisation, data centre demand, and a rising share of variable renewable generation. The traditional response was physical: build more transmission, deploy more storage, and connect more renewables. AI shifts the value toward orchestration, extracting more from assets already installed.

The deployed results are notable because they require no new construction. A battery storage fleet running AI-driven dispatch optimisation has earned 25 to 30 per cent more revenue from the same installed hardware. A transmission network using AI-enabled dynamic line ratings can unlock 10 to 30 per cent of additional capacity. A US utility facing wildfire exposure used a physics-based digital twin to reclassify 80 per cent of its transmission poles based on actual condition rather than incomplete records, projecting an 8 per cent reduction in average outage duration and recovering more than 5 million dollars in annual revenue from unauthorised attachments and clearance violations. The total value pool here is around 32 billion dollars. AI-optimised dispatch displaces fossil-fueled peaker plants while raising storage revenue, predictive monitoring cuts wildfire ignition risk while extending asset life, and better demand forecasting reduces renewable curtailment while lowering balancing costs. In each case, the same intervention makes the grid lower-carbon, more reliable, and more profitable.

 

Education and materials applications

 

Two further deep dives extend the argument. Inclusive education qualifies under the same resource-efficiency test: AI optimises teacher capacity and instructional time, and the social and financial returns flow from the same intervention. In low-income countries, primary school completion stands at 43 per cent against 99 per cent in high-income economies, with pupil-to-teacher ratios of 71 to 1 versus 17 to 1. An adaptive learning platform deployed across more than 15,000 government schools in India reached 2 million students at 20 to 25 dollars per child per year and delivered between 22 and 23 months of learning progress in 17 months, more than five months of acceleration relative to peers outside the program. The report puts the directly measurable commercial value for private schools at around 13 billion dollars and states that this figure is conservative, since it excludes the value of redeployable teacher capacity, improved learning outcomes, and expanded access.

Materials discovery is the smallest near-term pool, at roughly 3 billion dollars, split across battery chemistry at 1.1 billion, agricultural and biological materials at 0.9 billion, carbon management materials at 0.6 billion, and pharmaceutical applications at 0.4 billion. AI compresses the cost and timeline of laboratory R&D; one company cut lab-to-production timelines for next-generation battery cathode materials by over 90 percent. The authors describe materials discovery as a long-term investment area adjacent to industrial decarbonization and renewable energy, a position in the enabling science rather than the deployment layer, where breakthroughs in battery chemistry, carbon capture, or nitrogen-fixing microbes could reshape entire industries.

 

Limitations and uncertainties

 

The report does not overstate its case. AI carries its own resource demands, with compute-intensive models drawing significant electricity and data centers placing pressure on land, water, and local energy infrastructure. The authors frame the relevant test as whether the systems AI optimizes consume fewer resources in aggregate than they do today, and they treat this as an empirical question to be evaluated sector by sector and revisited over time rather than a settled conclusion.

Several constraints could slow adoption:

• Validation and trust: AI is difficult to validate in regulated, high-stakes settings, where users need to understand why a model reached its conclusion before acting on it.
• Data quality: Operational data is often fragmented, siloed, or incomplete, particularly in emerging markets and legacy industrial environments.
• Model drift: Models trained on historical conditions lose accuracy as weather, equipment, and market rules change, which makes continuous monitoring and retraining a core operational requirement.
• Liability: Responsibility for an AI recommendation that contributes to a failure must be clearly assigned.
• Regulatory variation: Disclosure rules and AI governance frameworks are evolving at different speeds across jurisdictions.
  
  

None of these constraints is unique to climate applications. What distinguishes the space is the strength of demand: hazard losses, energy waste, grid strain, and learning gaps are growing more urgent, not less.

The conclusion is direct. AI will not solve climate change, and it will not replace the physical infrastructure, policy frameworks, and capital these challenges require. What it does is improve how scarce resources are used, and in the applications examined here, it delivers financial value and sustainability outcomes from the same intervention. That alignment, rather than any single dollar figure, is the report's central finding: in sectors where inefficiency carries both a financial and an environmental cost, profit and sustainability point in the same direction.

Source: The private capital opportunity in AI-enabled climate and sustainability sectors

Full report here