Scaling AI sustainably: Four missteps leaders must avoid

ARTIFICIAL intelligence (AI) is scaling at breakneck speed across Asia – but the region’s infrastructure, energy systems and sustainability frameworks are struggling to keep up. 

In August 2025, Singapore introduced a new data centre standard to improve IT energy efficiency in tropical climates, targeting at least a 30 per cent reduction in energy use through smarter equipment selection and optimised operations.

China is experimenting with underwater data centres to cut cooling costs, while markets such as Indonesia and Malaysia are tightening environmental, social and governance (ESG) reporting rules. 

The message is clear: AI can no longer grow on an unsustainable foundation. To scale AI responsibly, business and technology leaders in Singapore and Asia-Pacific should avoid these four mistakes: 

1. Choosing oversized models for simple tasks 

Many chief technology officers and engineering heads are embracing cutting-edge, high-capacity AI models to harness the full potential of emerging innovations.

However, running inference – the process where the AI model actually generates an answer or prediction – on massive models can consume 10 to 100 times more energy a query than smaller, optimised alternatives.

Using a 175-billion parameter model to classify customer emails or extract basic data points is like using a freight truck for grocery shopping.

Consider whether your use case truly requires cutting-edge capabilities, or if a smaller, more efficient model, or even traditional machine-learning could deliver similar results with a fraction of the environmental cost. The smartest AI leaders focus not on size, but on suitability. 

2. Ignoring energy-efficient infrastructure and deployment 

Chief information officers and IT leaders play a pivotal role in determining AI’s environmental footprint. Running the same workload in coal-powered data centres versus renewable-backed ones can result in 10-times differences in carbon footprint.

SEE ALSO

Google Deepmind launches research lab in Singapore to better study S-E Asia’s cultural nuances

Google opens new AI engineering centre in Taiwan

Similarly, failing to implement model optimisation techniques such as quantisation (a process that decreases memory and computational requirements), pruning (removing less important or redundant parts of a model), or knowledge distillation (a machine-learning technique) means the models consume unnecessarily high amounts of energy during inference. 

Smart organisations audit their data centre or cloud providers’ renewable energy commitments and optimise model architectures for efficiency, not just accuracy. You should ensure that sustainability is a core part of your infrastructure strategy, not an afterthought. 

3. Neglecting data management and storage practices 

Chief data officers and analytics teams often overlook how storing massive datasets indefinitely without governance policies leads to unnecessary energy consumption from data centres. 

Data is the fuel of AI, but poor data hygiene creates enormous hidden costs. Many organisations collect and store vast amounts of data “just in case”, without clear lifecycle policies.

Redundant, obsolete or trivial data still requires energy to store, back up and maintain. Implement data governance frameworks that regularly assess data value, establish retention policies and use compression techniques.

Consider whether you need to store raw data permanently or if processed, smaller datasets would suffice for your AI applications. 

4. Failing to measure and monitor AI’s complete sustainability impact 

The responsibility for measuring AI’s full impact often falls between the chief financial officer, chief sustainability officer and technology teams. Yet without clear measurement frameworks, organisations lack the visibility needed to optimise or justify further investment. In short, you can’t manage what you don’t measure. 

It’s time to stop treating AI sustainability as an afterthought. The lack of visibility into the energy use and carbon emissions of AI systems makes it hard to optimise performance or track sustainability goals. 

Building measurement and monitoring your AI operations should start from day one. Implement tools that monitor key efficiency metrics, such as energy consumed per task, carbon emissions per interaction, and processing speed per kilowatt-hour.

Tracking both environmental indicators, such as renewable energy usage and carbon reduction, and productivity metrics such as time saved and process efficiency, can provide a complete picture of AI’s return on investment.

This dual-focus approach enables continuous improvement and helps justify investment in more efficient, sustainable AI practices. 

Building sustainability into AI from the start 

Ultimately, AI sustainability is smart business – going beyond carbon emissions and environmental risk mitigation.

More efficient AI systems reduce operational costs, improve performance, and future-proof technology investments against increasing energy costs and climate regulations. By avoiding these common mistakes, organisations can harness AI’s transformative power while building a more sustainable digital future. 

The key is to embed sustainability thinking from the start, not bolt it on later. That is how businesses can gain a significant competitive advantage tomorrow.

The writer is senior director and general manager, advisory and professional services, at HPE Asia-Pacific