As AI adoption accelerates across Asia Pacific, the region is facing an urgent need to implement more sustainable AI practices. AI workloads – particularly those running on hyper-scale data centres – are energy-intensive. Given that Asia Pacific has accounted for the largest share of newly added data centre supply over the past decade, ensuring AI systems are sustainable is a growing priority.
AI’s environmental impact is a subject that is well-established, no longer a conversation limited to expert forums and panel discussions. Although AI technologies promise breakthroughs in healthcare, climate science, energy transition acceleration, and more, it’s crucial to address AI’s predicted environmental downsides.
At its core, AI models are trained and run on extremely powerful, energy-hungry computers that rely on electricity that largely comes from carbon-intensive sources. Left unchecked, this could lead to significant carbon emissions.
There are many ways that we can address AI sustainability, but one often overlooked lever we can pull to improve the efficiency of AI workloads is data optimisation, or “data efficiency”. AI models rely on vast pools of data to be effective, but indiscriminately dumping disorganised, irrelevant, or even duplicate data into AI models leads to systems having to do extra work processing excess information. We cannot afford to be wasteful when it comes to AI.
By optimising the data before we feed it to AI models, we can help better manage the environmental footprint of AI. This requires careful forethought and expert planning that looks for sustainability gains along the entire AI lifecycle and prioritises data efficiency when planning AI projects.
How to tackle data efficiency for AI workloads
- Map out your data strategy upfront
Begin by clearly defining what data you need, where it will come from, how often it will be collected, and how it will be processed. Consider if data can be consolidated, stored using low-impact techniques, such as tape or other backup methods, or discarded if no longer necessary. Offloading non-essential data to more energy-efficient storage methods can reduce power consumption.
- Clean up before you start
Data efficiency in AI goes beyond just storing useful data. Data sets should be cleaned and optimised before training a model. Using raw, off-the-shelf data sets or repositories without minimising them results in unnecessary work and inefficiency. Cleaning data upfront ensures the model works more effectively and requires fewer resources.
- Get the training data set right
Data efficiency starts with an optimised data set for training, and using customer-specific data during model tuning helps further refine the model. By ensuring that data is as concise as possible from the start, you set a foundation for efficient processing throughout the entire AI lifecycle.
- Process data only once
Once data is processed for training/tuning, avoid redundant processing. Any additional training or fine-tuning should only occur on new data, minimising repeated energy-intensive operations.
Also Read: How a data-driven approach can optimise decarbonisation in the built environment
- Avoid data debt
Managing data is especially critical for AI workloads due to the massive volumes of data, including unstructured data. One of the key strategies for reducing the environmental burden of data is eliminating inaccurate, erroneous, out of date, or duplicated data. Like technical debt, data debt – where outdated or unnecessary data accumulates – can severely impact AI systems’ performance and sustainability.
- Location matters
Processing data as close to its source as possible minimises the energy required to move it. Optimising data movement reduces both the environmental and time-related costs, ensuring faster, greener AI operations.
As AI becomes integral to industries like manufacturing, logistics, and smart city initiatives, the need for more sustainable AI practices across Asia Pacific becomes more pressing. Singapore is already making strides in this area, with a focus on sustainable data centre innovation and initiatives like the . This approach is critical to ensuring AI systems can scale responsibly.
Asia Pacific’s growing dependence on AI-driven technologies presents a unique opportunity for the region to lead by example. Through initiatives that promote energy-efficient data management and more sustainable AI strategies, Singapore is positioning itself as a global leader in creating sustainable AI ecosystems.
To build a sustainable AI ecosystem in Asia Pacific, organisations must start with clean, lean data. As AI technologies become more embedded across industries, ensuring the data feeding into these models is optimised will not only help reduce energy consumption, but also foster innovation in a way that is more environmentally responsible.
For businesses in Singapore and the wider APAC region, prioritising data efficiency today will help ensure AI’s potential is fully realised without compromising the planet’s future.
—
Editor’s note: e27 aims to foster thought leadership by publishing views from the community. Share your opinion by submitting an article, video, podcast, or infographic.
Join us on Instagram, Facebook, X, LinkedIn, and our WA community to stay connected.
Image courtesy: Canva Pro
The post The hidden barrier to AI sustainability: Why clean data matters appeared first on e27.