The transformative potential of Artificial Intelligence (AI) is undeniable globally and especially in the Asia Pacific. In Singapore, the rapid rise of AI could become one of its biggest economic catalysts. According to EY’s recent report, AI-driven solutions are projected to unlock up to S$198.3 billion by 2030, close to 30% of the nation’s 2023 GDP.
Lurking behind the scenes is the uncomfortable truth that these powerful models take staggering amounts of energy to train and deploy large language models (LLMs). With energy restrictions targeting data centers and AI workloads mounting, developers and businesses are facing the reality that we must innovate for an energy-constrained world and abandon the old paradigm that assumed unlimited resources.
Recent regulations signal an irreversible shift towards a new reality that heavily favors energy efficiency in AI. Several countries are introducing regulatory measures aimed at stemming data centers’ strain on local energy grids. For example, in China, new data centers in national hub nodes must use at least 80 percent green electricity. In Thailand, the aim is to implement regulatory frameworks, including an enhanced carbon tax framework and mandatory corporate emissions reporting. Singapore also introduced new energy efficiency standards for data centers, aiming to save at least 30 percent in energy consumption. For AI developers, this means considering alternative approaches that prioritize energy constraints and IT efficiency.
The rise of small language models as high-efficiency alternatives
Fortunately, AI innovation is proving that models don’t need to endlessly scale to be effective. While ballooning LLMs have been the rockstars dazzling audiences in Asia, AI teams are turning to Small Language Models (SLMs) as small yet mighty alternatives that can be better suited for specific enterprise use cases. SLMs typically include models with fewer than 10 billion parameters, tiny in comparison to LLMs’ staggering hundreds of billions, or even trillions. Lower counts mean SLMs can serve as viable alternatives when computational resources are limited or when fast processing and low latency are critical. In fact, OpenAI’s GPT-5 can auto-switch to use smaller, internal models to answer simple questions, relying on larger models for more complex tasks. This way, various models can provide accurate, yet efficient, answers to more queries.
Assuming an inability to deliver enterprise-grade performance, organizations dismissed SLMs as being hard to scale and reduced general knowledge due to narrower training data. Many worried that these models would disappoint when it came to complex reasoning, multilingual capabilities, and effectively handling nuance and ambiguity. However, many of the misperceptions have been disproven over the last eighteen months. Underpinning this pivot toward SLMs is a growing recognition that data quality, not quantity, is the key to model performance. LLMs are often burdened by vast pools of raw, unfiltered data, much of which is duplicated or entirely irrelevant. SLMs instead rely on so-called ‘data efficiency’ principles, where datasets are meticulously curated with precision and relevance in mind.
The targeted approach to purpose-built SLMs trained on curated, high-quality datasets has improved accuracy for domain-specific tasks while reducing the inefficiencies burdening larger model behemoths. Smaller models stand out for their ability to be fine-tuned more quickly and updated more often, making them flexible and ideal for dynamic settings. SLMs are also appealing to enterprises because they are typically easier to deploy and manage, and are preferred in edge environments, such as in smart factory settings, where energy efficiency is essential.
Numerous examples of successful SLMs are emerging across industries. Retailers are leveraging leaner AI models to run customer support chatbots, while smartphones and wearable devices are being deployed with SLMs to generate real-time translation. Clinicians are also implementing SLMs to support resource and time-constrained physicians in medical diagnostics by rapidly analyzing patient symptoms, lab results, and medical systems to produce possible diagnoses and propose next steps.
Open-weight architectures reinforce the shift to efficient AI
Along with the rise of SLMs, open-weight models are also revealing themselves to be a viable energy-efficient alternative to LLMs. Those models using Mixture of Experts (MoE) architectures are attractive to energy-conscious AI developers because they only activate a small subset of their parameters during inference, drastically reducing the compute and energy required for each task. As open-weight models can be deployed locally and fine-tuned for specific use cases, they also relieve the energy burden on cloud infrastructure, making them well-suited for edge environments and enterprise applications where energy efficiency is important. The learnings of the last two years have demonstrated that SLMs can be just as effective, cost less, and excel at tasks that don’t demand the extensive knowledge base of an LLM.
To survive the next phase of development, organizations must mandate that AI teams start with energy efficiency as a foundational operating principle and consider energy implications at every stage of the AI lifecycle, from data curation to deployment. By applying holistic thinking to IT challenges, developers will be able to refine their techniques and spark renewed creativity in the next wave of AI’s evolution.
Srikanth Seshadri is the Chief Solution Architect, Enterprise Architect at HPE. He has over 25 years of IT experience providing services and solutions to customers in telecommunication, manufacturing, public sector, natural resources, banking and finance industry in various capacities. His career experience ranges from systems engineering to IT project management and service delivery management to technical solutions and architecture.
TNGlobal INSIDER publishes contributions relevant to entrepreneurship and innovation. You may submit your own original or published contributions subject to editorial discretion.
Featured image: fabio on Unsplash