A growing movement away from Silicon Valley's resource-intensive AI approach is proving that smaller, purpose-built models can deliver real value while using a fraction of the energy and water. As major tech companies race to spend hundreds of billions on massive data centers, researchers and startups in India, Indonesia, and other regions are pioneering what experts call "frugal AI," creating leaner systems that run on inexpensive, offline hardware and prioritize local data sovereignty over raw computing power .

Why Is AI's Energy Footprint Becoming a Global Problem?

The scale of AI's resource demands is staggering. Model sizes have exploded in recent years, driving significant increases in energy and water consumption. Meanwhile, computing power for building advanced AI systems is heavily concentrated, with U.S. and Chinese companies operating more than 90% of AI data centers that businesses and institutions rely on globally . Africa and South America have almost no AI computing hubs, creating a widening digital divide. At Arizona State University, researchers found that data centers in the Phoenix area are warming surrounding neighborhoods by several degrees, highlighting the real-world environmental impact of AI infrastructure .

The current trajectory is unsustainable. As adoption accelerates, billions of people remain excluded from AI's benefits while the systems that do exist consume enormous amounts of electricity and water. This creates a paradox: the technology that could help solve environmental challenges is itself becoming an environmental burden.

How Are Frugal AI Systems Different From Traditional Large Language Models?

Frugal AI takes a fundamentally different approach. Instead of building massive models trained on billions of parameters, developers create smaller systems optimized for specific tasks and regions. These models use less compute, less memory, and less energy by design, which directly translates into a smaller carbon footprint . They can run on low-end devices like Raspberry Pi hardware, operate offline for extended periods, and function on low-bandwidth networks, making them practical for communities without reliable internet access.

A concrete example illustrates the approach's power. The Saving Voices Project, working with the Indian Institute of Information Technology Dharwad, built a speech AI system for the Indigenous Soliga tribe in southern India. With just five hours of voice data, researchers created a text-to-speech model that runs on sub-$50 hardware and operates completely offline. The voice data never left community devices, ensuring data sovereignty in a way that closed cloud systems simply cannot match .

"The current trajectory of AI development is unsustainable economically, environmentally, and socially. Model sizes have exploded, leading to significant energy and water consumption, and yet billions of people remain excluded from AI's benefits. Frugal AI addresses these failures," said Arjuna Sathiaseelan, founder of the Saving Voices Project nonprofit and chief technology officer of the Frugal AI Hub at Cambridge University.

Arjuna Sathiaseelan, Founder of the Saving Voices Project and CTO of the Frugal AI Hub at Cambridge University

What Are the Key Advantages of Building Smaller AI Models?

• Environmental Impact: By design, frugal AI systems use significantly less compute, memory, and energy, which directly reduces carbon footprint and water consumption compared to large language models trained on billions of parameters.
• Data Sovereignty: Smaller models can run entirely on local devices and offline infrastructure, ensuring that sensitive community data never leaves the region or gets extracted by external companies, addressing historical concerns about data exploitation.
• Cost Efficiency: Frugal AI eliminates proprietary API margins and can run on any infrastructure, making advanced AI accessible to startups, developing nations, and communities that cannot afford expensive cloud services or specialized hardware.
• Accessibility: These systems operate on low-bandwidth networks and inexpensive devices, bringing AI capabilities to regions without reliable internet or the capital to invest in cutting-edge infrastructure.
• Task-Specific Performance: Smaller models trained on specific data for specific uses can be nearly as effective as massive general-purpose models, while consuming a fraction of the resources.

The launch of DeepSeek in China last year energized the frugal AI movement globally. China is developing its own AI cloud and semiconductor supply chain, and its open-source models have quickly become the foundation for developers and builders worldwide . Countries including India, Mexico, and Malaysia are now actively working to reduce their reliance on expensive chip imports and proprietary systems from Silicon Valley.

How Can Organizations Use AI More Responsibly Today?

• Evaluate the Right Tool: Before generating content, ask whether AI is actually the best solution for your task, or if a simpler approach would suffice and use fewer resources.
• Write Clear Prompts: Well-defined prompts reduce the need for repeated requests and unnecessary computation, cutting both costs and environmental impact.
• Combine and Reuse Outputs: Group related questions into a single request and save valuable AI responses for future use rather than generating new content repeatedly.
• Focus on What You Need: Request summaries or targeted outputs rather than large, unfocused results that require more processing power.
• Share Best Practices: Encourage colleagues and teams to adopt responsible AI habits, multiplying the impact of thoughtful use across your organization.

At Arizona State University, administrators are taking these principles seriously. Kyle Bowen, the deputy chief information officer at ASU, explained that the university developed CreateAI to help users choose AI models suited to specific tasks rather than defaulting to large-scale models that consume more resources . The platform displays the size of models to users and places a green leaf next to smaller models to help people make environmentally conscious choices.

"We actually put a little green leaf next to smaller models to help people make choices around, 'Hey, these are smaller models that have a smaller impact in terms of the ways they consume resources,'" said Kyle Bowen, deputy chief information officer at ASU.

Kyle Bowen, Deputy Chief Information Officer at Arizona State University

What Challenges Still Limit Frugal AI's Growth?

Despite its promise, frugal AI faces real obstacles. Data scarcity, compute limitations, and funding gaps prevent scaling in many regions. Success also depends on domestic AI infrastructure, including access to efficient, cost-effective graphics processing units (GPUs) and data centers . Performance trade-offs are real, though experts argue that most tasks do not actually require frontier-level AI capability.

In the Soliga language preservation project, the word error rate in the speech models is slightly higher than commercial systems, but the trade-off was worth it. As Arjuna Sathiaseelan noted, "what we gained is complete data sovereignty, offline deployment on sub-$50 hardware, and a governance structure that elders and community leaders actually trust" .

The Saving Voices Project aims to reach nearly 500 million Indigenous people across 90 countries, preserving languages and cultures that are intricately tied to community identity. The Frugal AI Hub is setting up a lab in the Indian state of Andhra Pradesh and is in talks with officials in Kenya and Nigeria about expanding the approach .

Meanwhile, research into algorithmic efficiency is advancing the field. Microsoft Research scientist Lingjiao Chen developed FrugalGPT, an algorithmic framework that automates model selection, helping users pick the large language model (LLM) that suits their budget and accuracy goals while reducing costs significantly . As Chen explained, "Given the huge financial cost, energy consumption, and environmental impacts of LLMs, a major issue is how sustainable they can be in the long-term."

As Chen

"Given the huge financial cost, energy consumption, and environmental impacts of LLMs, a major issue is how sustainable they can be in the long-term. There is also a risk that AI models become unaffordable to more users due to their high cost. FrugalGPT and other frugal AI tech are thus increasingly important," said Lingjiao Chen, researcher in the AI Frontiers group at Microsoft Research.

Lingjiao Chen, Researcher in the AI Frontiers group at Microsoft Research

The frugal AI movement represents a fundamental shift in how the world thinks about artificial intelligence. Rather than assuming that bigger is always better, researchers and developers are proving that thoughtful, efficient systems can deliver real value while respecting both environmental limits and community autonomy. As AI adoption accelerates globally, the question is no longer whether we can afford to build smaller models, but whether we can afford not to.