October 17, 2026, (Inside AI) — The global race for artificial intelligence is dominated by a narrative of scale: bigger models, larger data centers, and billions in capital. But a growing counter-movement argues that this Silicon Valley playbook is not just unsustainable for most of the world, but fundamentally misguided. At the forefront of this push is the Frugal AI Hub at Cambridge University's Judge Business School, co-founded by Serish Gandikota. The Hub's mission is to shift the focus from building ever-larger models to deploying AI effectively under real-world constraints of cost, computing power, and organizational capacity.
Gandikota's journey to frugal AI began not in a computer lab, but on an ice sheet in Svalbard during the 2008 International Polar Year. Tasked with drilling an ice core to analyze past climates, his team faced a critical failure. The drill, sent the previous summer, was bent. With no way to repair it on the ice sheet and time running out, they had to abandon the expedition. "That experience taught me a lesson—that one can have a lot of resources and yet fail," Gandikota said. This realization, that resource abundance does not guarantee success, became a cornerstone of his philosophy.
This insight is gaining traction as organizations worldwide confront the gap between AI's promise and its practical implementation. The Frugal AI Hub, launched in April 2025, is positioning itself as a bridge over that gap. Its work with the United Nations International Computing Centre (UNICC) led to a report titled Total Cost of Ownership to Social Impact, which provides a framework for evaluating AI projects not just on ROI, but on their contribution to the Sustainable Development Goals (SDGs). The Hub is also expanding its footprint through deployment labs, including a partnership with the Andhra Pradesh government in India under the Amaravati Quantum Valley initiative, signed in November 2025.
The core of Gandikota's argument is that the current AI discourse is dangerously unbalanced. "The fundamental issue is that the US and Silicon Valley are setting the agenda, and everyone else is simply reacting," he said. "It does not matter whether you are in the UK, Europe or India. The conversation coming out of Silicon Valley is taking up all the oxygen in the room." This reactive stance, he warns, leads to a fear of missing out that drives organizations to adopt AI without a clear understanding of the outcomes they need.
Read: Pension Funds AI Governance
Why The Binding Constraint Is Absorption, Not Intelligence
Gandikota's forthcoming book, Frugal AI: Revolutionary Practices For Making AI Work, Scale, and Endure, argues that the primary bottleneck in AI adoption is no longer the intelligence of the models, but the ability of organizations to absorb and govern them. "The binding constraint in AI is no longer intelligence but absorption," he said. This perspective challenges the prevailing narrative that equates progress with model capability. Instead, it suggests that the real work lies in the unglamorous realms of governance, workflow redesign, and change management.
This view is echoed by a growing number of practitioners who see a widening gap between AI's supply side and demand side. While companies release increasingly powerful models, the majority of organizations lack the infrastructure and expertise to deploy them effectively. The Frugal AI Hub's approach is to start with the desired outcome, not the technology. "Instead of starting with technology as the input and asking what output you want to achieve, you should start by asking what outcomes you actually want to achieve," Gandikota explained. Only after establishing governance, workflows, and change management should organizations consider which model to use, whether frontier, open-source, or a small language model.
This philosophy is not about settling for less, but about rethinking the solution entirely. Gandikota points to the Embrace baby warmer, a frugal innovation that replaces a $20,000 incubator with a $200 sleeping-bag-like device that requires no electricity. "Frugal innovation is not about taking something that already exists and making a cheaper version," he said. "It is about rethinking the solution itself." Applying this to AI, he argues that countries like India should not simply build a cheaper version of Silicon Valley's AI, but develop solutions tailored to their unique needs and constraints.
The Hub's work with the UNICC and the Andhra Pradesh government exemplifies this approach. In India, the focus is on building local capacity and co-creating pilots that can scale. One example Gandikota cites is MahaVISTAAR, a voice AI advisory service for farmers. The system must handle noisy environments, diverse dialects, and provide accurate, up-to-date information on weather and fertilizers. The solution involved building an orchestration system where institutions and designated individuals are responsible for maintaining the data. "It is a great example of frugal AI in practice," Gandikota said.
This emphasis on orchestration and data maintenance highlights a critical but often overlooked aspect of AI deployment. The success of MahaVISTAAR depends not on the sophistication of the voice model, but on the reliability of the information it delivers. This aligns with Gandikota's view that many challenges are not technological. "There is a lot that organizations can do to prepare themselves for AI deployment," he said. "Finally, deployment is the key."
Read: AI Boom Risks
As the AI race accelerates, the frugal AI movement offers a pragmatic alternative for organizations and governments that cannot compete on scale. It is a philosophy that prioritizes sustainability, governance, and local context over raw capability. For Gandikota, the lesson from Svalbard is clear: having the most advanced drill does not matter if you cannot adapt to the conditions on the ground.
