As artificial intelligence advances day by day at astonishing speed, an essential question persists: who is AI truly serving, and at what cost to society, environment and opportunity? That is the challenge at the heart of the first episode of the new series of the Cambridge Executive Business Insights: Rethinking AI podcast. In conversation with host Jaideep Prabhu, Serish Gandikota, Elizabeth Osta, and Arjuna Sathiaseelan draw on their collective expertise across frugal innovation, corporate transformation, and technical development to propose a compelling blueprint: AI that is not just bigger, but better, because it is viable, accessible, and designed for everyone.
Rethinking innovation: the power and promise of constraints
Since the earliest days of industrial progress, the prevailing narrative has been simple: more is better. AI, with its insatiable appetite for data and compute, has reached heights unimaginable even five years ago but, as Serish Gandikota notes, much of this progress comes from “innovation of abundance”: world-class data centres, enormous budgets, and highly specialised talent. Yet most of the world operates in a reality defined by constraints: infrastructure, cost, energy, and skill. For Serish Gandikota, whose background in frugal innovation ranges from resource-scarce communities in India to international conference halls, the urgent need is clear: “What needs to be done if you don’t have resources like ChatGPT?”
Rather than treating such limitations as setbacks, the Frugal AI movement views constraints as creative catalysts. As with the visionary urbanists and planners who sought inclusivity and resilience in city design, frugal innovation in AI calls for solutions that are intentionally shaped by the needs and resources of everyone, not just the privileged few.
Defining frugal AI: reimagining value through the lens of efficiency and impact
But what, in practice, does “frugal AI” mean? For Elizabeth Osta, with her extensive experience leading AI programmes in banking, retail and beyond, it begins with specificity: “One-size-fits-all, or generic tools don’t really work. They need to be very specific to a particular problem.”
Crucially, frugal AI is not about “making do”, but about maximising value by matching the solution tightly to the task. This means:
- Optimising for constraints: whether the limitation is energy, compute, bandwidth, or linguistic diversity, frugal AI actively designs for these realities.
- Promoting sustainability and accessibility: lowering the cost of AI does more than free up a balance sheet; it also reduces environmental impact and enables solutions to spread to remote oil rigs, rural clinics, or urban schools.
- Benchmarking for impact, not just accuracy: the value of an AI solution lies as much in its cost-to-value ratio and social benefit as in its technical prowess.In Arjuna Sathiaseelan’s words, frugal AI operates as “a design philosophy… to perform at a scale and accuracy of traditional systems, but with constraints in place: energy, water, data, cost.”
The three domains of failure and the urgent need for change
The team identify three key domains where today’s AI falls short, and where frugal alternatives are not just desirable, but essential:
- The hyperscaler challenge
Tech giants might have seemingly endless resources, but the environmental consequences are sobering. By one projection, “the energy consumption of all their data centres for deploying and running AI will be greater than India’s energy consumption as a whole” within the next eight years. Designing for abundance is no longer defensible, nor sustainable. - The enterprise gap
While corporates were early AI trailblazers, 95% of Generative AI pilots fail to move beyond test phases. The reasons are varied, from hidden costs, to lack of governance and unclear data, but often stem from the assumption that “bigger is always better”. - The consumer divide
For the majority of the world’s population, local content and language support are scarce. Out of 7,000 global languages, only about 100 are supported by today’s leading models. If AI is to truly be universal, it must be “small enough”, decentralised, and sensitive to linguistic and cultural nuances.
Frugal AI offers a path forward: energy savings, commercially viable production, and AI that speaks to every culture. The aim, says Arjuna, is “so the next three or four billion can actually have access to GenAI systems that actually speak their own languages, that represent their culture.”
Turning the tables: innovating for the many, not just the few
The drive for ever-larger AI models draws on the classic “bigger is better” approach, yet as Serish argues, much of human progress comes from innovating under pressure. Indeed, in places like India, where a Google data centre’s energy usage may rival that of a tier-2 city, replicating Western or Chinese patterns of AI development is not just impractical, but unjust. It risks reinforcing old divides and excluding billions from the data-driven future.
Instead, frugal AI takes inspiration from past successes in “doing more with less”: edge-based solutions, device-level intelligence, or “tiny machine learning” – all built around local realities, not global averages.
Who cares about frugal AI? Governments, businesses, and the global South
What might surprise outsiders is just how wide and urgent the demand is for this new approach. From governments aiming for digital sovereignty to companies struggling with cost and value, and from UK hospitals to emerging markets in India or Africa, the interest is “serendipitous and very high”.
Neo-industrial competition in the form of sovereign AI and national strategies shows that “it’s not just a technical issue, but deeply political, too”. Companies, meanwhile, are rapidly learning that general-purpose AI is expensive, opaque, and sometimes ill-suited to their most pressing needs. Elizabeth Osta sees a “race” among organisations to measure and prove the value of their AI investments, not just financially, but in terms of social impact.
Global adoption, local nuance: narrowing the digital divide
Perhaps nowhere is the case for frugal AI more urgent than in the developing world, where the digital divide persists even amid the rapid adoption of mobile and internet technologies. For Arjuna, who has spent years working with underprivileged communities in Africa and Asia, the rapid spread of AI risks widening old fault lines: “If you want to basically start adopting AI within specific communities… you need the right data infrastructure and grid. At this moment in time, that is lagging”.
What’s more, the linguistic diversity of places like Africa and India is dramatically underserved by current AI. Even as pilot projects collect local voice recordings and data to train models in tribal languages like Sholinger (spoken by just 40,000 people), the central challenge is not just translating algorithms but empowering communities to build and govern their own AI systems. True accessibility demands a shift from top-down provision to a participatory, decentralised ecosystem.
The Frugal AI Hub and its “adoption labs” seek to bridge this divide not as a philanthropic afterthought, but as a new paradigm. By partnering with governments and universities to build local capacity, train new cohorts, and develop solutions that are measurable and benchmarked for frugality – not just technical prowess – frugal AI becomes not just desirable, but inevitable.
Corporate reality check: from hype to impact
For many corporates, the narrative around AI has oscillated between hype and disappointment. Early wins in “numbers AI” such as credit risk scoring in banking or product recommendations for ecommerce, were clear, measurable, and transformative. The rise of large language models (LLMs), however, has proved both technically dazzling and practically confounding: impressive, but frequently all-purpose, generic, and difficult to connect directly to business value.
As Elizabeth reflects, “with the large language models… it’s been a bit of a struggle to find the right place to drive value with an all-purpose tool. And sometimes, these tools were generic, not trained on the local data companies usually need”. The result is an arms race without a clear roadmap: immense investments, limited upskilling, escalating costs, and growing ethical anxieties.
Frugal AI offers an antidote. By focusing on:
- Measuring and optimising for real impact (including social as well as financial returns)
- Upskilling teams to understand and own the technology
- Embracing architectures that are right-sized and context sensitive (not just “the biggest model available”)
companies can shift from pilot fatigue to sustainable adoption. The role of measurement, portfolio-level tracking (inspired in part by the team’s work with UN agencies), and the relentless focus on matching model to business case will define the winners of this new era.
Technology, process and people: the four pillars of successful AI adoption
Sathiaseelan distils the challenge for enterprises into four fundamental needs: data, workflows, governance, and people. Too often, failure to diagnose strengths and weaknesses in any of these leads to unclear use cases, pilot failures, or wasted investment.
The solution lies in what the Frugal AI Hub is seeking to provide:
- A measurement and monitoring framework to help organisations track total cost of ownership, return on investment, and alignment with wider impact goals (such as the UN Sustainable Development Goals)
- A technical stack and a forthcoming marketplace for tools that make it easier to build right-sized, responsible solutions
- Capacity building and adoption labs to demystify the technology and equip teams to innovate within their own resource constraints
This integrated approach recognises that AI is as much about people and process as about code and compute.
Human transition: managing the psychology of organisational change
Yet, as Elizabeth Osta argues, the transition isn’t just technical, it’s human. “Sometimes there is a reticence of the workforce to adopt AI, because there is a genuine concern about loss of jobs… sometimes it’s loss of meaning or purpose”. For companies to succeed, the journey has to account for the psychological realities of work: the meaning, the anxieties, and the need for purpose.
Research with MBA cohorts at Cambridge Judge is exploring how to map these factors and design successful change management journeys. The message is clear: lasting success depends on owning both the technical and human sides of the AI revolution.
Looking ahead: what does success look like?
What might the future hold if frugal AI principles are widely adopted? For Serish Gandikota, the vision is ambitious: a global network of 100 Frugal AI Adoption Labs, a thriving marketplace of both software and hardware solutions carrying the “frugal AI mark of approval”, and, critically, widespread adoption by organisations and communities who have developed skills and frameworks to make better decisions about their own futures.
But there is also a note of caution. If the world locks itself into “very expensive, energy-intensive AI infrastructures, they become very hard to unwind and difficult to justify economically and environmentally”. The challenge is not just one of technology, but of storytelling: ensuring that frugal solutions are not dismissed as “good enough” alternatives, but are recognised as high-quality, affordable, and accessible choices for everyone.
The five pillars of frugal AI: a systems framework
Over the course of the conversation, the team identifies five pillars that underpin the frugal AI philosophy:
- Resource efficiency: using less compute, less data, and less energy—by design.
- Environmental sustainability: lowering the AI footprint across the lifecycle.
- Accessibility and inclusion: lowering barriers to adoption, not only for privileged sectors.
- Impact and scalability: focusing on measurable value beyond the pilot stage.
- Context sensitivity: ensuring that AI adapts to local realities, not the other way round.
For leaders navigating the complex terrain of AI investment and adoption, the advice is clear: measure what matters, invest in people, build for real-world constraints, and always design for impact.
A new paradigm for intelligence: by everyone, for everyone
The future of AI does not lie in ever-larger models, but in collectively building intelligence that is smarter, leaner, and more responsible. Frugal AI does not mean “settling for less” – it means doing more, for more people, with less waste. It is a philosophy, a toolkit, and a movement for inclusion.
As cities face up to climate change by learning from their own streets and neighbours, so too must AI evolve by listening to the lived realities of its global users. By advancing the five pillars of frugal AI, the Cambridge Judge Business School and its partners are charting a roadmap for AI that works, not just for a privileged elite, but for everyone.
To explore these ideas further and hear from the thinkers at the forefront of this new movement, listen to the full episode, available wherever you get your podcasts.
