Serish Gandikota is the co-founder of the Frugal AI Hub at Cambridge University-Judge Business School.
An honorary fellow at the school, he also leads the Hub’s Frugal AI Deployment Labs, which work with partners globally to move AI from pilots to sustained real-world use under constraints of cost, computing power and organisational capacity.
His work over the past decade has spanned frugal innovation, technology deployment and impact investing across Europe, India and Africa.
Co-founder of InnoFrugal in Finland, Gandikota built a community around frugal innovation and has earlier worked in environmental science and glacier research, including fieldwork in Svalbard during the International Polar Year.
He is the author of the forthcoming book, Frugal AI: Revolutionary Practices For Making AI Work, Scale, and Endure.
Gandikota spoke to indianexpress.com on his book, his journey towards Frugal AI, its deployment labs and the challenges governments and companies face in their journey to make AI adoption and deployment sustainable.
Venkatesh Kannaiah: Tell us about your journey towards Frugal AI.
Serish Gandikota: I studied chemical engineering in India and then moved to the US to study environmental engineering before moving to Finland to pursue glacier and climate research. In 2008, I was part of an International Polar Year expedition to Svalbard, where we went onto the ice sheet to drill an ice core, collect it, and analyse past climates. But when we reached the ice sheet, we discovered that the drill, which had been sent the previous summer, was bent.
We had no way to repair it on the ice sheet. We were short of time, and the weather conditions were not favourable. Eventually, we had to abandon the entire exercise. That experience taught me a lesson–that one can have a lot of resources and yet fail.
The baby incubator is an example of frugal innovation that I admire. A conventional baby incubator can cost at least 20,000 dollars, making it expensive and unworkable in places with unreliable electricity.
A group of students from Stanford University came up with an alternative: the Embrace baby warmer, which looks like a sleeping bag. At the back of the sleeping bag is a heating pack containing a special gel that maintains a constant temperature for six hours at a time. The Embrace team developed an alternative that costs around 200 dollars, is portable, and does not require electricity.
The key takeaway is that frugal innovation is not about taking something that already exists and making a cheaper version. It is about rethinking the solution itself. That is also how I think about frugal AI.
Silicon Valley has developed AI using enormous resources that countries like India may not be able to replicate. But that does not mean we should simply build a cheaper version of what Silicon Valley has created. We should develop AI solutions that suit our needs and deliver similar kinds of outcomes through a different approach.
Venkatesh Kannaiah: What does the Frugal AI lab do, and what are its goals?
Serish Gandikota: In 2023, I moved to Cambridge University after being invited to serve as an honorary visiting fellow at the Judge Business School. In April 2025, we officially launched the Frugal AI Hub at the business school.
The United Nations, particularly the UN International Computing Centre (UNICC), approached us for insights into the costs associated with AI projects. UNICC undertakes AI and digital projects, including setting up on-premises data centres when required.
This led to a report we published together with them, called Total Cost of Ownership to Social Impact. In the report, we examined the costs that go into an AI project and how a for-profit organisation should evaluate its return on investment and assign a value score to it. For projects aligned with the Sustainable Development Goals (SDGs), we explored how to calculate a net SDG score. The learnings from this report are being used extensively by companies and organisations.
Another idea was that, while the hub would be based in Cambridge, we wanted to establish deployment labs in other geographies with like-minded partners. These would be co-creation and incubation centres where we would help build local capacity, identify problem statements with our partners, and work together to design pilots that could eventually scale.
In November 2025, we signed an MoU with the Andhra Pradesh government under the Amaravati Quantum Valley initiative. We are now working on developing a digital course on frugal AI and collaborating with the government to set up a Frugal AI lab in Amaravati. We are also in discussions with the Telangana government.
The fundamental issue is that the US and Silicon Valley are setting the agenda, and everyone else is simply reacting. 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.
Instead of being reactive, we need to focus on specific questions. Do you really need AI? Everyone seems to think they need AI, but does the outcome you want to achieve really require it? Just because someone else is doing it, you do not need to have a fear of missing out. These are the kinds of conversations we are initiating.
Venkatesh Kannaiah: Can you explain your frugal AI philosophy?
Serish Gandikota: When people talk about frugal AI, do you know what they usually focus on? They talk about edge computing, running AI on phones or laptops, on-premises computing, federated learning, and distributed computing. All of that is relevant, but it is secondary to the larger question.
At the Frugal AI Hub, we approach it more as a design philosophy, a systems architecture, and a framework.
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.
To achieve those outcomes at scale, you will need technology, the right kind of model, and so on. But before even looking at the model, you need to ask whether you have everything else in place. Do you have governance, workflows, orchestration, and change management in place?
Only then should you look at which model is appropriate for your needs. That process should help you decide whether you need frontier technology, an open-source model, or your own small language model. It should also inform your decision on whether to use cloud servers, on-premises infrastructure, or distributed computing.
For us, Frugal AI is fundamentally a design philosophy.
Venkatesh Kannaiah: Tell us about your forthcoming book, Frugal AI: Revolutionary Practices For Making AI Work, Scale, and Endure.
Serish Gandikota: The book is for practitioners responsible for making AI work in organisations, private and public leaders making investment decisions, tech leaders designing deployment architectures and policymakers shaping public systems.
The book is designed to help institutions ask better questions before commitments form and before dependencies become difficult to reverse.
The durable advantage in AI will belong not only to those who build powerful systems, but to those who can sustain, govern, and absorb them over time. The binding constraint in AI is no longer intelligence but absorption.
We have spent extraordinary amounts of effort improving the intelligence of AI systems. But at the same time, organisations don’t have a ready-made toolkit or a magic wand that they can pick up and use to make things happen.
I see Frugal AI as a response to the deployment-related problems organisations are facing. AI is advancing exponentially, but the ability of institutions, companies, and even governments that want to deploy AI solutions for their citizens is lagging.
There is so much happening on the supply side, with everybody creating something new, but on the demand side, organisations are still struggling to understand how to use all of this.
Venkatesh Kannaiah: So, where do you see yourself in the spectrum of AI implementation?
Serish Gandikota: The frugal AI philosophy is intended for organisations looking to deploy AI at scale.
We are looking at the architecture of the entire system. It also helps decision-makers, executives, and managers of AI projects understand how to move from pilots and proof of concept to large-scale deployment.
We believe frugal AI provides a lens through which organisations can examine what happens when a solution is scaled up, right from the beginning.
Venkatesh Kannaiah: How are the UK and Europe readying themselves for this race?
Serish Gandikota: In terms of technology and resources, the UK and Europe are better placed than the Global South, but they cannot compete with the US and China.
Take the UK, for example. If it wants to build hyperscale data centres, it needs land, water for cooling, and additional electricity. A one-gigawatt data centre requires one gigawatt of electricity, along with a power grid capable of supporting that demand, particularly when multiple hyperscale data centres are involved.
Then there is the familiar ‘not in my backyard’ problem. People may oppose such infrastructure near where they live. Similar challenges exist across mainland Europe.
These countries are also recognising the relevance of frugal AI. They understand that, although they may have better access to technology and resources than the Global South, they still need to think frugally and work within their constraints.
Their constraints may not be the same as those in India, but they certainly have constraints of their own.
Venkatesh Kannaiah: From a frugal AI perspective, how do you see the new AI developments?
Serish Gandikota: Everyone is reacting to what is coming out of Silicon Valley. We have to stop doing that.
Frugal AI and Frontier AI occupy different spaces. I am not against hyperscale data centres or frontier technology. They have their own place. But confusing the philosophy behind Frontier AI with what is required for deployment is the issue.
Companies and executives are also making the mistake of assuming that personal use of AI at the task level will automatically translate into organisational AI usage and transformation. They are discovering that this is not true.
Venkatesh Kannaiah: Where does India stand in this scenario?
Serish Gandikota: India has several factors going for it.
First, there is a growing recognition that computing tech and computing resources matter. If you don’t have access to computing capacity, you become dependent on someone else. AI has become an issue of sovereignty, not just for governments but also for companies.
Second, countries like India have experience building digital public infrastructure. We have examples such as Aadhaar and UPI.
Third, language is a major issue. Fewer than 50 languages have properly developed AI models, while the world has anywhere between 5,000 and 7,000 languages. India itself has a large number of languages.
From that perspective, it is encouraging that India is taking the issue seriously. We are seeing foundational models and other AI initiatives emerging in the country.
Venkatesh Kannaiah: Can you give some examples of where frugal AI principles could be of use in an Indian context?
Serish Gandikota: One example is the voice AI advisory service for farmers, MahaVISTAAR. I have documented it in my book.
When people speak on the phone, there are pauses, gaps, and different ways of speaking. Farmers also don’t necessarily call for advice from quiet homes. They might be out in the fields, where there is a lot of background noise. The voice model needs to understand what the person is saying in that particular dialect and then provide an appropriate answer.
But, more importantly, the information the farmer receives must be accurate and up to date. The farmer’s livelihood may depend on it.
What they did was build an orchestration system in which institutions and designated individuals were made responsible for keeping information on weather, fertilisers, and other relevant subjects up to date. They had to ensure that the databases were maintained properly. All these elements had to work together.
It is a great example of frugal AI in practice.
Venkatesh Kannaiah: How frugal can we really get in our AI journey?
Serish Gandikota: We can take inspiration from how China has been doing things. DeepSeek is an example in itself.
On the technology side, there are things we can do much more cost-effectively and resource-efficiently. It could be about how we capture data, how we build a model from that data, or other aspects of the technology.
Then there is the computing side, although we have to consider supply chain issues and other constraints.
As we build our capabilities and capacity in these areas, we also need to remember that many of the challenges are not technological. There is a lot that organisations can do to prepare themselves for AI deployment. Finally, deployment is the key.