India must emerge as AI innovator, not just consumer: Google’s Harsh Dhand

Could you talk a little about the key partnerships Google is building in India and what these partnerships are focused on?

We have the largest presence of an innovation lab — Google DeepMind and parts of Google Research based out of India. The country is an important base for us, where AI innovation is happening at breakneck pace across APAC.

I lead a team of senior partnership development managers who work across different partners ranging from government to academia, startups, and enterprises. The idea is lab-to-impact, which means bringing all the innovations from our AI innovation labs, whether global or in India, to the forefront of this geography.

One big part of our job is to make sure that the best AI models land here and are utilised at the applied frontiers with the challenges in, for instance, healthcare, agriculture, education, and SMBs.

The second part is ensuring that India’s voice and experiences are reflected in how AI models are developed and deployed, so we can identify the challenges they face in the Indian context and use those insights to improve the models.

Google provides foundational capabilities through both its frontier models like Gemini and open models like Gemma. The foundational capabilities also expand to datasets, evals, and benchmarking. Google has invested heavily in setting up AI infrastructure with a large data centre, which forms the foundational infrastructure layer.

We also work with third parties to ensure impact-focused solutions are developed. Often, once we have these base open models, the ecosystem can take it and run with it. But some cases may need some hand-holding, funding, and additional engagement with Google, which is where the partnerships team comes in. This is the second layer.

The third layer is that research cannot be done in isolation but happens at an open ecosystem level. We are also invested in making sure that India emerges not just as a consumer, but also as an innovator, and that every AI research-focused startup has the right tools, capability, funding, and support.

The fourth layer is responsible AI — safety, bias removal, and making sure there is a trust layer, which is so important when we not only develop these models, but also deploy them. We ensure that model deployment, both in research and production, is done safely and securely. Those are the four tenets by which we support Google’s innovation in AI in India.

Why is it important to democratise AI, and what kind of real-world impact do you hope to create through these collaborations?

When we say democratise, we’re not just talking about the 20-30 per cent of the population living in metros, who are graduates and well-read. We’re also speaking of empowering smallholder livelihoods.

Google is well known for AnthroKrishi, where we have these foundational models. Imagine Google Maps, but to provide farmers with the farm boundaries. The whole ag-tech ecosystem takes this data and multilingual farm advisory to deliver to the last mile of farmers.

Or imagine the ASHA or Anganwadi workers, who, through HealthVani — our partnership with Wadhwani, built on a combination of Gemini as a conversational layer, MedGemma and Gemma as the core reasoning model — have a lightweight, low-cost hardware model that serves the interaction. They don’t have to carry logbooks everywhere.

Another thing about democratisation, which is core to our lab and DNA, is voice as a universal interface. Our partners, IISc, built a model called SraVaani, built on the Vaani dataset, which caters to 65-plus long-tail languages. These are the three elements of democratisation.

How does Google’s AI innovation and research presence in India differ from its setups in other parts of the world?

Google has a long-standing presence in India. When we first thought about expanding to APAC, we thought about setting up a lab in India. There are two aspects to how our innovation labs work.

We work closely with the global labs, whether in London or Mountain View. We contribute to the frontiers, like Gemini development, model optimisation, or how to reduce inferencing costs. All that work is happening in the lab here. We are also working on advanced frontiers of agentic AI. Because of the lab’s presence here, we also work on fundamental problems in India, for instance, the agriculture-related AnthroKrishi project. This year, we’ve launched the same model developed in the India lab across several countries in Southeast Asia.

For example, we built models specific to chest X-ray and tuberculosis identification, and gave that technology to Apollo and said, ‘Be a partner. Take it on and scale it.’ While Google is not a healthcare company, we have developed health AI solutions which can be scaled with partners. These are two ways by which the India Lab is both unique and also tied into the DNA of our global AI innovation labs, which we work directly with.

What support does Google provide startups beyond foundational capabilities and datasets, particularly in terms of infrastructure access and funding?

We work closely with our Google.org grant-making entity. For example, we contributed $2 million each to all AI centres of excellence in India. We invested about $10 million in the AI CoEs and one dedicated language centre of research in IIT Bombay. The Google.org team can support these.

The Cloud and Google for Startups team have several tiers of support ranging from a couple thousand all the way to $200,000 for startups who are adopting and are on the forefront of AI adoption, but need that level of cloud credits.

We also invest through country-specific programmes. For example, we know that adoption of MedGemma and Gemma as sovereign models for India is important, and we have invested this year close to $400,000 for MedGemma adoption.

You mentioned agriculture and healthcare. Are there other sectors where Google is looking to expand its AI use cases?

An important one is the education sector. One example is the Atal Tinkering Lab, but we are also working on AI in education and AI for education. In both cases, we make sure it’s responsibly injected into the education sector so that students, teachers, and everyone benefits.

India has led the way for DPI, but when you talk about AI as a DPI, it can touch all these sectors — health, education, agriculture, and finance or the MSME sector. Maybe a broad horizontal layer, but then vertical-specific is also coming up quite fast.

We are looking at making sure we continue to be deeply embedded in the ecosystem of languages, open models, and agriculture and healthcare, and are doubling down on a lot of these efforts. Languages continue to be a frontier that we improve on in all aspects, including accessibility. There are lots of things to be solved for India, including accessibility, fairness, bias removal, and the cost factor.