Google’s strategy includes integrating its advanced Gemini AI model into its Android app and focusing on vernacular language support to engage a largely untapped market, the head of its DeepMind research division in India told Mint, even as it lags behind its competitors in appealing to developers in India and around the world.
Manish Gupta, Google’s AI boss in India, explained that the company wants regular users to experience its AI directly, not just businesses and programmers. For that, the company has put its latest AI technology, Gemini 2.5, into the Gemini app that works on all Android phones.
“Through DeepMind, we’ve integrated video and image generation into the single Gemini app, which is available to all users now. DeepMind India has also worked on making video generation, live object recognition and all other capabilities work with nine Indian languages. The app is natively multimodal, and our experimental attempt at a universal digital assistant, called Project Astra, is being increasingly made available to more users,” Gupta said.
Also read | Mint Explainer: Can Google get its AI mojo back with Gemini?
He, however, did not comment on whether Google will look to make Gemini a default offering on Android phones going forward. “It is already available as a free app for all—and Samsung has already integrated Gemini as a built-in offering on their latest phones,” he said, without commenting upon whether more such smartphone integrations are on the way.
India sells roughly 150 million smartphones every year, and has around 1.2 billion active smartphones in the country. Of this, nearly 95% run on Google’s Android operating system.
On Friday, Google highlighted through a report in partnership with consultancy firm Kantar India that two of three smartphone users in India are not clear on AI’s use cases, while one in every three such users has never used a generative AI tool. The company’s latest push in India is aimed squarely at the consumer, with vernacular language integration in Gemini a key focus for the search and mail giant from Mountain View, California, US.
But, Google’s consumer push comes after the company fell behind the likes of OpenAI, Meta, Anthropic and even China’s DeepSeek in the race for AI supremacy among large technology firms. On Wednesday, data from Google’s own internal analysis at its landmark US Department of Justice hearing on the monopoly of Search said that Gemini, as of last month, had 35 million daily active users—paling in comparison with OpenAI’s ChatGPT with 160 million daily active users.
Read this | ‘Google tax’ abolition: No major revenue gains for Big Tech, but a moral victory
“Even among developers, Anthropic’s Claude and OpenAI’s GPT family of models are the pick of the lot among developers, while open-source developers have moved to China’s DeepSeek—thus hurting Meta’s early run with Llama. Google, however, is behind this lot in terms of both developer and consumer integration, and pushing for consumer usage might not necessarily promise success,” said Kashyap Kompella, AI analyst and consultant, and founder of advisory firm RPA2AI.
Ironically, Google is the founder of the transformer model—the very basis of generative AI and the reason for OpenAI’s existence, having funded the now-iconic ‘Attention is all you need’ research paper published by researchers from Stanford University in June 2017.
Google’s Gupta acknowledged this. “As powerful as our models are, we do see a gap in their developer uptake. We’re looking to increase how we reach out to developers, as a result,” he said.
Pointing to how Google failed to capitalize on its early advantage, Kompella said that despite Google’s massive success with search, email and browser (through Chrome), “the company’s product graveyard shows that not every field and product Google pursues sees success.”
Read this | With AI in Indian iPhones, can Apple rival Google’s appeal for coders?
“With AI, Google’s right strategy would be to court developers and bring them in, to build third-party applications and create an AI ecosystem that can go well with its Android user base,” Kompella added.
Such a turnaround is possible. “You can’t write Google off so early, given their might, size, and access to funds, markets and talent,” Kompella emphasized.
Gupta said that India’s engineering team and developers alike can play a big role going forward.
“DeepMind’s India developers developed an AI technique called Matrioska representation learning, or Matrioska transformers—named after the Russian dolls. We built the latest Gemini models in a similar style, where models of varying sizes are nested one inside the other. Now, we’re integrating them into Android phones,” he said.
“We’re looking to offer multiple points of entry into AI for users, developers and businesses into the AI ecosystem. Through Google Cloud’s Vertex AI APIs, the Gemini ‘Flash’ models are seeing the highest uptake in India—as per benchmarks, Gemini’s Flash models are the most energy efficient. This is encouraging for developers.”
And read | A Google GenAI expert weighs in on why companies are clamouring for AI agents
“We’re also seeing small companies experiment with AI in advertising already, and we expect this vertical to eventually generate substantial revenue. Ultimately, it all boils down to a product adding value, and that can be a clear business driver for our AI efforts,” he added.
For now, though, Google’s focus is on bringing AI into everyone’s screens.
“Public services will play a key role in bringing AI to the masses. The Gemini app, which bundles all of Google’s AI capabilities, is available for all. We’re not commenting on making Gemini a default feature on Android in general as yet, but the app is available for everyone already,” said Shekhar Khosla, vice-president of marketing at Google India.
