Google expands India AI push with focus on agentic systems, local infra

Google has unveiled a broad set of AI initiatives for India that go beyond launching new AI models, signalling a stronger focus on helping developers, businesses and public institutions deploy agentic AI.

The announcements, made at Google I/O Connect India 2026 in Bengaluru, span education, healthcare, enterprise infrastructure, cybersecurity and developer tools, reflecting the company’s effort to address some of the practical barriers to AI adoption at scale.

The developments also highlight Google’s view that the next phase of AI will be defined not only by more capable models, but by the infrastructure, security and local partnerships needed to use them responsibly.

Alongside new products and research collaborations, the company pointed out the economic contribution of its existing platforms, saying third-party research estimated that the Google Play and Android ecosystem generated Rs 5.3 lakh crore ($60 billion), in revenue for Indian app publishers and the wider digital economy in 2025, representing 28% growth over the previous year.

Opening the event, Google India Country Manager Preeti Lobana said India had emerged as one of the world’s most significant AI markets because of both its scale and its ability to apply technology to real-world problems.

“We are watching India shape the global tech landscape in real time, driving a new era of AI leadership defined both by purpose and true India scale,” she noted.

“The country not only outlined AI’s immense potential, but also demonstrated that the global advances of this powerful technology are most meaningful when they are anchored in tangible impact beyond the theory and beyond what's in labs to actual impact on the ground,” Lobana added.

She said Google’s strategy was to support India with a “full-stack approach” that combines computing infrastructure, AI models, developer tools and partnerships with government, academia and industry.

That broader strategy was reflected across the announcements, many of which centred on what Google calls the “agentic era”.

Lobana said the transition towards these systems would require greater emphasis on security and governance.

“India’s builders are already deploying AI faster than almost anywhere else. As we drive the shift into the agentic era, where AI moves from answering queries to securely executing tasks, our focus is on providing the underlying infrastructure and guardrails the ecosystem needs to scale safely,” she added.

Laying the groundwork

A significant part of the announcements focused on enterprise AI infrastructure. Richard Seroter, Chief Evangelist at Google Cloud said Gemini will become available through Google Distributed Cloud, allowing organisations, including government agencies and businesses operating in regulated sectors, to run frontier AI models entirely within Indian data centres. 

According to Seroter, supporting services can also operate without connecting to the public internet, while Gemini 3.5 Flash will be offered with in-country machine learning processing commitments to help organisations meet localisation requirements.

Google also announced that Sec-Gemini V3, a specialised cybersecurity agent, will be made available to selected government and enterprise testers, including Flipkart. The system is designed to analyse complex security information and assist security teams in investigating cyber incidents more quickly.

The company also introduced CAPSEM, short for Capabilities Security for Agents, an open-source runtime environment intended to improve the security of autonomous AI systems. According to Google, CAPSEM isolates each AI agent inside its own virtual machine, reducing the risk that a compromised agent could affect an entire system.

Alongside these technologies, Google announced work on open standards intended to support secure AI interactions. These include Device Bound Session Credentials, which are designed to prevent stolen browser session cookies from being reused, and Agents-to-Payments, which works with the Agent2Agent protocol to enable secure, low-value financial transactions carried out by AI agents.

Google is also expanding research partnerships with IIT Delhi and IIT Madras to study agentic AI safety, including methods for identifying emerging threats and developing what it calls Guardian Agents to help reduce institutional risks.

From classrooms to clinics

Beyond enterprise computing, education formed another major pillar of the announcements. Google DeepMind said it is bringing its AI Research Foundations curriculum to India. The free 56-hour programme teaches learners how to build and fine-tune large language models (LLMs). Participants can earn Google Cloud skill badges and certificates after completing the programme.

The curriculum will be delivered through partnerships with NASSCOM and the Indian Institute of Science, while AVPN will support wider access through the Google.org APAC AI Opportunity Fund.

Google DeepMind also introduced ATL Saathi, a desktop web application developed with the Atal Innovation Mission. The Gemini-powered tool is designed to assist teachers in Atal Tinkering Labs by helping them prepare lessons and recommend practical classroom experiments. Google said the programme will initially reach 100 schools before expanding towards its long-term goal of serving 10,000 schools.

Healthcare was another area where Google highlighted locally relevant AI applications. Researchers at AIIMS Delhi are using Google’s MedGemma multimodal open models to develop India-specific AI models focused on leprosy and sexual and reproductive health.

“We have deeply appreciated our partnership and support from Google. MedGemma's multimodal capabilities have already delivered encouraging results in our dermatology screening. We are confident that expanding its use to leprosy and sexual and reproductive health will further bridge healthcare gaps, bringing vital clinical decision support to even more communities across India,” said Dr Somesh Gupta of AIIMS.

Google also announced that Gemini Live now supports more than 25 Indian languages and dialects, including Sanskrit, Bhojpuri and Maithili, as part of its effort to improve accessibility across India's linguistic diversity.

Partner cases

The event also highlighted how Indian companies are already applying Google’s AI models across different industries.

Policybazaar has rebuilt parts of its insurance platform using Google’s AI technologies, reporting a 50% increase in policy conversions and a 2.6-fold rise in advisor calls. Travel platform redBus demonstrated a conversational booking assistant that allows users to search for bus journeys in Hindi and other Indian languages through natural language conversations.

Other companies including Emergent, Adya, VideoSDK, Sivi, Superjoin and Knit showcased applications spanning software development, voice AI, creative design, financial modelling and enterprise integrations.

Google also said it will expand Google Play Academy training to 10,000 developers through partnerships with the governments of Rajasthan, Odisha and Madhya Pradesh. The programme is intended to help developers use the latest AI technologies while learning how to build and scale applications and games.