Google Deploys Trillium TPUs in India to Boost Local AI Infrastructure and Digital Sovereignty

Google Brings Trillium TPUs to India to Power Local AI Growth | AIM

Google DeepMind announced a collaboration with IIT Madras to support the launch of Indic Arena, a benchmarking platform developed by the AI4Bharat centre. Google on Tuesday announced a major expansion of its AI infrastructure and investments in India, unveiling new local compute capacity, AI tools, and collaborations designed to strengthen India's digital and AI sovereignty. The company said it is deploying its AI Hypercomputer architecture, powered by Trillium TPUs, within India to support businesses and public sector organisations in training and serving advanced Gemini models locally. The move aims to support India's data residency and sovereignty requirements while reducing latency for AI workloads. "India's developer community, vibrant startup ecosystem, and leading enterprises are embracing AI with incredible speed," said Saurabh Tiwary, vice president and general manager of Cloud AI at Google. "To meet this moment for India, we are investing in powerful, locally available tools that can help foster a diverse ecosystem and ensure compliance with AI sovereignty needs." The company said that Gemini 2.5 Flash, already available to regulated Indian customers, now supports local machine learning processing. Google Cloud has also opened early testing for its latest Gemini models in India and committed to launching its most advanced versions with full data residency support, marking the first time Google Cloud will host such models locally. The announcement also includes a suite of AI capabilities built for India's context. These include batch support for Gemini 2.5 Flash to handle large-scale AI tasks cost-effectively, Document AI for automating document processing, and real-time grounding on Google Maps for location-aware responses. In a move to strengthen India's AI research ecosystem, Google Cloud and Google DeepMind announced a collaboration with IIT Madras to support the launch of Indic Arena, a benchmarking platform developed by the AI4Bharat centre. Indic Arena will allow users across India to anonymously evaluate and rank AI models on multilingual tasks unique to India's linguistic diversity. "At AI4Bharat, our mission is to build AI for India's specific needs," said Mitesh Khapra, Associate Professor at IIT Madras. "A critical part of this is having a neutral, standardised benchmark to understand how models are performing across our many languages. Indic Arena will be that platform." Tiwary said the new TPU infrastructure and research partnerships reflect Google's commitment to India's long-term AI ambitions. "We're committed to bringing our latest AI advancements to India faster than ever, with the controls and capabilities that reflect the country's unique business and cultural context," he said. Google invited startups, research institutions, and government organisations to leverage the new Trillium TPU-powered infrastructure through Vertex AI, aiming to help build AI systems "by Indians, for Indians."

Google expands AI infrastructure and ecosystem efforts in India

* Google has significantly invested in local AI hardware * The company is also expanding access to pre-built AI tools India's developer community, startups, and enterprises are moving fast with AI and Google is clearly trying to match that pace. The company has announced a major expansion of its local AI hardware capacity in India, built on its AI Hypercomputer architecture and powered by the latest Trillium TPUs. The goal is to give Indian businesses and public sector organizations the ability to train and deploy advanced Gemini models within the country helping address India's growing demand for high-performance, low-latency AI while meeting data residency and sovereignty needs. Beyond infrastructure, the focus is also on control and context ensuring Indian organizations can build and fine-tune models that actually understand local languages, business logic, and cultural nuances. Google also emphasized that true digital sovereignty isn't just about servers or models -- it's about people. The company says it's doubling down on efforts to support researchers, developers, and startups who are building for India's diverse needs. One of the key initiatives in this direction is a new collaboration between Google Cloud, Google DeepMind, and IIT Madras to support Indic Arena, an open benchmarking platform developed by the AI4Bharat center. Indic Arena will allow anyone to anonymously evaluate and compare AI models on multilingual Indian datasets -- creating a transparent standard for how AI performs across the country's many languages. Google is supporting the project with cloud credits to power its early phase. "At AI4Bharat, our mission is to build AI for India's specific needs," said Mitesh Khapra, Associate Professor at IIT Madras. "A critical part of this is having a neutral benchmark to measure how models perform across our languages. Indic Arena will be that platform." From startups to large enterprises, and from academia to government institutions -- Google is opening up new compute capacity and sovereign-ready infrastructure in Vertex AI to help India build its next generation of AI models. The direction is clear: AI that's trained, deployed, and built in India -- for India.