Google launches Market Access Program to scale Indian AI startups globally with new AI models

Google Bridges the Gap Between Pilots and Production

Google has launched the Market Access Program at the AI Startups Conclave held in India's capital, addressing a critical bottleneck that Indian AI startups face when transitioning from successful prototypes to repeatable enterprise contracts1

1

. The programme targets AI-first companies that have moved beyond the prototype stage and are preparing to scale Indian AI startups globally. "Indian startups are building serious deep technology and solving population-scale problems with AI," said Preeti Lobana, VP and country manager for India at Google1

1

. While the journey from labs to prototypes has improved, many startups still struggle with scaling—a gap this initiative seeks to close.

Enterprise Readiness Through Structured Training and Network Access

The Market Access Program focuses on enterprise readiness by providing structured training on global enterprise sales, pricing strategies, and buyer behaviour1

1

. Indian AI startups will gain facilitated introductions to Google's global network of CIOs and CXOs, offering the access and credibility that often determine long-term success at the critical scale-up stage2

2

. The initiative also includes international immersion programmes in partnership with ecosystem organizations like TiE Silicon Valley and Alteus, with applications now open1

1

. This full-stack approach spans infrastructure, models, safety, skilling, and capital, positioning "Bharat-tested" AI as a global benchmark for resilience2

2

.

MedGemma 1.5 and FunctionGemma Expand Open-Source AI Models

Google introduced two open-source AI models to support the AI ecosystem: MedGemma 1.5 and FunctionGemma1

1

. MedGemma 1.5 is a 4-billion-parameter model designed for healthcare AI applications, supporting high-dimensional medical imaging workflows including CT and MRI scans, whole-slide histopathology, longitudinal chest X-ray analysis, anatomical localization, and extraction of information from medical lab reports1

1

. The model builds on Google's Health AI Developer Foundations programme and its collaboration with All India Institute of Medical Sciences, which is using MedGemma to develop health foundation models as part of India's Digital Public Infrastructure1

1

2

2

. FunctionGemma, a lightweight variant of the Gemma 3 270M model, is optimized for function calling and on-device AI agents1

1

. It enables applications to convert natural language commands into executable actions locally, allowing AI systems to operate with low latency, limited connectivity, and enhanced user privacy1

1

.

Infrastructure Investment and India's $126 Billion AI Market Projection

Google emphasized its ongoing investments in cloud and AI infrastructure, including the Global AI Hub in Visakhapatnam, which provides a one-gigawatt compute foundation powered by green energy and Google's AI chips1

1

2

2

. The facility is positioned as a long-term anchor for India's AI economy, enabling startups to build on the same cloud and AI stack that powers Google globally2

2

. The company also highlighted progress on data availability through Project Vaani, a collaboration with the Indian Institute of Science that has released more than 27,000 hours of speech data across over 100 Indic languages through the government's Bhashini platform1

1

. According to the 'Bharat AI Startups Report 2026' released by Inc42, India's AI market is projected to reach $126 billion by 2030, with nearly 47% of enterprises already moving AI use cases from pilot to production2

2

. Falling compute costs and public digital infrastructure are allowing founders to shift spending from infrastructure to product innovation, while startups like Cloudphysician have reduced ICU mortality rates by 40%, demonstrating outcome-led AI adoption2

2

. FunctionGemma can be fine-tuned using tools such as Hugging Face Transformers, Keras, and NVIDIA NeMo, and deployed across environments including LiteRT-LM, vLLM, Llama.cpp, and Vertex AI1

1

.