This report consists of five sections: 1. The Landscape 2. Policy & Government Push 3. The Incumbents 4. The startups to watch for 5. The Challenges 1. The Landscape India’s AI ecosystem holds around 6200+ AI startups, with Bengaluru maintaining the dominance followed by Delhi, Mumbai, Hyderabad, Chennai, Pune, Kolkata, Ahmedabad, Chandigarh & Visakhapatnam.

In 2024, the focus was on "experimentation." Experimentation as in, applying AI to different areas like most of them are API driven products, like funding folks with a track record of having a good software development experience who is coming up with an idea of "AI for x or y or and so" and mostly they're categorized as "wrappers". In 2025, the money moved towards "Verticalization." Investors limited funding for "general AI tools" and started funding the AI for a specific vertical, like "AI for Healthcare," "AI for Legal," and "AI for Manufacturing" and so A defining trend of the year was the "83% Rule" that is approximately 83% of all GenAI startups in India are now focused on the application layer, as in solving specific business problems using the existing models (like OpenAI, Anthropic, Deepseek.. etc) and only 5% of the startups are working on building foundational models, that is building the models from scratch. The main goal & theme for Indian foundational models is that the models understand all the different Indian languages, dialects and cultures and building the foundational models goes by three pillars: - Linguistic Sovereignty (to understand different languages and nuanced cultural queries) - Data Security (the sensitive Indian data stays in India) - Cost Efficiency (should be optimized to run on cheaper hardware, even CPU's) The datasets used for these foundational models come from Indian newspapers, publications, books, and similar local sources. But in practice, only around 20–30% of the training data is actually Indian. The rest still comes from global datasets. 2. Policy and Government push The core boils down to three things 1. IndiaAI Mission (The Capital/Infra & Data) 2. Sovereign & Ethical Framework (The Rules) 3. State-Level Push (The Ecosystem) IndiaAI Mission The main aspect of IndiaAI Mission is like the government realizing that buying GPUs (the chips that power AI) is becoming too expensive for anyone who would like to work in the ecosystem. So, instead of letting only big tech players participate in the game, the government decided to spend ₹10,372 Crore (about $1.25 billion) to build a massive "public AI cloud" for everyone else. The core promise is simple: Democratized Compute and provide Data. Right now, if you are a small startup, renting an NVIDIA H100 chips will cost a fortune like sometimes it’s about $2–3 per hour per chip and multiplying that by thousands of hours, you’d go bankrupt before even launching and buying the chips is out of question. The IndiaAI Mission gets into play and fixes this by procuring  GPUs (10,000+ as of now) directly and offering them to startups, researchers, and public sector organizations at a heavily subsidized rate or even for free. It’s essentially a "compute subsidy" to ensure Indian innovators aren't priced out of the AI race. and along with giving the compute, the government also has setup a platform called IDP (IndiaAI Dataset Platform) that gathers high quality Indian datasets like local languages, agricultural data, and health records that foreign models do not have access to. The goal is that Indian startups build "sovereign" models that solve specific Indian problems, rather than relying on Western AI. Sovereign & Ethical Framework (The Rules) India has chosen a different approach to AI governance, different from what the EU or US or China has implemented. The EU has implemented a strict regulation through “The AI Act”, which basically says ban all risky AI with some heavy compliance requirements. The US has opted for market driven development with minimal intervention. China has implemented tight state control with built-in censorship mechanisms. India, however, is solving a different problem: a massive population, 22+ official languages, thousands of dialects, and vastly different socioeconomic realities. In response, India developed a 'Techno-Legal' framework built around the India AI Governance Guidelines (released in late 2025). Rather than creating restrictive AI specific legislation, the government established Seven Sutras (Principles): Trust as Foundation, People First, Innovation over Restraint, Fairness & Equity, Accountability, Understandable by Design, and Safety, Resilience & Sustainability. These function as guiding values rather than hard regulations, allowing startups to move quickly while maintaining accountability to core ethical standards. To give these principles enforcement power, the government amended the IT Rules in October 2025, specifically targeting deepfakes and AI generated misinformation. Platforms must label AI generated content with visible markers covering at least 10% of screen space. Failure to detect or remove deepfakes after user complaints strips platforms of Safe Harbour protection, exposing them to direct legal liability. Underlying this framework is a strategic emphasis on Data Sovereignty. By establishing the IndiaAI Datasets Platform and encouraging models trained on Indian data, policymakers aim to build AI systems that understand local languages, cultural contexts, and regional realities. For example, when a farmer is talking to an AI loan assistant, the responses should be based on his actual conditions and reality, not on assumptions derived from Western datasets. State-Level Push (The Ecosystem) India used to be center driven model where Delhi decided who could produce, how much, and where. Today, the Indian AI ecosystem is defined by a fierce "cooperative federalism," where states compete aggressively for capital and talent and that pushes the Innovation too. While Karnataka (Bengaluru) and Telangana (Hyderabad) are in a heavyweight title fight for general dominance, other states are carving out smart, specialized niches. Karnataka (Bengaluru) Karnataka is managing incumbency. Bengaluru is already known as the silicon valley of India, it has got the companies, the people, the startup vibe but does have the issues like traffic, housing costs & water shortages. so now the challenge isn't attracting new companies (that's happening anyway), it's keeping the companies/founders that are already there from leaving to different cities & managing overflow which is evident in it’s strategy of Retention & Decentralization. The Karnataka Startup Policy (2025-2030) allocates ₹518 Crore to launch 25,000 new startups, with a specific goal of pushing 10,000 to emerging clusters like Mysuru and Hubballi keeping growth in the state while relieving pressure on Bengaluru. To prevent deep tech founders from leaving entirely, the state is pairing this with some deep‑tech incentives, including significant R&D reimbursements and a 1000 crores dedicated deep‑tech & AI fund, so that companies building AI products can scale without relocating. Telangana (Hyderabad) Hyderabad is playing offense. Telangana knows it doesn’t have the legacy advantage of Bengaluru, so it is trying to win by building brand new AI infrastructure and positioning Hyderabad as the place where global AI companies and serious applied AI startups set up shop. The state came up with AI City Hyderabad project. It is a 200‑acre AI-focused zone (around the areas of Maheshwaram, Serilingampally, Chevella, Ibrahimpatnam) inside the new Bharat Future City, a 30,000 acre net‑zero greenfield smart city being developed on the southern side of Hyderabad. To make this more than a real‑estate story, Telangana is pairing the land with capital. The government has announced a ₹1,000 Crore Startup Fund in December 2025 structured as a Fund of Funds that backs venture funds investing in Telangana based, product‑led and AI‑first startups, with an ambition of creating 100 unicorns by 2034.  At the same time, Google has launched a Google for Startups hub at T‑Hub in Hyderabad, offering selected companies workspace, access to Google’s AI stack, cloud credits and mentoring. So Combining all of this, the pitch to founders is simple: “we’ll give you space, capital, and infrastructure if you build your AI company here.” The Niche Players (Specialization Strategy) While the big two fight for general dominance, other states are playing a smarter game by owning specific verticals. Maharashtra (The "Fintech & Infra" Hub) Leveraging Mumbai’s financial DNA, the Maharashtra Startup Policy 2025 established an "AI Sandbox" where fintechs can test risky AI financial products (like algorithmic lending) in a safe, regulated environment. Simultaneously, they are building a "Maharashtra Innovation City" in Navi Mumbai and kinda rivaling Hyderabad’s infrastructure play. Andhra Pradesh (The "Greenfield" Bet) Andhra Pradesh is running a two city deep‑tech strategy with Amravati being positioned as "Quantum City" being built from scratch with CRDA clearing ₹1,300 crores in late 2025 for core infrastructure and a quantum computing centre. Visakhapatnam, on the other hand, already has hard capital committed: Google and AdaniConneX announced a USD 15B global AI hub and data‑centre campus in 2025, making Vishakhpatnam one of the most important AI infrastructure nodes in the country for the coming decade. Tamil Nadu (The "Governance" Hub) Instead of chasing consumer apps, Chennai is building "AI for Governance." The Tamil Nadu AI Mission is releasing anonymized public data (health, transport, sanitation) to startups. The goal is to build SaaS solutions that make the government smarter, creating a unique B2G (Business-to-Government) market. Uttar Pradesh (The "Tier-2" Bet) UP is making a massive real estate play with India’s first designated "AI City" in Lucknow (₹10,700 Cr investment). The pitch is simple: "Bengaluru talent at Lucknow costs." It aims to attract back-office AI operations of global IT giants looking for cost arbitrage. Odisha (The "Engine Room") AI needs electricity. Odisha’s IT Policy 2025 offers huge power subsidies and 100% electricity duty exemptions to attract Data Centers. While other states build the software, Odisha wants to house the physical servers that power the AI revolution. and, - Delhi is framing a ₹400 crore venture capital fund for frontier tech (AI, robotics, fintech, climate tech, etc.) under its 2025–35 draft industrial policy. - Gujarat is using GIFT City’s IFSC status to build an AI‑driven fintech and algorithmic trading hub, tying AI to capital‑markets infrastructure. - Kerala is branding around GenAI design, digital creativity and AI ethics, emphasising talent, education and responsible AI over mega data‑centre builds. - Rajasthan is fusing AI with its existing gaming, animation and creative industries push to create a “Creative AI” niche around content, VFX and interactive media. 3. The Incumbents There are two kinds of incumbents in this story, and they used to make money in very different ways. On one side are the IT and consulting majors like TCS, Infosys, Wipro, HCLTech, Tech Mahindra and Cognizant that grew by selling large teams of engineers to global clients on multi year projects, taking responsibility for entire chunks of banking, retail or telecom tech stacks end to end.  On the other side are the infra, cloud and telecom majors like Microsoft, Amazon, Google, NVIDIA, Reliance Jio, Bharti Airtel, Tata and Adani whose business is to build and rent out the underlying compute, data centers and networks that everyone else runs on. The IT and consulting incumbents are basically trying to make sure GenAI bends their world without breaking it. They are setting up dedicated AI and GenAI practices, pulling AI into every digital transformation pitch and quietly changing delivery so that coding, testing, documentation and support are all AI assisted rather than pure human grind. A lot of effort is going into re skilling large chunks of their workforce and turning some of their domain know how into reusable AI “solutions” and platforms built on top of Azure, AWS or Google Cloud, so they can still charge for outcomes instead of just throwing ever larger teams at a problem. Underneath the branding, the core move is simple: shrink the number of people needed per project with AI, but keep the client relationship and the transformation mandate firmly in their hands. The infra, cloud and telecom incumbents are playing a much more concrete game around steel, silicon and land. Microsoft, Amazon and Google have announced tens of billions of dollars of India investments for 2025 to 2030, largely to build new hyperscale data centres and AI regions that can host the next wave of models and enterprise workloads. Telecoms like Reliance Jio and Bharti Airtel are teaming up with NVIDIA, Google and IBM to create gigawatt scale AI hubs in places like Jamnagar and Visakhapatnam, extending their networks into full blown AI ready cloud and data centre offerings. Around this, IndiaAI and similar public initiatives are adding a national layer of shared GPUs and datasets, which these players plug into so that, in practice, most serious AI in India rides on infrastructure built or co owned by this second bucket. 4. The startups to watch for These are the few startups that are interesting because they’re building around Indian constraints: language, cost, workflows, and real-world deployment. Sarvam AI Sarvam is focused on building foundational models that understand Indian languages and contexts. The emphasis isn’t on beating global models on benchmarks, but on making AI usable across India’s linguistic diversity. It’s one of the more serious attempts at sovereign AI done with practical intent. Krutrim Krutrim is building large language models optimized for Indian languages at scale. Backed by Ola, it’s positioned less as a research lab and more as an applied AI stack meant to power consumer and enterprise products across India. Qure.ai Qure.ai works on AI for medical imaging, particularly in radiology. Their systems help doctors detect conditions like tuberculosis, stroke, and lung disease faster and more consistently. This is one of the few Indian AI companies with real global clinical adoption. Niramai Niramai uses thermal imaging and AI for non-invasive breast cancer screening. It’s a strong example of AI being applied to a very specific, high-impact healthcare problem, rather than trying to be a general medical platform. Yellow.ai Yellow.ai builds conversational AI systems for customer support and enterprise workflows. The strength here isn’t novelty, but scale. They operate across languages, channels, and large customer volumes, which makes them a real infrastructure layer rather than a demo product. Observe.AI Observe.AI focuses on AI for contact centers, analyzing calls for quality, compliance, and performance. It’s an example of AI quietly reshaping large operational workflows, not by replacing humans, but by changing how work is monitored and optimized. CoRover CoRover builds conversational agents across a wide range of Indian languages, often used in government and public-facing services. It’s less flashy, but it solves a very real problem around accessibility and scale. 5. The Challenges India has plenty of technical AI talent. Engineers who understand machine learning pipelines, fine-tuning, and model deployment. The challenge isn’t technical capability. It’s workflow adaptation. Working effectively with AI requires changing how problems are approached. AI needs to be treated as the central tool, not an auxiliary feature. That shift only comes from hands-on practice. Trying different tools, breaking workflows, and rebuilding them around AI-first processes. The problem is that most AI productivity tools are built and priced for Western markets. Subscriptions for tools like Cursor, Claude Code, CodeRabbit, Suno, Notion AI, or specialized vertical AI applications add up quickly. For an Indian professional trying to experiment across coding assistants, writing tools, video generation, music, and more, this becomes prohibitive. Especially for students, early-career professionals, or people outside well-funded companies. This is compounded by structural dependence. Most of these tools, models, and platforms sit on foreign AI stacks and cloud infrastructure. Costs, access, and policy decisions are set outside India, while their impact is felt directly by Indian users and startups. Even when intent and talent are present, sustained experimentation depends on infrastructure and pricing that India does not fully control yet. Sources: https://www.cci.gov.in/images/marketstudie/en/market-study-on-artificial-intelligence-and-competition1759752172.pdf https://nasscom.in/knowledge-center/publications/india-generative-ai-startup-landscape-2025-mapping-momentum https://aign.global/ai-governance-insights/aign-global/indias-ai-governance-guidelines-2025/ https://pib.gov.in/PressReleasePage.aspx?PRID=2012375®=3&lang=2 https://aicityhyderabad.in/ https://aicityamaravati.com/ https://www.startup-movers.com/blog/ai-startups-in-india-opportunities-challenges-and-compliance-issues Models & AI tools used: Perplexity (Research) Claude & ChatGPT (Structure or Simplifying Explanations) Notion AI (Editing & Charts)