Here is a number that should change how every AI founder in India thinks about their infrastructure costs.
₹65 per GPU hour.
That is what the India AI Mission charges for access to high-end GPUs — including NVIDIA H100s, H200s, and Google Trillium TPUs. Meanwhile, commercial cloud GPUs cost ₹300-600 per hour on AWS and Azure. That makes the government’s offering somewhere between 5 and 9 times cheaper than commercial alternatives — and that is before accounting for enterprise pricing tiers and hidden data egress fees that can inflate cloud bills by 20 to 40%.
That is a 70% cost advantage for training the exact same models.
And the scale is real. India’s original IndiaAI Mission target was 10,000 GPUs. They have deployed over 38,000 — more than triple that goal. More than 38,000 GPUs have been empanelled through 14 AI service providers under IndiaAI’s shared AI compute infrastructure.
If you are training an AI model in India and you are still paying AWS or Azure rates, you are either uninformed or allergic to saving money. The India AI Mission has built one of the world’s largest public-sector GPU clusters — and it is open for business.
But subsidised GPUs are just one piece of a much larger system. The IndiaAI Mission is being implemented through seven pillars: IndiaAI Compute, IndiaAI Innovation Centre, IndiaAI Dataset Platform, IndiaAI Application Development Initiatives, IndiaAI FutureSkills, IndiaAI Startup Financing, and Safe and Trusted AI. Together, these pillars represent a total outlay of ₹10,371.92 crore — approximately $1.25 billion — deployed over five years.
Here is what each pillar actually offers founders, and how to access it.
Pillar 1: IndiaAI Compute — the GPU infrastructure that changes your cost structure
38,000+ GPUs at ₹65-92/hour through 14 empanelled providers — with up to 40% government subsidy
The mission will subsidise up to 40% of computing costs, significantly reducing the financial burden on startups, researchers, and public institutions. ₹4,563.36 crore is earmarked specifically for enhancing compute capacity — nearly half of the entire mission’s budget.
NVIDIA is collaborating with next-generation cloud providers Yotta, L&T and E2E Networks to deliver advanced AI factories. Yotta is building large-scale sovereign AI infrastructure branded as Shakti Cloud, powered by over 20,000 NVIDIA Blackwell Ultra GPUs.
The practical implications for an AI startup are enormous. If you are training a language model, fine-tuning a domain-specific AI, or running large-scale inference experiments, the cost difference between ₹65 per hour and ₹330 to ₹590 per hour adds up to lakhs — or crores — depending on your compute needs. A training run that costs ₹50 lakh on AWS could cost ₹6 to 10 lakh through the IndiaAI portal. That is not a marginal saving. That is a fundamentally different business model.
Eligibility extends to researchers, academic faculty, IndiaAI fellowship awardees, students, startups, MSMEs, and government entities, each with specific criteria and supporting documents.
How to access the compute portal — step by step
Users must register on the IndiaAI compute portal using Meri Pehchaan through their Digilocker, Parichay, or ePramaan account. Users fill a registration form and upload relevant documents. Once verified, users can submit a request for accessing compute infrastructure. Post-registration, users must submit a detailed project proposal, including technical approach, envisioned impact, and estimated bill of materials.
Approvals are valid for 30 calendar days, and users must start using the services within this period to avoid expiration. Move quickly once approved — the clock starts ticking.
Pillar 2: AI Kosh — the sovereign dataset platform
AI Kosh — a unified platform providing curated, privacy-compliant datasets for training AI models on Indian data
GPUs without data are useless. This is where AI Kosh enters.
AI Kosh is a unified data platform offering high-quality, non-personal datasets to researchers and startups for training AI models. AI Kosh enables controlled and secure access to India-based datasets, thereby supporting data governance and privacy.
According to government clarifications, there are no plans to monetise this data. It is free. For startups building India-focused AI products, this eliminates one of the biggest early-stage costs: data acquisition and labelling.
The platform covers multilingual datasets, domain-specific data across healthcare, agriculture, education, legal, and governance, and real-world Indian contexts that global models miss entirely. If you are fine-tuning a model for Indian languages, regional dialects, or culturally specific use cases, AI Kosh provides the training data that would otherwise cost months of collection and lakhs in labelling.
The platform also features AI sandbox capabilities — integrated development environments with tools and tutorials for building and testing models using the available datasets.
Pillar 3: Foundation Models — 12 startups selected, 500+ proposals received
Building India’s own large language and multimodal models — with 12 startups already selected across two phases
More than 500 proposals were received, and twelve startups were selected across the first two phases. These include Sarvam AI, Soket AI, Gnani AI, Gan AI, Avaatar AI, the IIT Bombay-led BharatGen consortium, Zenteiq, Gen Loop, Intellihealth, Shodh AI, Fractal Analytics and Tech Mahindra Maker’s Lab.
The resulting AI models will contribute to the open-source ecosystem and be available for use by Government organisations and also support innovation across India’s startup and research community.
For founders who are not building foundation models themselves: these models — once released as open-source — become the base infrastructure you build on top of. Instead of training from scratch, you fine-tune an Indian foundation model that already understands Hindi, Tamil, Bengali, and 19 other scheduled languages. Your development time drops from months to weeks. Your compute costs drop by orders of magnitude.
For a nation as multilingual as India — with 22 constitutionally recognised languages and over 1,500 more recorded by the country’s census — frontier AI models are a powerful tool to help its more than 1.4 billion residents interact with technology in their primary language.
Pillar 4: Startup Financing — ₹2,000 crore earmarked for deep-tech AI ventures
₹2,000 crore allocated for supporting deep-tech AI startups — plus a global acceleration programme with Station F in Paris
Approximately ₹2,000 crore will be allocated for supporting deep-tech startups. This pillar supports and accelerates deep-tech AI startups by providing streamlined access to funding for innovative AI projects.
The IndiaAI Startups Global initiative, in partnership with Station F and HEC Paris, offers 10 selected Indian AI startups a 4-month funded programme in Paris for global market expansion, mentorship, and investor connections.
NVIDIA is also partnering with prominent venture capital firms including Peak XV, Z47, Elevation Capital, Nexus Venture Partners and Accel India to identify and fund promising startups that are building AI solutions for India and international use. More than 4,000 of India’s AI startups are already part of the NVIDIA Inception programme.
For founders, the Startup Financing pillar is not a single grant application — it is an ecosystem of capital channels. The ₹2,000 crore allocation flows through multiple mechanisms, from challenge-based grants to VC-facilitated investments. Creating deep-tech AI requires long, risk-taking capital. This pillar is intended to streamline the funding process for deep-tech AI startups and fast-track the transition from idea to implementation.
Pillar 5: Application Development — 30 India-specific AI applications funded
The IndiaAI Application Development Initiative promotes AI application development to address India-specific challenges. Priority sectors include healthcare, agriculture, climate change, governance, and assistive learning technologies. By July 2025, thirty applications had been approved.
If you are building an AI application that addresses an Indian problem — crop disease detection, rural health diagnostics, vernacular education, governance automation, disaster prediction — this pillar provides both funding and a pathway to government deployment.
Pillar 6: FutureSkills — 13,500 scholars and 31 Data & AI Labs in Tier 2/3 cities
IndiaAI Future Skills aims to develop a strong pool of AI talent. It supports 500 PhD fellows, 5,000 post-graduate students, and 8,000 undergraduate students. More than 200 students have been provided with fellowships, and 73 institutes have admitted students to the PhD programme. Data and AI Labs will be established in Tier 2 and Tier 3 cities, with 31 Data and AI Labs launched in association with NIELIT and industry partners.
For AI startups struggling to find talent outside metros, this is the pipeline being built. 31 labs in Tier 2 and 3 cities, 13,500 scholars in AI training, and a massive push to decentralise AI education across the country. In two to three years, the graduates from these programmes become your hiring pool.
Pillar 7: Safe and Trusted AI — the governance framework
The Safe and Trusted AI pillar focuses on the responsible adoption of AI. Thirteen projects have been selected to address issues such as machine unlearning, bias mitigation, privacy-preserving machine learning, explainability, auditing, and governance testing.
For founders building AI products that touch sensitive data — healthcare records, financial information, government data — this pillar provides the governance frameworks, tools, and guidelines that make your product compliant and trustworthy by design.
The infrastructure behind the GPUs — who is building it
Understanding who is physically building the compute infrastructure helps you make better decisions about which provider to use through the portal.
- Yotta (Shakti Cloud): Over 20,000 NVIDIA Blackwell Ultra GPUs across campuses in Navi Mumbai and Greater Noida — GPU-dense, high-bandwidth AI cloud services on a pay-per-use model
- Larsen & Toubro: Building sovereign, gigawatt-scale NVIDIA AI factory infrastructure with expansions in Chennai (30 MW) and Mumbai (40 MW)
- E2E Networks: Building an NVIDIA Blackwell GPU cluster on its TIR platform at the L&T Vyoma Data Center in Chennai — featuring HGX B200 systems
- Netweb Technologies: Manufacturing NVIDIA GB200 NVL4 platforms in India under Make in India — featuring four Blackwell GPUs and two Grace CPUs per system
Analysts expect private orders for another 100,000 GPU devices through 2026. The infrastructure is not static — it is expanding rapidly, which means access is likely to improve and costs may drop further as capacity grows.
The honest assessment: what works and what does not yet
The infrastructure is real and the subsidies are genuine. But there are honest caveats founders should know.
Only about ₹800 crore was spent last year against the allocated budget — suggesting that the disbursement machinery has not yet caught up with the ambition. Startups complain about complex allocation paperwork. The approval process requires project proposals, technical documentation, and verification through Digilocker — which is more bureaucratic than signing up for AWS.
IndiaAI subsidises approved projects, but users must bear additional costs beyond the subsidy. The 40% subsidy does not mean everything is free — you pay the remaining 60% directly to the service provider.
And the FY27 budget earmarked ₹1,000 crore for IndiaAI — which analysts noted was a sharp cut from the previous year, though the total approved multi-year outlay remains ₹10,372 crore.
The practical advice: apply now, because early adopters get the smoothest experience. But budget for the non-subsidised portion, prepare thorough project proposals, and do not expect instant approval. The bureaucracy is real — but the cost savings are also real.
Who should use the India AI Mission — and how
✅ IndiaAI is built for you if you are:
- Training or fine-tuning AI models: The compute subsidy makes frontier experimentation viable at a fraction of commercial cost
- Building for Indian languages and contexts: AI Kosh datasets and the 12 selected foundation models are specifically designed for India’s linguistic diversity
- A deep-tech AI startup: ₹2,000 crore earmarked for your category, plus the Station F global programme and NVIDIA Inception access
- Building AI applications for healthcare, agriculture, education, or governance: The Application Development pillar funds India-specific AI deployments
- A researcher or academic: Full access to compute, datasets, and the FutureSkills programme’s fellowship and lab infrastructure
- A DPIIT-registered startup or MSME: Explicit eligibility for subsidised compute access through the portal
Your action plan — step by step
✅ Step 1: Register on the IndiaAI Compute Portal
Visit the IndiaAI Compute Portal (accessible via indiaai.gov.in). Register using your Digilocker, ePramaan, or Jan Parichay account. Complete your profile with required documents — DPIIT registration, company incorporation, and project details.
✅ Step 2: Submit a project proposal for compute access
Your proposal must include a technical approach, envisioned impact, and an estimated bill of materials based on the published rate list. Proposals aligned with national priorities — multilingual AI, healthcare, agriculture, governance — receive stronger consideration. Include specific milestones and measurable outcomes.
✅ Step 3: Access AI Kosh for training data
Visit aikosh.indiaai.gov.in to browse available datasets, models, and sandbox capabilities. The platform is free. If you are building for Indian markets — especially vernacular or domain-specific applications — AI Kosh datasets can eliminate weeks of data collection and thousands in labelling costs.
✅ Step 4: Explore the Startup Financing pillar
Monitor indiaai.gov.in for open calls under the Startup Financing pillar. The Station F global programme selects 10 startups per cohort for a 4-month funded programme in Paris. The NVIDIA Inception programme is already accessible to 4,000-plus Indian AI startups — join if you have not already.
✅ Step 5: Watch for Foundation Model calls and Application Development challenges
Phase 3 calls for foundation model proposals have been running — track indiaai.gov.in for new rounds. The Application Development Initiative funds AI solutions for India-specific problems. Prepare challenge proposals with clear sector alignment and deployment plans.
How the India AI Mission stacks with other government support
IndiaAI operates independently of state startup schemes. This means you can access subsidised GPU compute and AI Kosh datasets while simultaneously claiming Karnataka ELEVATE grants, Tamil Nadu TANSEED seed funding, Maharashtra MSInS support, or any other state-level programme. Central and state benefits are additive — they do not cancel each other out.
For AI founders specifically, the stacking potential is significant: IndiaAI subsidised compute plus state startup grants plus DPIIT recognition benefits (Section 80-IAC tax holiday, CGTMSE collateral-free loans) plus BIRAC grants (if biotech-AI) plus SISFS seed funding through incubators. Each channel operates independently, and a disciplined founder can access multiple channels simultaneously.
Why this window matters in 2026
Here is the honest strategic picture for AI founders in India right now.
The infrastructure is live. 38,000 GPUs are deployed. The subsidies are active. The dataset platform is free. Twelve foundation models are being built on Indian data and will be open-sourced. And the global competition to attract AI startups is intensifying — with countries from Singapore to the UAE to France all offering compute subsidies and acceleration programmes.
India’s advantage is structural: the system offers high-power computing resources at a subsidised rate of ₹65 per hour — one of the lowest rates globally. Combined with India’s cost advantage in talent, the multilingual dataset infrastructure, and the growing domestic market for AI applications, the economics of building AI in India have never been more favourable.
The founders who register on the compute portal now — while the system is still scaling and competition for allocation is growing — will have the strongest position as the infrastructure matures. The GPUs are live. The datasets are free. The foundation models are coming. The startup financing is allocated. The question is not whether the infrastructure exists. It does. The question is whether you will use it.
₹65 per GPU hour versus ₹330 to ₹590 on commercial clouds. 38,000 GPUs deployed — three times the original target. Sovereign datasets free to access. ₹2,000 crore for deep-tech AI startups. 12 startups building India’s own foundation models. Stanford ranks India third globally in AI competitiveness. The infrastructure is live. Register at the IndiaAI Compute Portal and start building.
Access the India AI Mission this week
Step 1: Register on the IndiaAI Compute Portal via indiaai.gov.in using Digilocker or ePramaan. Step 2: Submit a project proposal with your compute requirements. Step 3: Browse AI Kosh at aikosh.indiaai.gov.in for free training datasets. Step 4: Monitor the Startup Financing pillar for open funding calls and the Station F global programme. Step 5: Stack with state startup schemes and DPIIT recognition for maximum support.
38,000 GPUs. ₹65/hour. Up to 40% subsidy. ₹10,372 crore total mission outlay. ₹2,000 crore for deep-tech startups. Free sovereign datasets. 14 empanelled service providers. The infrastructure is the largest public AI compute deployment in the Global South.
The cost barrier to building AI in India just collapsed. The founders who move now have a 12-24 month structural cost advantage over everyone who waits.