IndiaAI has launched a Call for Proposals (CFP) to foster the development of Indigenous foundation AI models trained on Indian datasets. This initiative seeks to empower startups, researchers, academics, and private enterprises in building large language models (LLMs), Small Language Models (SLMs), large multimodal models (LMMs), and other advanced AI frameworks aligned with India’s linguistic and cultural landscape.

An informational webinar was held to clarify the application process and address key questions from prospective applicants. It featured an in-depth Q&A session led by Shri Abhishek Singh, Additional Secretary, MeitY & CEO, IndiaAI Mission and Aakrit Vaish, Advisor, IndiaAI Mission. The session provided valuable insights into the scope of funding, technical expectations, evaluation criteria, and strategic focus areas.

You can find the video of this entire session here 

Q&A Session: Key Insights

Q: What's the range of funding available for the RFP?

A: The justification of cost and funding that you provide as part of your proposal will be examined by experts. Your approach to building the model will be the criteria for funding. A total of Rs. 1,971.37 Cr has been allocated by the Cabinet for the Foundation Model pillar, as part of the IndiaAI Mission.

Q: Can there be a consortium of companies coming together?

A: Yes, a consortium can be formed between individuals, researchers, a group of experts, scientists or a group of companies. Once a proposal by a consortium is selected, it will need to be incorporated as an entity within a month.

Q: What will be considered as a foundation model?

A: A foundation model that uses large amounts of training data (text, video, audio), to perform downstream tasks. It can be a horizontal model that can be used for general purposes across domains or a vertical model for use cases in a specific domain.

Q: Are there any minimum requirements of participation?

A: There are no specified minimum requirements, but the capability of the team and past background in research will be very important. We will select a finite number of proposals based on team, approach and overall milestones.

Q: Would there be a provision for providing access to compute? What is the biggest single cluster of H100s readily available for training in the AI Mission?

A: As part of the AI Compute pillar of the IndiaAI Mission, AI cloud service providers have been empaneled with lowest rates across various GPUs. You can find the discovered L1 rates here.

The current biggest single cluster is roughly 4,000 GPUs, but it is expected this will increase with more demand from teams/entities building Foundation Models.

Q: How will the data sets for training be sourced? Who will own the data sets post the models. If the bidders tend to procure proprietary data sets, what will that cost also be covered in the proposal?

A: IndiaAI Mission is developing India Datasets platform to make high quality AI ready data accessible to startups & researchers. In general, bidders must include all costs, including proprietary data procurement, in their proposals. IP rights for self-created/procured data belong to the entity.

Q: Does the source of external funds need to be India based, or can be outside India? Does the entity that is applying for this have to be India, or can it be outside?

A: The applying entity must be based in India, whether it's a for-profit company, research group, or nonprofit organization. However, additional funding from legal sources can be from anywhere, including foreign sources. The Indian registration requirement is a strict criterion for applicants.

Q: What are the milestones expected at which grants will be dispersed?

A: The justification of cost and funding that you provide as part of your proposal will be examined by experts. Your approach to building the model will be the criteria for funding. The quality of your proposal will be one of the key evaluation criteria.

Q: Will a SaaS product be funded, or does the proposal need to focus on hardware?

A: The funding is specifically for building LLMs and foundation models.

Q: What are the expectations regarding cultural fit? How many languages should the model support, and are there any mandatory languages?

A: Applicants must propose their approach, which will be evaluated against other submissions in the same category. The more languages supported, the better.

Q: Can funding be used for healthcare data collection?

A: Yes, but applicants must specify their data collection approach and cost components in the proposal.

Q: Are proposals limited to LLMs, SLMs, and LMMs, or can other models be considered?

A: Any Foundation Model, which uses large datasets and machine learning to generate downstream tasks will be considered. This includes Large Language Models (LLMs), Small Language Models (SLMs) focused towards a specific domain, or Large Multimodal Models (LMMs), or even Large Quantitative Models (LQMs).

Q: How is this funding initiative different from Startup India Funds?

A: This funding is exclusively for LLM development, distinct from Startup India’s broader funding programs.

Q: Is there a focus on supporting models that reflect the Indian cultural context?

A: Yes, models should be aligned with the Indian context, and supporting multiple Indian languages is encouraged.

Q: What are government-preferred cloud, compute, and storage environments for AI workloads?

A: As part of the AI Compute pillar of the IndiaAI Mission, AI cloud service providers have been empaneled with lowest rates across various GPUs.

You can find the discovered L1 rates here 

Q: Are foundation models in the healthcare sector eligible for funding?

A: Yes, models focused on healthcare, including those with pending patents, are encouraged.

Q: Can students submit proposals?

A: Students are encouraged to apply, based on their capability and approach towards building foundation models.

Q: Is dataset creation an eligible expense?

A: Yes, please include all your potential expenses including those for dataset procurement or dataset creation. Funding will be provided for the project as a whole based on the evaluation framework.

Q: If sensitive datasets (e.g., financial or transactional data) are unavailable, can IndiaAI facilitate access through industry partnerships?

A: Yes, IndiaAI will assist in ensuring compliance and secure access where possible.

Q: What is the funding timeline? Will it be disbursed in phases?

A: Applicants must define project milestones and expected funding timelines in their proposals. The goal is to have the initial versions of the models to be ready within 9-12 months.

Q: Can the approach and architecture be adjusted if the AI model development evolves during training?

A: Yes, applicants should provide a precise initial plan, but modifications based on evolving needs are allowed.

Q: Does the entity applying have to be in India?

A: Yes, the entity for which the grant will be provided has to be headquartered in India.

Q: Can distillation and further SFT with GRPO on domain specific data can be considered as a foundation model?

A: Yes, a model can be considered a foundation model if trained on domain-specific data, provided the core training data is proprietary to the model. However, specific details and nuances would need to be evaluated on a case-by-case basis once proposals are received.

Q: Can people who are submitting proposals be part of multiple non-related proposals?

A: It is advisable that people submitting proposals are part of one team only.

Q: Under the proposal it is stated compliance with DPDP Act and all relevant regulations is required. What does this mean?

A: Applicants must ensure their data set complies with the Digital Personal Data Protection Act and other applicable regulations.

Q: What is the approximate timeframe for proposal evaluation, let's say, if somebody submits it on the 15th Feb. When will they hear back on whether they're selected or not?

A: Evaluations will be completed preferably within one month of submission, with the goal of completing reviews before the next round begins. While the exact timeline depends on the proposal volume, applicants will receive feedback. The team aims to process all evaluations within a month of submission

Q: Is section 8. company acceptable?

A: Yes

Q: Does the Foundation model need to be disruptive or incremental?

A: The Foundation model must be built from the ground up, not fine-tuned from existing models. It should be developed independently ensuring a unique and original base model for the intended purpose.

Q: Can a foundation model for vision systems be considered?

A: Yes, absolutely. That's a different modality in the multimodal. So it absolutely can be considered.

Q: What are the specific domains of proposals?

A: There is no predefined domain for proposals.  You can explore any domain based on your expertise and capabilities. The scope remains open to diverse topics.

Q: If you already have funding from another government agency but need more funding. Can we apply for a grant?

A: Yes, you can apply for a grant even if you have funding from other government agencies, the private sector, or venture capital. Just be sure to disclose your existing funding as part of your background and team information.

Q: How many proposals can be accepted?

A: The number of accepted proposals is open-ended, depending on quality and feasibility. It could be as few as one or two if submissions are limited, but potentially up to 6–10 if many strong proposals meet project milestones. The focus is on selecting high-quality proposals that can successfully complete requirements.

Q: Can startups collaborate with academic research institutes?

A: Yes, startups can collaborate with academic research institutes. Consortiums are also welcome, and companies are free to partner with any entity they find suitable for collaboration.

Q: Are there any priority use cases or specific domains where India wants to see foundation models being developed?

A: While social sectors such as Agriculture, Healthcare, Education, Climate and Governance are generally important to the IndiaAI Mission, for foundation models the capability of the model is of prime importance and there is no such preference to any sectors.

Q: How is the capability of the team assessed? Are there any preferred qualifications or past experiences?

A: The team’s capability is assessed based on track record, relevant experience, research background, and published work. Both early-career and senior researchers are valued, with emphasis on the quality of contributions rather than years of experience, given the fast-evolving nature of the field.

Q: Will IndiaAI provide ongoing support post funding to ensure long-term success?

A: IndiaAI will assess sustainability plans in proposals to ensure long-term success. Funding for sectoral models may arise later if adopted. The approach is open-ended, allowing diverse solutions. Proposals must clearly outline sustainability, use cases, and foundational model aspects to convince evaluators of their independent viability.

Q: How will evaluations be done? Will there be a committee? Will there be presentations?

A: Yes, there will be a committee which will initially screen the proposals, and those who are found suitable at the 1st level of screening will go to the next round of presentations with the Technical Committee.

Q: Does the proposal need to include cost of inference for the model also? Or is this just for training?

A: You must include all costs related to your approach, including inference and headcount payroll. The proposal covers both foundation model development and its application. While all costs can be included, additional funding sources are allowed. Consider this as gap funding rather than exclusive funding.

Q: Since the IP belongs to the entity. Can they explore monetizing the model?

A: Yes, they can monetize the model, but the government retains the right to use it for public purposes. For example, if the model relates to agriculture and has relevant applications, the government can utilize it within that sector.

Q: How will the models be evaluated?

A: The evaluation will rely on benchmarks relevant to each use case, as defined by the teams. Initially, proposals should outline target benchmarks. After development, the same or updated benchmarks will be used.

Conclusion

The IndiaAI Call for Proposals represents a significant opportunity to drive innovation in large-scale AI models tailored to India’s linguistic, cultural, and economic needs. This initiative aims to strengthen India's position as a global AI leader by encouraging participation from startups, academia, and enterprises.

With a focus on linguistic diversity, ethical AI, and real-world applicability, the program provides a structured pathway for researchers and developers to secure funding and build AI solutions that address India-specific challenges.

Prospective applicants are encouraged to carefully review the CFP guidelines and submit well-structured proposals that align with IndiaAI’s vision of self-reliant and cutting-edge AI innovation.

For more information: Guidelines