India's AI Ambitions: Government Receives 67 Proposals for Indigenous Foundational Models

India to develop its own AI model

Hot on the heels of DeepSeek, India has announced plans to develop its own AI model, aiming to compete with industry giants in the United States and China. The country's first indigenous AI foundational model will be ready within 10 months, the information technology minister, Ashwini Vaishnaw, said last month at a press conference. "We are all set to launch our own safe and secure indigenous AI model at an affordable cost," Vaishnaw said. "This will help India emerge as a reliable technological powerhouse of ethical AI solutions." The model is expected to be not only secure but also cost-effective. While global AI models charge between US$2.50 and $3 per hour for computation, India's AI system is projected to cost less than $1.15 per hour after a 40% government subsidy, making it a competitive alternative. A startup ecosystem and government initiatives have spurred India's push into AI, with the country now ranked fourth in AI innovation, according to Stanford University's Global AI Index. The index, which assesses AI ecosystems across 36 countries based on research papers, private investments, and patents, places India just behind the United States, China, and the United Kingdom. Recognising AI's potential to transform industries, the Indian government has pledged to strengthen AI infrastructure and fund more research and development. In March last year, the Indian cabinet, chaired by Prime Minister Narendra Modi, approved the IndiaAI mission with a budget of ₹10,372 crore. The initiative aims to create a robust AI ecosystem through strategic programmes and public-private partnerships. Key priorities include improving data quality, developing local AI resources, and establishing a framework for ethical AI development. The Union Budget 2025, announced earlier this month, reinforced this commitment, allocating ₹2,000 crore to the IndiaAI mission -- a huge jump from the Rs ₹551 crore assigned to the mission in 2024, earmarking funds for the creation of three AI Centres of Excellence in education. These will aim to integrate AI into academic institutions, strengthening AI research, and promoting digital learning. India has made significant strides in AI research, yet large-scale adoption of the technology remains sluggish. Experts argue that a multi-pronged approach involving collaboration between policymakers, academia, and the private sector is necessary to accelerate AI adoption. "India is one of the best prepared nations for AI transformation," said S. Krishnan, secretary of the ministry of electronics & information technology, speaking at a panel discussion on AI innovation at the Indian Institute of Science. "There is a need to boost investment and build a robust AI infrastructure. Policy interventions should be made to promote AI, while addressing ethical concerns." AI development in India faces persistent challenges, including limited access to high-quality data, a shortage of skilled professionals, privacy and security concerns, and a lack of clear regulatory frameworks. Luke Preskey, chief revenue officer of market intelligence firm, Resonance, emphasized the need for greater public-private collaboration. "India has the talent and scale to be a global AI leader, but unlocking its full potential requires stronger investment in AI research and a regulatory framework that supports innovation, while ensuring ethical use," he told Nature India. Despite strong government backing and increased collaborations, India's AI ecosystem continues to face structural hurdles. There are concerns over data accessibility, ethical risks, and regulatory ambiguity, which could slow AI's full-scale implementation. "Collaborative efforts from various stakeholders will ensure that AI propels inclusive and sustainable economic progress," Krishnan added. The success of India's homegrown model will depend on the country's ability to strike a balance between innovation, affordability, and ethical AI governance, Vaishnaw pointed out.

Indian Govt Receives 67 Proposals to Build Foundational Models

The electronics and IT ministry is organising a high-level technical committee to evaluate the proposals. Ever since China's DeepSeek made its mark on the global AI space, India has been exploring building its own foundational models. This is important, especially considering the security issues revolving around DeepSeek. India recently received a significant boost with a ₹2,000 crore investment in the Union Budget 2025 for the IndiaAI mission. In January, the mission also started inviting startups, entrepreneurs, and researchers to collaborate on building foundational AI models trained on Indian datasets. Moreover, at the Utkarsh Odisha Conclave held last month, IT minister Ashwini Vaishnaw said, "Very soon, we will have our own large language model (LLM)." Now, it seems that the proposals for it have started coming in, as reported by The Indian Express. A senior government official reportedly said that the Centre had obtained at least 67 proposals to build an AI model focused on India, 20 of which were LLMs. It is also known that the electronics and IT ministry is organising a high-level technical committee to evaluate the proposals. The committee will include external domain experts, who will review the proposals and decide whether to accept or reject them within a month. The report further states that Sarvam AI, CoRover.ai, and Ola are some of the prominent names among the proposals that have expressed interest in building a foundational model. A government official, under the condition of anonymity, told The Indian Express, "Of the total 67 proposals, 20 proposals have been received for building LLMs. The remaining are sector-specific small language models, including one which has been created by a group of doctors to assist medical professionals working on breast cancer issues." With the proposals arriving, it seems that the process of building India's own foundational model should start soon. It will be exciting to see the outcome of when that happens.

This Govt Funded ₹235 Cr AI Initiative is India's Real Answer to DeepSeek

Spearheaded by the government and driven by top researchers, BharatGen is not just another large-scale AI project -- it is the cornerstone of India's technological sovereignty. China's DeepSeek, which is claimed to be built under $6 million, has taken India by storm. The race to build its own foundational AI model has started in the country for which the IndiaAI Mission has received 67 proposals, with 20 of them planning to build LLMs. But amidst all the proposals and startups, BharatGen, the Department of Science and Technology-funded initiative, stands out as the nation's strongest bet. Spearheaded by the government and driven by top researchers, BharatGen is not just another large-scale AI project -- it is the cornerstone of India's technological sovereignty. "We have made significant technical progress, and the announcement of our models has already come from the DST secretary," Prof. Ganesh Ramakrishnan, the head of the BharatGen initiative, told AIM. He said this while discussing the project's rapid progress and growing role as a national AI mission. Unlike private entities, BharatGen operates with a clear mission -- 'GenAI for Bharat, by Bharat'. With an investment of under ₹235 crores, which is close to $27 million, leveraging cost-efficient computing and attracting top talent from graduates from IITs. The BharatGen consortium comprises IIT Bombay, IIT Kanpur, IIT Mandi, IIT Madras, IIIT Hyderabad, and IIM Indore. Rather than being confined to a single department, BharatGen has been structured to benefit from a whole-of-government approach, ensuring seamless inter-ministerial collaboration. "The first goal was to get BharatGen into a national mission, which has happened. The second is to build from the ground up while ensuring all intellectual property (IP) remains with India," Ramakrishnan explained. However, Ramakrishnan clarified that the funding goes towards model building and supporting the broader AI ecosystem. "People often look at numbers in isolation. The real story is about provisioning for long-term research, infrastructure, and talent development," he added. This explains why DeepSeek is unable to sustain itself with minimal funding and the cost of inference when it comes to AI models, which is the highest. The endorsement from Union Minister Ashwini Vaishnav further strengthens this vision. "The Minister was briefed about BharatGen, and he clearly said, 'India is building its own foundation model,' not just Indian companies," Ramakrishnan emphasised. Vaishnaw recently stated that India would have its own foundational AI models within 7-8 months, and BharatGen is a key part of that vision. "Yes, we are very much on track. The Minister has been briefed, and we are aligned with the timeline," Ramakrishnan confirmed. Unlike private players like Sarvam or Krutrim, BharatGen maintains a stronger academic foundation. "Private players operate in silos with no academic connection. How sustainable is that in the long run?" Ramakrishnan questioned. BharatGen, on the other hand, integrates research institutions and faculty members into its ecosystem, ensuring long-term innovation. While private startups have been receiving cloud infrastructure support, BharatGen primarily leverages Indian providers like Yotta and Neysa. "Our goal is not just to build AI models but to provide resources that startups and system integrators can leverage," said Ramakrishnan. The team is already on track to building Bharat Datasagar, a multilingual repository for Indian AI research and has already released e-VikrAI, a solution built on Vision Language Models, tailored for product images in Indic e-commerce. A key point of contention in the AI discourse is whether India should rely on private firms to develop foundational AI models or if the government should take charge. Some argue that startups should take the lead, like in the US and China. Ramakrishnan, however, sees this as a misguided approach. "This is not about competition; it's about necessity," Ramakrishnan concluded. "India cannot afford to depend on foreign AI models. The future of AI in India must be built in India." "Inclusiveness is at the heart of BharatGen. Private companies, no matter how well-intentioned, will always prioritise business interests. Why would they focus on India's dialects, agriculture, or public services?" he asked. For BharatGen, AI is more than just a business opportunity -- it is a strategic necessity. The initiative is working to build models that cater to India's diverse linguistic and economic landscape. "This is not just about creating models; it's about ensuring they serve the people," he said. Moreover, security concerns make it imperative for India to own its foundational AI models. "Relying on foreign companies is not viable for mission-critical applications like defence. It's similar to nuclear power -- we need our own AI infrastructure," Ramakrishnan stated. With a team of 50-60 researchers and a large network of student contributors, BharatGen is moving forward rapidly. The initiative constantly holds discourse on infrastructure, including GPU allocations, with the government.

Seeking the right model: What's keeping India from building its own LLMs

Swathi Moorthy and Annapurna Roy look at the major bottlenecks in building Indian foundational models and what's needed to bolster the innovation ecosystemSpurred by the launch of Chinese-made foundational model DeepSeek, reportedly built with a few million dollars and outdated GPUs, the Indian government called for proposals to build an Indian foundational model capturing the country's languages, supported by grants in the form of compute credits, equity investments and co-financing options. People from around the world hailed DeepSeek for demonstrating that a foundational model can leverage innovative techniques without having to shell out big bucks, and is pegged as an example of how every country including India can make it big in the world of foundation models. But India's innovation ecosystem still confronts several challenges in going the foundational model route. A historically lagging AI research backbone, insufficient quality data repositories and not enough capital may keep India from having its own DeepSeek moment any time soon. Yet, some pockets are making bets on opportunities they see in smaller models, while others are diving into research to create a truly representative foundational model for India. Late to the game A point often overlooked is that China's ascent to rival the US is on the back of years of research and funding into its AI ecosystem -- a relatively recent focus in India. China first announced its ambitious AI programme in 2017, where it laid out a roadmap to become a major player in AI by 2030. By 2020, China had already poured $60 billion into AI development, and the economic value of these investments are expected to reach $600 billion by 2030, according to multiple reports. According to a report from the World Intellectual Property Organisation, between 2014 and 2023, 38,000 GenAI inventions came out of China, 6X more than the second place, the US. Compared to this, India had filed only 5,000 in the last decade, according to a Nasscom report. Industry voices believe India does not have this kind of research ecosystem. Naganand Doraswamy, co-founder, Ideaspring Capital, said that while companies such as Sarvam AI have been building Indian language AI models on top of open-source offerings, it requires years of research, discussions and exchange of ideas, which has not happened in India at the scale required. DeepSeek is the culmination of years of research and investments that have gone into the sector, he said. Data unlocks needed When it comes to AI, experts believe that the biggest moat is data. Theoretically, India with its 1.4 billion population, of whom 600 million have smartphones, should have the largest repository of data. The challenge, however, is that much of the data is split across multiple Indian languages and dialects. An investor, on the condition of anonymity, said that in addition while India had the advantage of volume, the digital footprint compared to a country like the US, Europe or China is still very new and that puts us at the disadvantage. While companies like Sarvam are using synthetic data to solve the issue to a certain extent, the lack of quality data in Indic languages will impact synthetic data generation as well. Unadventurous capital One of the reasons DeepSeek made headlines was the cost of training, pegged at just $5.576 million as per the DeepSeekV3 whitepaper released in December 2024. But some flagged that this only includes training of the V3 model and does not include other costs such as that of GPUs and research. The company did not disclose the training cost for its DeepSeek-R1 model. People in the industry said that the cost of training the models could be far greater than $5 million -- a significant investment. Somshubhro Pal Choudhary, co-founder and partner, Bharat Innovation Fund, told ET that it only makes sense to invest if companies are building on top of original research and differentiating technology moat. "Most of what we have seen is incremental on the AI side and very soon it has the risk of commoditisation," he added. The way forward While the industry is on the fence about large language models, SLMs or small language models are slowly creating waves and investors are taking bets on them. Mohamad Faraz, managing partner, Upsparks Capital, said they are looking to invest in SLMs that are solving for a specific use case. The firm has invested in Smallest.ai, which is building voice models. On the fundamental innovation front, Pushpak Bhattacharya, professor of computer science and engineering, IIT Bombay, said that while building a foundational model from scratch is possible through a collaboration between government, industry and academia, this would be expensive and time-consuming. A more practical, compute-efficient and cost-effective approach that would address India's diversity would be a mixture of "trinity" models, he said -- models that bring together a specific language, domain and task. "For example, a model in Manipuri, for the sports domain, for the task of sentiment analysis," he said. He explained that these can be built cheaper, smaller and faster, each requiring perhaps 80-100 GPUs. IIT Bombay has built a large language model for consumer grievance redressal with National Law School of India University, Bengaluru, funded by Meta. It is working on another for the railway complaint handling system. "We are using government funding from Meity for machine translation models. We are part of the Bhashini project, for example," Bhattacharya said. "We are also applying for funding for these kinds of trinity models." On Indian talent in global tech and AI companies, who may already have the knowhow of foundational models, Sangeeta Gupta, SVP & chief growth officer, Nasscom, said, "I think many are keen to contribute. Whether they will start up in India is a different question, but even if we can engage with them meaningfully on some of our big ambitions, that's a good starting point."

Govt Receives 67 Proposals For Building AI Foundational Model

This comes a month after MeitY invited applications from Indian startups for building an India-specific AI foundational model under the INR 10,370 Cr IndiaAI Mission The union government has reportedly received at least 67 proposals for building an indigenously-designed and developed artificial intelligence (AI) foundational model. A senior government official told The Indian Express that the applications also include proposals to build 20 large language models (LLMs). As per the report, homegrown AI startups such as Sarvam AI, CoRover.ai and Ola's Krutrim have submitted applications for building the LLMs. "Of the total 67 proposals, 20 proposals have been received for building large language models. The remaining are sector-specific small language models, including one which has been created by a group of doctors to assist medical professionals working on breast cancer issues," the official reportedly added. Meanwhile, the Ministry of Electronics and Information technology (MeitY) is reportedly setting up a "high-level technical committee" to evaluate the said proposals. The panel, which will reportedly comprise experts from "outside", will review the applications and take a call on their feasibility in a month. This comes a month after MeitY invited applications from Indian startups and enterprises for building an India-specific AI foundational model under the government's flagship INR 10,370 Cr IndiaAI Mission. The development comes weeks after union IT Minister Ashwini Vaishnaw said that India would build its own foundational AI model in the next 10 months. At the time, the minister also said that the country would be able to develop its own high-end graphics processing unit (GPU) in the next three to five years. While inviting proposals from startups and researchers earlier this month, IndiaAI Mission said that the project will look to develop a scalable foundational AI model, which should be trained on diverse datasets and demonstrate capabilities to address India-specific challenges and opportunities across sectors. As part of this, the Centre has expressed willingness to offer both equity-based funding as well as compute credits for the selected entity(s) that will undertake the task. Meanwhile the union government has also empanelled ten companies to supply 18,693 GPUs for developing the foundational model including the likes of Jio Platforms, Yotta, Tata Communications, E2E Networks, among others.