IndiaAI Mission runs into challenges in building indigenous technology

Bernstein, a wealth management firm, in a recent report raised sharp questions about India’s ₹10,372-crore artificial intelligence (AI) mission, warning that the programme risks becoming inconsequential on the global stage. 

The report said the “highly publicised government fund — spread thinly across a handful of startups building foundational models — barely registers globally.” It feared that American organisations with deeper pockets and better infrastructure are likely to dominate AI, as they did in earlier waves of technology. 

The report flagged risks from allowing companies such as OpenAI, Perplexity and Gemini to enter India’s market with discounted versions of their AI models, which could undercut domestic startups. “India is at a crucial point — allowing access to foreign LLMs could cripple the local ecosystem, while banning them would seem far too escalatory,” the report noted, referring to large language models, foundation blocks of AI trained on massive amounts of text and code to understand, summarise, and generate human-like language. 

“The IndiaAI ‘mission’ paper looks ambitious, but the reality is a repeat of tech dominance 1.0: US players, with deeper pockets and stronger infrastructure, are poised to win all over again — by using low prices to bury any homegrown alternative, just like last time,” it said. 

A review of what the IndiaAI Mission has achieved and comments from industry experts highlight efforts the country needs to make for the technology. 

The Mission was launched in March 2024 with seven stated pillars to provide access to modern computing facilities, promote impactful AI solutions, establish a framework for safe and trusted AI, and create a national platform with anonymised datasets for training LLMs, among others. 

Of these, the government’s push to boost AI computing power has had the most traction. Bids have come in for procuring nearly 40,000 graphics processing units (GPUs), well above the initial target of 10,000. A GPU is a specialised electronic circuit designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer for output to a display device. 

Bidders like CMS Computers, Ctrls Datacentres, E2E Networks, Ishan Infotech, NxtGen Datacenter and Cloud Technologies, Tata Communications, and Yotta Data Services are ready with GPUs, but they can’t rent them because the usage of these high-end computing machines is yet to take off, according to industry executives and government officials. 

One key reason for the slow uptake of GPUs is the lack of high-end research and development projects needed to enable large usage amount of computing power, said a government official. 

“Take, for example, Sarvam. It has been allotted nearly 4,100 GPUs for about six months, and the government has subsidised the costs. Additionally, three other startups are also developing LLMs. But barring these, which have in total occupied just about 8,000-9,000 GPUs, there are no big clusters of usage,” the official said. 

AIKosh, a dataset platform with contributions from several academics, institutes of higher education, companies, and individual researchers, too has had modest success as a pillar of IndiaAI mission. As of August 31, the platform had more than 2,000 datasets uploaded across 20 sectors. 

Contributors have uploaded 200 AI models trained on various datasets. These include pre-trained transformer models for Hindi, Marathi, Gujarati, Tamil, Malayalam and other Indian languages. 

Other objectives, such as developing future skills for AI, the IndiaAI Innovation Centre, and startup financing, are trailing, said an industry executive. 

“We have four companies developing indigenous LLMs. A few more are likely to be selected in the next cohort. There is, however, no clarity on the applicability of the LLMs being developed by these companies,” the executive said, asking not to be named. 

The government is also likely to award contracts to build LLMs to Tech Mahindra, Fractal, BharatGen, Genloop and four other companies, said another official. Sarvam, Soket, Gan.ai and Gnani.AI were given deals earlier. 

“As of April 30, as many as 506 applications were received. We are approving applications based on the [companies’] proposals and their applicability in the real world instead of the applicant’s company size or their proposed investment,” said the official. 

Similarly, the IndiaAI Application Development Initiative, which aims to support the development, scaling, and promotion of impactful AI solutions that address real-world challenges, is largely dormant following the selection of 30 applications from 900-plus participants. 

“Currently, these projects are undergoing a technical assessment and need external evaluations as well. The problem there is the lack of qualified external evaluators and their time,” said the second official. 

Some industry executives believe that the IndiaAI Mission requires additional funding to support startups that can compete on a global stage, but others argue against this and claim that the country’s only advantage in developing the technology is talent-cost arbitrage. 

“There is an unintended consequence of having very well-funded startups, especially in the AI space. Since the pace of technological development in the country is not at par with other companies across the globe, most of the revised funds may go into hefty salaries, which may turn out to be counterproductive,” said an executive of one of the companies supplying GPUs to the IndiaAI Mission.