IIT Madras professor Balaraman Ravindran has forged doubt on the federal government’s formidable plan to develop an indigenous giant language mannequin (LLM) below the IndiaAI mission inside six months.
In an unique dialog with AIM, Ravindran indicated that the timeline could also be too brief to construct a top-notch mannequin and doubted whether or not the initiative can really place India on the worldwide AI stage.
“I feel six months is simply too aggressive a timeline for us to essentially construct tremendous succesful fashions. What we’re most likely going to get are proper or respectable fashions; we’re not going to shake the world,” he stated.
This comes within the backdrop of Union minister Ashwini Vaishnaw anticipating India’s LLM to be prepared throughout the subsequent ten months. The federal government allotted ₹2,000 crore within the Union Price range 2025 for the IndiaAI mission and has up to now obtained 67 proposals to develop indigenous AI foundational fashions, together with 22 LLMs, for India’s various linguistic and cultural panorama.
Though startups are nonetheless unsure in regards to the timeline, the Ministry of Electronics and Data Know-how (MeitY) stated it’ll proceed accepting proposals until the fifteenth of every month for the following six months, or till it receives a ample quantity—whichever comes first.
The federal government on Thursday launched AI Kosha and Compute Portal below the IndiaAI Mission. These initiatives are a part of the IndiaAI Mission to offer startups and researchers with entry to datasets and high-performance computing sources for synthetic intelligence improvement.
Taking academia because the reference level, Ravindran, who as soon as taught Perplexity CEO Aravind Srinivas, harassed how brief six months are. “[In that time] one semester occurs—assignments and exams occur. Until the analysis has already been taking place, you’ll be able to’t anticipate an LLM in six months.”
In the meantime, IIT Madras, in partnership with a startup, has additionally despatched a proposal for the IndiaAI mission. “Whoever submits a proposal has to arrange as an organization as a result of six months of delivering one thing requires you to have a startup mindset,” the professor stated.
He opined that the federal government ought to have invested extra in academia. “You can’t actually do cutting-edge automotive analysis in academia alone,” he stated, including that it has come to some extent the place if one needs to do cutting-edge AI analysis, one must associate with the trade.
Ravindran acknowledged that the present plan of the IndiaAI mission is best than the unique plan of shopping for plenty of GPUs instantly from NVIDIA. “Now, at the least, we’ve these cloud distributors who can hopefully present extra subsidised issues,” he stated.
The IndiaAI mission will characteristic 18,000+ GPUs by means of public-private partnerships with gamers like Jio Platforms, NxtGen Information Middle, Locuz Enterprise, E2E Networks, CtrlS DataCenters, CMS Computer systems, Orient Applied sciences, Tata Communications, Vensysco, and Yotta Information Providers.
AIM discovered that AI startups equivalent to Sarvam AI, Krutrim, CoRover.ai, TurboML, LossFunk and IIIT Hyderabad had additionally utilized below the proposal.
Notably, Soket AI lately launched Challenge EKA, which seeks to unite AI researchers, engineers, and establishments throughout the nation to develop multilingual, high-efficiency AI fashions that cater to India’s wants whereas competing globally.
Equally, TurboML, a San Francisco-based AI firm, has launched a brand new initiative to carry collectively Indian-origin AI researchers from around the globe. The purpose is to develop basis fashions to help the nation’s future in AI.
“We’re collaborating with the very best minds to construct basis fashions that can energy India’s AI future”, wrote TurboML CEO Siddharth Bhatia on X. In his put up, he talked about discussing the initiative in a gathering with IT minister Ashwini Vaishnaw.
Bhatia informed AIM that his firm plans to construct an LLM below $12 million.
Indian SaaS firm Zoho has additionally entered the race. In a latest interview, group chief govt officer Shailesh Davey stated the corporate plans to launch two foundational fashions by the top of the 12 months. Zoho can also be creating foundational fashions for Indian languages however has not set a timeline on this but.
Extra Funding and Compute Wanted
Ravindran stated that one other benefit of the mission is that it’s going to create a pool of educated engineers who can work with extra researchers sooner or later and develop higher options. Nevertheless, he’s nonetheless not satisfied that India can construct a DeepSeek-like mannequin in six months however is hopeful that one thing fruitful can come within the subsequent 12 months and a half. “In the event you construct DeepSeek in a 12 months and a half, it needs to be one thing very, very totally different,” he stated.
The professor identified that India has a dearth of companies investing cash in elementary analysis. “They’re joyful to do utilized analysis,” he stated.
He defined that China’s DeepSeek didn’t come out of nowhere and that analysis had been happening for an extended interval. “So far as China is worried, there have been at the least 5-6 corporations, possibly 10, that have been on the degree of DeepSeek, they usually may simply launch anytime,” Ravindran stated.
He even instructed that the timing of DeepSeek’s launch appeared strategic. “They stayed silent till the US introduced the Stargate initiative.”
This, he stated, was a transparent response to the developments within the US. “Stargate stated it wanted $500 billion, so the Chinese language countered with, ‘We are able to do it in $4 or $6 billion’,” he stated, suggesting a political angle to the entire affair.
“It’s not like DeepSeek V3 simply got here out of the blue. In the event you have a look at plenty of issues that they utilized in DeepSeek V3, lots of the parts are there in DeepSeek V2. They added RL fine-tuning as an alternative of supervised fine-tuning in the long run,” Ravindran stated.
He added that American corporations have been attempting to create a moat round themselves by saying that you simply want some huge cash to get to the identical degree as us, and that’s maybe one of many causes they didn’t wish to discover cheaper alternate options.
Ravindran stated that when discussing India needing extra compute, he means reaching a sustainable degree, not an ‘Elon Musk-level’ of compute. Notably, Musk’s xAI educated Grok 3 on 200,000 GPUs. The IIT professor believes India must look past Transformer-based autoregressive coaching and even reinforcement learning-based fine-tuning. “I feel they’ll quickly be hitting a ceiling. Individuals who labored on symbolic AI have been manner forward of their time in how they considered intelligence,” he concluded.
The put up IIT Madras Prof Balaraman Ravindran Questions IndiaAI Mission’s Timeline to Construct an Indigenous LLM appeared first on Analytics India Journal.
