[Exclusive] Two LLMs will be launched by February: IndiaAI M…

In March last year, the launch of the IndiaAI Mission drew a lot of attention at the global stage. Aimed to create a robust AI ecosystem in the country by building sovereign large language models (LLMs), the mission has had its share of hits and misses. Twelve companies, including Sarvam AI, BharatGen, and Fractal Analytics, were selected to develop foundational models tailored to India’s linguistic and regional diversity, with the government backing them with grants, subsidised GPU credits, and other forms of support.  

As the mission and the government gear up for the India AI Impact Summit scheduled for February 2026, IndiaAI Mission CEO Abhishek Singh projected confidence in the country’s AI ecosystem.

“The mission has progressed most desirably, and the team is now gearing up for product launches and inference at scale,” he told AIM in an exclusive conversation.

Addressing the mission’s earlier plan to launch the first LLM by November this year, Singh said, “Making the compute infrastructure operational itself took six months, and only after that could we start training the models. I don’t think there is any cause for concern regarding models coming up. Both Sarvam AI and BharatGen will launch their models before the India AI Impact Summit.”

He also hinted at a voice-based LLM being developed for cybercrime, calling it the first such application to be built in India, though he did not share further details.

Expressing overall satisfaction with the mission’s progress, Singh said he would have preferred to move faster on setting up a fund of funds for startups. In the Union Budget 2025-26, the government had allocated an outlay of ₹2,000 crore out of the planned ₹10,372-crore budget over five years, with plans to double that to ₹20,000 crore.

Statutory Licensing Framework

Welcoming the Department for Promotion of Industry and Internal Trade’s proposed statutory licensing framework for AI training, Singh said such a framework would strike a balance between the need for AI models to access copyrighted data for training and the requirement of mandatory licensing.

“Ideally, revenues should be shared with those who hold copyrights and royalties, but the real challenge will be estimating who gets what. It will definitely face implementation challenges,” he warned.

He clarified that since the proposal is still in the draft stage, further refinements are likely.

On experts repeatedly mooting for the need for a data protection law, he said, “We recently published our AI governance guidelines, and I believe the current legal provisions are sufficient to handle the risks AI poses. AI is just another technology, and it’s never wise to regulate technology itself,” Singh said.

The seven core ethical principles—or “sutras”—guiding India’s AI strategy focus on trust, people-first design, innovation over restraint, fairness and equity, accountability, explainability by design, and safety and resilience. To support implementation, the government is developing tools for bias mitigation, ethical AI certification, stress testing, and machine learning audits.

“The right way to limit potential harm caused by AI is to regulate use cases and applications. For deepfakes, there are already provisions under the Bharatiya Nyaya Sanhita and the IT Act,” he added.

Facing Challenges

One of the key challenges the mission faced was GPU procurement. “Setting up the GPUs required private investment of nearly ₹20,000 crore. That kind of investment takes time to materialise,” Singh noted. High-end GPUs also require data centres with advanced cooling capabilities, which typically take 12–18 months to build.

On the talent front, Singh pointed to a growing skills crunch. “There are very few people qualified to train large AI models, and most are absorbed by Big Tech, offering exorbitant salaries. Many startups under the IndiaAI Mission struggle to find the right talent, and retaining skilled professionals remains a major challenge,” he said.

Working on Actuals

Responding to criticism that the mission has not yet shown a tangible impact due to limited GPU usage in the absence of large-scale research projects, Singh said it was too early to conclude.

“In a country like India, even if 100 million people start using these services, we will need hundreds of thousands of GPUs. Once the models are launched, the next phase will be large-scale inferencing,” he said.

As of August, the IndiaAI Mission invested in over 38,000 GPUs.

He also dismissed comparisons that portray India’s $1.2-billion AI investment as insignificant compared to the US’s $500-billion Stargate initiative or China’s $137 billion spend. “Stargate is backed by OpenAI, SoftBank, Oracle, and Abu Dhabi’s MGX. In China, it’s difficult to distinguish between public and private funding. India’s $1 billion, by contrast, is purely government funding and cannot be compared directly,” he said.

Singh acknowledged that China spends heavily on R&D—around 3% of GDP, compared to India’s 0.7%—while also pointing out the vast difference in economic scale.

India’s AI Needs

Acknowledging the AI hype, Singh said the IndiaAI Mission is not chasing that narrative. “We are focused on building use cases and models that meet India’s requirements. For that, we don’t need extremely large models. A 70-billion-parameter model trained on Indian data can deliver strong outcomes,” he said.

The government, he added, is following a use-case-led roadmap to guide how AI is deployed and for what purpose.

“Whenever a new technology emerges, questions about survival and sustainability arise. But the speed at which AI has been adopted shows it is here to stay. Companies that build products with real value will survive,” Singh said.

Making Money

Though confident that AI companies will eventually turn profitable, Singh predicted that monetisation will take time. “This is true for almost every new technology. Initially, companies focus on user acquisition by offering free or low-cost services and burning capital. Over time, products that create real value survive, and people are willing to pay for them just as Microsoft Office evolved into a paid subscription,” he said.

India’s approach to building AI—particularly voice-based applications—will be distinct from the rest of the world, shaped by the country’s scale and diversity.

“India has the opportunity to show the world, especially the Global South, how AI can move beyond hype, achieve mass adoption, and generate real revenue, much like what we achieved with digital public infrastructure,” Singh said, signing off.

The post [Exclusive] Two LLMs will be launched by February: IndiaAI Mission CEO  appeared first on Analytics India Magazine.

Generated by RSStT. The copyright belongs to the original author.

Source