IPO-Bound Fractal Proposes To Build India’s First Reasoning AI Model

Riding on the AI wave, IPO-bound Fractal Analytics

is looking to build India’s first large reasoning model (LRM) with a total investment of INR 118 Cr, its chief AI Research and Platforms officer Suraj Amonkar told Inc42.

While the Mumbai-based AI startup will infuse INR 42 Cr in the project from its own coffers, it is seeking a capital subsidy of INR 76 Cr from the Centre under the IndiaAI Mission, Amonkar said.

Fractal will deploy a substantial portion of the capital towards procurement of graphic processing chips (GPUs), high-end chips that go into development of large language models (LLMs). 

“There is a provision of about INR 94 Cr to be spent on the GPU compute only,” the Fractal executive said. The company anticipates that 700K hours of GPU would be needed to train its upcoming LRM models over 9 months.

The development was first reported by the Economic Times, which said that Fractal has submitted a proposal to the Centre to build a reasoning-focussed AI model. It plans to release it in three versions – small, medium and large. These models would be trained using datasets ranging from 2 Bn to 1 Tn tokens.

Fractal To Join Startup IPO Wave: Fractal is among the growing list of Indian startups looking to make public debut as it seeks to capitalise on India’s booming economy, access to a larger pool of investors and better initial public offering (IPO) aspects in the country.

Reports surfaced last year that Fractal was looking to raise $500 Mn via its IPO at a likely valuation of $3.5 Bn with an eye on a public listing by March 2025. However, the company’s IPO plans haven’t yet materialised.

Thirteen new-age tech companies, including Swiggy and Ola Electric, went public last year and raised over $29,000 Cr via their public issues. The startup IPO wave is to continue this year, with more than 20 new-economy ventures including the likes of Physics Wallah, Ola Consumer, BlueStone, boAt, Ather Energy, among others, eyeing a Dalal Street debut in 2025.

How Reasoning Models Are Different From Conversational AI: Reasoning models have been the talk of the town since OpenAI released a model called o1 aka Strawberry, in September last year. Unlike conversational AI models such as ChatGPT, Bard and Bing Chat, which are trained to mimic patterns from training data, LRMs are touted to have reasoning and analytical capabilities.

For instance, OpenAI trained its o1 LRM model to solve complex problems on their own using reinforcement learning, a technique that teaches the system using rewards and penalties. It then uses “chain of thought” to process queries in a step-by-step manner.

OpenAI isn’t the only company working on reasoning-focussed AI models. China-based DeepSeek’s v3 and R1 reasoning models have caused a lot of buzz and upended the idea about AI costs. While OpenAI’s ChatGPT and o1 models cost more than $100 Mn, training DeepSeek R1 cost only $5.6 Mn.

The advances from DeepSeek models show that Indian startups such as Bhavish Aggarwal-led Ola Krutrim, Sarvam AI and Pixis, can compete with bigger competitors like Meta and Google.

What India Is Doing To Catch Up With China, the US As AI Race Heats Up: The demand for AI is on the rise in India. AI adoption was particularly prominent in the pharma and healthcare industry at 52% in FY24, followed by the FMCG sector at 43% and manufacturing at 28%, as per industry estimates.

While Indian heavyweights TCS, Wipro and Infosys have been less than bold in embracing AI, homegrown startups are increasingly integrating AI, particularly generative AI (GenAI), into their products and services, as per Inc42’s ‘The Rise Of India’s GenAI Brigade’ report. 

Amid growing demand for AI services, investor interest in the sector is gaining significant momentum. India’s GenAI startup brigade has raised $1.2 Bn in funding since 2014.

Behind the AI boom is also the Centre’s policy shift. In March last year, the union cabinet approved the IndiaAI Mission with an outlay of INR 10,372 Cr spread over the next five years. As part of the initiative, the Centre aims to offer incentives and subsidies to private companies to scale up India’s AI compute capacity.

The IndiaAI Mission has also invited Indian startups, researchers, and entrepreneurs to collaborate on developing foundational AI models trained on indigenous datasets. So far, it has received 187 applications for this initiative.