While it has famously been reported that OpenAI spends about USD 700,000 daily to keep ChatGPT online, closer to home, India’s indigenous generative AI platform, Hanooman, reportedly requires an investment of about USD 3.3 million monthly for operations.
“Currently, the bill we incur for managing Hanooman is USD 3.3 million per month. Though we don’t need billions, like how ChatGPT does, but it’s still really millions,” Dr Vishnu Vardhan, co-founder and CEO, SML India said during an event in Bengaluru on Friday.
The expenses associated with operating a generative AI platform usually include computational resources, maintenance, data storage, energy consumption, research and development, scalability, and regulatory compliance.
“That’s the reason nobody is willing to bet on this. Indians are been very good in application building and doing wrappers on top of it. So, people are okay with using existing models and build it. But India needs a foundational model,” he added.
In May, Abu Dhabi-based 3AI Holding and Dr Vardhan-led SML India launched Hanooman, with the capability of catering to 98 global languages, including 12 Indian languages.
Initially, Hanooman’s Gen AI platform faced challenges due to the unavailability of necessary GPUs such as H100s, leading Vardhan to pursue H200s instead. The company was communicated of a waiting period of 52 weeks for Nvidia’s GPUs. “We can’t even blame them because India is such a tiny market for generative AI like and GPUs. When people are buying tens of thousands elsewhere, when you order something like in hundreds, nobody’s going to give you,” shared Dr Vardhan at the Matrix Global Summit 2024.
Despite these obstacles, Vardhan procured GPUs from Taiwan to commence model development. Today, Hanooman relies on around 1,600 GPUs for operations and more compute via Azure. Dr Vardhan noted the increasing availability of compute resources, including partnerships with companies like Yotta and the introduction of new GPU cloud services, pointing to a positive trend in overcoming compute limitations in India.
The Hanooman generative AI platform has been rolled out on the web and is available for download on Android. The platform claims to have 15 million current users and aims to reach a user base of 200 million within the span of a year.
Several Indian startups, such as Sarvam and Ola's Krutrim, supported by venture capital firms like Lightspeed Venture Partners, are in the process of creating AI models tailored specifically for the Indian market.
Generative AI has gained big momentum in the country as funding soared from USD 15 million to USD 250 million in the 2022-23 period. Earlier this year, the Indian government also approved an allocation of over Rs 10,300 crore to encourage a public-private partnership-based AI ecosystem in India.