Last week, electronics and information technology minister Ashwini Vaishnaw announced at a press briefing that 10,000 GPUs at a common compute facility will soon go online. The facility will be available to startups, academia, and developers through an online portal. The Centre will review requests for access to the GPUs based on requirements.
GPUs are a critical electronic circuit that help speed up processing of images and videos in computing.
“Building a foundational model has to be government supported, because it’s really about capacity building and lifting the ecosystem,” Chopra told Mint.
The entrepreneur’s new company, Lossfunk, comes at a time when discussion about foundational AI models has reached fever pitch. China’s DeepSeek recently revealed its own open-source model, R1, which is quite cost-effective, posing a challenge to other top AI models like OpenAI’s ChatGPT.
However, privacy experts have suggested that DeepSeek may not be as secure as other models, with most pointing to potential data leaks to the Chinese government. DeepSeek’s privacy policy states that all user data is stored in China. In fact, Italy has instituted a complete ban on the company’s app across the Google Play and Apple App stores in the country.
Also read | DeepSeek: A Chinese marvel or OpenAI copy?
Bootstrapping a new AI venture
For now, the 37-year-old Chopra plans to use the $200 million he made off the sale of Wingify to Everstone to bootstrap his new project. His latest venture is all about exploring the most interesting “vectors”.
“Innovation is fundamentally unpredictable. You can’t say I want to do that and then backtrack. It needs to be more forward-looking. We are working on the most interesting thing. I’m hypothesising right now, and I expect something interesting to come out of it,” Chopra said. Broadly, he’s working on three main themes for his foundational model: creativity, reasoning and efficiency.
The Wingify founder says that while sitting on a not-insignificant chunk of capital, he doesn’t want to be reckless with it; instead, he is very strategic with allocating his money. “Private money like my own would go a long way in doing very pin-pointed, breakthrough kind of research which can happen at academic lab budgets.”
Also read | From bedroom to boardroom: How SaaS startup Wingify bootstrapped its way to success
With his new venture, Chopra doesn’t just want to build a foundational model but also change the way AI is being built in India. While his long-term goal remains to build a creative and efficient superintelligence, his short-term aim is to develop a culture of state-of-the-art work around AI in India.
“The initial road map is about catching up to what is state-of-the-art, and then start breaking it bit by bit, and then just go beyond,” the Lossfunk founder said.
Govt involvement in building AI for India
The government has reiterated its commitment to building AI in India, primarily through the FY25-26 Budget, which allocated ₹2,000 crore to the India AI Mission. However, its policy decisions have often been at odds with what it’s trying to achieve.
Last year, the government set up an expert committee to help develop legislation for AI. Later that year, however, the government changed its stance, opting instead to set up an institute that would provide recommendations.
Sudden changes in government policy or thinking regarding AI could adversely impact startups working on AI. “I have sympathy with knee-jerk reactions because the thing is that nobody really knows what to do,” says Chopra.
Even if policy changes do happen, it’s not a completely foreign concept for startups. Some may move jurisdiction, while in a worst-case scenario, others may just shut down. “It is a no-brainer that if you don’t have a consistent and clear policy, industry and startups are net losers,” says Ai&Beyond co-founder and CEO Jaspreet Bindra.
Also read | Minimalist and Wingify: deals with a difference
Enacting legislation is a challenging task in itself. However, leaning too heavily on legislation can kill innovation, as has happened in Europe because of the EU AI Act.
Bindra proposes a balance between legislation and recommendation where some issues, like deepfakes (AI-generated images/ videos of real or non-existent people), need binding rules, while development could have guidelines.
But is the government’s decision not to lean on legislation right? Primus Partners vice president Raizada Munish Vaid suggests that the government is on the right track. “They are trying to be adaptive and flexible. The government doesn’t want to miss out on bigger products that can come out of the Indian startup ecosystem.”
Also read | DeepSeek vs Stargate Project: How AI spending is evolving, in charts