As frontier artificial intelligence becomes increasingly centralized in the hands of a few Western companies, India is betting that innovation doesn’t have to come only from the biggest labs.
The country has launched a hackathon inviting startups, researchers, students, and academic institutions to build affordable, multilingual AI devices that work offline and run on open-source models.
The goal is to create AI tools for classrooms, farms, clinics, and villages where cloud connectivity is unreliable, data privacy is critical, or English-language models fall short.
Bhashini, the Indian government-backed AI language platform; French nonprofit Current AI; and Kalpa Impact, a Mumbai-based social impact consultancy, have partnered to create the initiative. It reflects a broader vision of AI as public infrastructure rather than a proprietary product.
“Cloud-based AI systems have enabled significant advances, but there are many situations where citizens operate in environments with limited connectivity, where privacy considerations are important, or where local language support remains critical,” Amitabh Nag, CEO of Bhashini, told Rest of World in an interview over email.
“By running models locally, we reduce dependence on continuous cloud connectivity, lower recurring compute and transmission costs, improve responsiveness, and provide greater control over data.”
Organizers of the hackathon will shortlist 20 teams, which will receive AI hardware kits, technical support, and mentorship to move beyond prototypes. Qualifying teams will get to pitch their ideas to senior government officials, and winners will deploy their tech inside government departments.
AI as digital public infrastructure
At a time when OpenAI tightly guards its trade secrets, Microsoft runs the world’s largest private corporate hackathon, and Anthropic restricts its ultra-advanced Claude Mythos model over cybersecurity concerns, Bhashini and Current AI are opening up their AI infrastructure to the public to fill a critical innovation gap.
The world’s biggest AI firms make products at lightning speed. But these companies often lack the incentive to create public infrastructure, Ayah Bdeir, CEO of Current AI, told Rest of World.
“With AI treated as a proprietary product, and the majority of these models informed by the English-speaking, internet-connected world, we see AI technology that is neither designed for nor accessible to the majority of the world,” Bdeir said.
High-income countries account for more than 80% of notable AI models, AI startups, venture funding, and data center capacity despite representing just 17% of the global population. As a result, many AI systems are optimized for English-speaking, always-online markets where companies can generate commercial returns.
The hackathon is a small attempt to tip the scales, the organizers said.
Current AI is a public-private partnership that has $400 million in pledges from government and philanthropic sources. It aims to raise a total of $2.5 billion over five years. Bhashini has partnered with 50 ministries, and powers more than 500 government websites. It collects language data across more than 500 districts.
“These kinds of investments matter because they fund the parts of the ecosystem that markets often underinvest in — things like public data sets, support for low-resource languages, safety and safeguards, and community-level training and digital literacy,” Priya Vora, CEO of Digital Impact Alliance, a nonprofit that advises African governments on digital development, told Rest of World.
Can this actually work?
The hackathon may generate ideas, but turning them into products is a much tougher challenge.
Experts question whether enough developers working in India’s many low-resource languages have the time and expertise to participate. Even if they do, promising prototypes would still need long-term funding, engineering talent, government support, and customers.
“Hackathons are an effective way to surface grassroots innovation and identify solutions tailored to local languages, affordability, and accessibility challenges. But breakthrough ideas alone are not enough,” Sagar Vishnoi, director and co-founder of think tank Future Shift Labs, told Rest of World. “The critical enablers are continued research, patient capital, and market access. Without these, promising ideas rarely translate into real-world impact.”
India is hardly alone in trying this approach, Astha Kapoor, co-founder of the Aapti Institute, a Bengaluru-based research and policy think tank, told Rest of World. Chinese companies have released open-weight models; the U.N.-backed Africa AI Hub supports local AI ecosystems; and Masakhane, a pan-African grassroots organization is building data sets for African languages. What remains unclear across all of these efforts, Kapoor said, is whether they can achieve meaningful scale.
“It’s impossible to say at this early stage, without clear demand, business models, or incentives, how any of this will scale,” Kapoor told Rest of World. “Use cases and linguistically diverse data sets are not the only considerations for scale. The realities of India’s land, water, talent, and governance must also be considered for replicability.”
Chrissy Martin Meier, director of products and practices at Digital Impact Alliance, a non-profit, emphasized the need to set up the “unglamorous infrastructure work” — consent frameworks, interoperability standards, continuous language data collection, government coordination — to make hackathons impactful. Without these systems, they remain “siloed, fragile, and ultimately dependent on whoever built them,” Meier said.
West versus the rest?
The organizers insist the hackathon is not an attempt to compete with OpenAI or replace Western frontier models. Instead, they see it as an effort to broaden who gets to build AI — and what problems it is designed to solve.
“Commercial actors do not have to be in opposition to DPI [digital public infrastructure]. They can also adopt its principles in lower-cost and more inclusive deployments,” Jonathan Dolan, senior director of strategic alignment and learning at Dial, told Rest of World.
That is already happening, Kentaro Toyama, a professor at the University of Michigan and the author of Geek Heresy: Rescuing Social Change from the Cult of Technology, told Rest of World. Models like Google’s Gemma and Microsoft’s Phi-4 offer “cheap, open, offline, and decentralized” systems for researchers and nonprofits. And their deep pockets help sustain these initiatives.
“Working products require not only sustained engineering, but ongoing maintenance — and all of that requires an organization, usually a company, that is willing to keep investing in the technology,” said Toyama, who helped set up Microsoft Research India, leading interdisciplinary research on tech and socioeconomic development.
The ecosystem is deeply intertwined at the infrastructure level: The hackathon relies on Nvidia hardware.
Pilots from hackathons “rely on the underlying compute infrastructures, which are built, in the end, by a few U.S.-based tech firms,” Abhineet Nayyar, senior associate for research and policy engagement at IT for Change, Bengaluru-based nonprofit, told Rest of World.