Vijay Shekhar Sharma offers personal funding for Indian AI model builders
Paytm founder and CEO Vijay Shekhar Sharma said he would write personal cheques of Rs 1 crore to Rs 2 crore to entrepreneurs building their own artificial intelligence models in India. Speaking at the Cypher 2026 AI conference in Bengaluru, he said Paytm, being a listed company, cannot commit corporate funds, but offered founders access to its distribution network. Sharma put the cost of building an AI model at under USD 100 million and said capital is not India's constraint.
Source
Technology · read the original report ↗
Desk check · compared with the source
What the desk checked (5)
- Sharma will write personal cheques of Rs 1-2 crore to founders building Indian AI models. — Directly attributed to Sharma at Cypher 2026 in the source, with quotes.
- Building an AI model costs under USD 100 million. — Sharma's own estimate as quoted in source; no independent costing given.
- Paytm optimised a 200-billion-parameter model into a 4-billion-parameter model for Indian languages. — Source says Sharma has previously stated this; figure appears in source, attributed to him.
- Paytm cannot commit company funds because it is a listed company. — Attributed to Sharma; stated as his reasoning in the source.
- Paytm's Q1 FY27 earnings release mentions function-specific models and agents from fine-tuned open-source models. — Attributed to a company earnings release named in the source.
Analysts’ view opinion
This is less an announcement about money than one about direction. Cheques of Rs 1–2 crore cannot fund a foundation model — but they work as a seed signal, with one of India's best-known founders publicly picking a side in the debate over whether Indian startups should build apps on others' models or build models of their own. The more valuable part of the offer may not be the cash at all, but access to Paytm's distribution network, since getting an AI product in front of real customers is often harder than training it.
- Rs 1–2 crore realistically funds experiments, small teams and fine-tuning work, which fits Sharma's own framing of it as a starting cheque against a model-building cost he put at under USD 100 million.
- The point that Paytm, as a listed company, cannot commit corporate funds matters: this is personal capital, not a corporate venture fund.
- Paytm's claim of taking a 200-billion-parameter model down to 4 billion parameters for Indian languages and running it on its own machines mirrors the current industry shift toward optimising and fine-tuning open-source models rather than training from scratch.
- Turning internal risk and fraud engines into enterprise services under Paytm Intelligence (Pi) fits a broader trend of fintechs repositioning themselves as AI infrastructure vendors.
- Sharma's stance in the sovereign-AI debate is hybrid rather than isolationist — keep using foreign models while building several domestic ones — which aligns with how many countries are currently thinking about this.
What to watch — Watch whether this converts from conference rhetoric into actual cheques, concrete Paytm distribution tie-ups and teams genuinely attempting model-building over the coming months.
The story does not establish how many founders will be funded, on what timeline, or under what terms, and no actual deal is reported as closed.
Deep dive
Research brief · 8 facts · 4 dates · exam-readyThe brief
Context
India's AI startup ecosystem has been debating whether to build foundational AI models of its own or build applications on top of models developed abroad. Speaking at Cypher 2026, an AI conference in Bengaluru, Paytm founder and CEO Vijay Shekhar Sharma turned that debate into a personal funding offer, saying he would write cheques of Rs 1 crore to Rs 2 crore from his own pocket to founders building Indian AI models. Sharma, who popularised QR-code payments and the Soundbox at Indian shop counters, argued India should own core AI capability rather than only build on foreign models. He also said capital is not the constraint, pegging the cost of building an AI model at under USD 100 million.
Key facts
- Vijay Shekhar Sharma offered personal cheques of Rs 1 crore to Rs 2 crore to Indian founders building their own AI models, speaking at Cypher 2026 in Bengaluru.
- Sharma said Paytm, being a listed company, cannot commit corporate funds to such bets, but offered founders access to Paytm's distribution network.
- He put the cost of building an AI model at under USD 100 million, saying the bigger test is long-term commitment, not money.
- Sharma said: 'There is enough capital, enough resources, and enough intention to support that.'
- At Paytm's 26th Annual General Meeting in September, Sharma called AI a 'native capability' and said, 'We don't need to buy AI from outside.'
- Sharma has said Paytm took a 200 billion parameter model and optimised it into a 4 billion parameter model for Indian languages, run on its own machines.
- Paytm's Q1 FY27 earnings release says the company has developed 'function specific models and agents' by fine tuning open source models.
- Paytm's AI-based services are being tested internally and piloted with some enterprise customers under Paytm Intelligence, or Pi.
Timeline
- About a decade agoSharma bet that India's shopkeepers would accept payments on a phone, taking mobile payments to shop counters via QR codes and the Soundbox.
- September (Paytm's 26th AGM)Sharma described AI as a 'native capability' for Paytm and said the company does not need to buy AI from outside; AI services piloted with some enterprise customers under Paytm Intelligence (Pi).
- Q1 FY27 earnings releasePaytm disclosed it had developed 'function specific models and agents' by fine tuning open source models.
- This week, at Cypher 2026, BengaluruSharma offered personal cheques of Rs 1 crore to Rs 2 crore to Indian AI model builders and said capital is not India's constraint.
Who has a stake
- Vijay Shekhar Sharma, Paytm founder and CEO — Committing personal capital and reputation to the idea that India should build its own AI models, not only applications on foreign ones.
- Indian AI founders and startups — Access to starting cheques of Rs 1 crore to Rs 2 crore plus Paytm's distribution reach to put products before customers.
- Paytm (listed company) — Cannot commit corporate funds to such bets; building risk engines, fraud systems and fine-tuned models internally and commercialising them via Paytm Intelligence (Pi).
- Paytm's enterprise customers — Early pilots of AI services built on Paytm's internal models under Pi.
- India's AI ecosystem at large — The build-your-own-model versus build-on-top debate shapes whether core AI capability sits inside or outside the country.
Why it matters
Much of India's AI activity today sits in the application layer, while core model-building capability largely resides abroad; Sharma frames this as a question of ownership rather than isolation, arguing India can use foreign models while developing several of its own. His claim that an AI model can be built for under USD 100 million, and that capital and intent already exist, shifts the debate from funding scarcity to founder ambition. It also positions AI as a source of national economic strength for a population of more than a billion, where higher productivity could outweigh job-loss fears.
UPSC angle
Prelims pointers
- Cypher 2026: AI conference held in Bengaluru where Sharma made the funding offer.
- Vijay Shekhar Sharma is the founder and CEO of Paytm; associated with QR-code payments and the Soundbox.
- Paytm Intelligence (Pi): Paytm's AI services unit, piloted with some enterprise customers.
- Sharma pegged the cost of building an AI model at under USD 100 million.
- Paytm reportedly optimised a 200 billion parameter model into a 4 billion parameter model for Indian languages.
- Paytm's Q1 FY27 release cites 'function specific models and agents' built by fine tuning open source models.
Mains framing
India's AI debate has hinged on whether to build foundational models domestically or layer applications over models developed elsewhere; Vijay Shekhar Sharma's offer of personal cheques of Rs 1 crore to Rs 2 crore at Cypher 2026 reframes the constraint from capital to conviction, since he argues there is 'enough capital, enough resources, and enough intention' and that a model can be built for under USD 100 million. The structural problem he identifies is capability location: Indian firms excel at applications while core model capability sits outside the country, which he casts as an ownership question rather than a call for isolation, India can keep using foreign models while developing several of its own. Paytm's own path illustrates a pragmatic route, internally built risk and fraud engines, open-source models fine tuned into 'function specific models and agents', and a 200 billion parameter model compressed to 4 billion parameters for Indian languages running on its own machines, now being commercialised through Paytm Intelligence (Pi). The limits are equally instructive: as a listed company Paytm cannot commit corporate funds, so the backing is personal and the cheque is only a 'starting' one, with distribution access as the real differentiator. The way forward implied by the source is long-horizon commitment by founders, leveraging distribution networks to reach customers, and treating AI as a productivity multiplier across a billion-plus population rather than a job threat.
Key terms
- Cypher 2026
- AI conference held in Bengaluru where Sharma announced his personal funding offer to Indian AI model builders.
- Paytm Intelligence (Pi)
- Paytm's AI offering under which services built on its internal AI are tested in-house and piloted with some enterprise customers.
- Fine tuning open source models
- Adapting freely available AI models to specific tasks; Paytm says it built 'function specific models and agents' this way.
- Parameter (model size)
- A measure of an AI model's scale; Paytm reportedly compressed a 200 billion parameter model into a 4 billion parameter one for Indian languages.
- Soundbox
- Paytm device that audibly confirms payments at shop counters, part of Sharma's earlier push for mobile payments in India.
- Listed company constraint
- As Paytm is publicly listed, Sharma said he cannot commit its funds to speculative AI bets, hence personal cheques.
Practice questions
- India has built strength in AI applications while core model capability largely sits abroad. Examine the implications of this gap for technological sovereignty and suggest a way forward.
- 'Capital is not India's constraint in AI; commitment is.' Critically evaluate this claim in light of recent private funding offers to Indian AI model builders.
- Discuss how private sector distribution networks, such as those built by payments firms, can accelerate adoption of indigenous AI products in India.
Grounded only in the source report — figures and dates are the source's, not inferred.