← All episodes · Y Combinator
India Can Create The Largest AI Companies
Overview
The closing panel of YC's Startup School event in India, hosted by YC partners (including Jared and Ankit) with two guests who helped YC expand into India: Puneet, who founded SuperDaily (YC W17, scaled to ~$100M annual revenue with essentially one engineer besides himself, exited to Swiggy) and later invested at Nexus, and Arnav, a former YC staffer now at Peak XV.
The through-line is a thesis: for the first time, global — not just local — companies can be built from India, because the AI wave rewards being at the edge of the technology rather than mastering local go-to-market. The panel distills the day's founder talks into tangible advice: sell into the US without warm intros (a third-year IIT student cold-emailed US insurance companies and won), ignore non-AI-native career advice, tinker on 'projects' until startup ideas surface, exploit second-mover advantage with superior product velocity, and spend aggressively on tokens to feel the frontier. It ends with what YC actually looks for in applications — clarity, taste, agency — and announcements about compute credits for attendees.
Topics
Why this wave is India's to win
Puneet frames the last decade of Indian startups as the mobile wave: mobile phones 'tokenized labor,' letting anyone sell an hour of their time to Swiggy, SuperDaily, or Zepto — but those network effects were hyper-local, producing strong local companies in India, LatAm, and the US separately. The AI wave is different in kind: it is global, and it rewards understanding the technology 10x better than everyone else rather than owning a local go-to-market motion. With India's technical talent depth, he predicts some of the largest companies in the world will come out of India, echoing Mukund of Emergent's 'look at the ceiling' framing.
On the intimidation of selling into an unfamiliar US market: the old SaaS playbook required warm intros or years living in SF. Now everyone worldwide simultaneously understands AI matters, so buyers are open to meritocratic, outcome-driven solutions from anywhere. His proof points: Giga and Emergent's founders never lived in SF before founding; a recent YC batch company run by a third-year IIT student sold to US insurance companies via cold email. And YC itself remains 'the great conduit' — going through the batch 'raises your ambition 10x.'
Escaping the Indian education system's safe path
Arnav's argument to the students in the room: the future of India's AI ecosystem will be defined by them, not the prior generation, and traditional advice — banker, consultant, engineer, doctor — points at jobs that may not exist in ten years. The historically safe path may now be the risky one, while owners and entrepreneurs are the most insulated. He caveats honestly that an average young Indian can't take Silicon Valley-grade risks given thinner social safety nets, and that landing a prestigious job from a humble background is already a top-10% outcome worth celebrating.
His mechanism for developing an independent point of view is environmental: surround yourself with people at the cutting edge. YC's magic is people network effects — founders leave the batch as 'the 007 versions of themselves' (PG's line). It's not cool to be ambitious in most education systems; in a YC-like environment ambition is expected. In AI specifically, following advice from non-AI-native people is dangerous 'because they're just not in the game with you.'
Younger founders, tinkering, and the definition of a project
YC's batches are skewing younger — not by policy, but because AI leveled the playing field: you're no longer limited by ability to build, only by pace of learning, which favors the young. The best young founders tinker: they follow curiosities and work on what's barely possible for today's models ('live in the future and find what's missing'), which surfaces bottlenecks that are themselves startup ideas.
A recurring empirical pattern from the day's talks: almost no founder's big idea was their first idea. Good startup ideas aren't visible from a whiteboard; they're noticed while building. With coding agents, a college student messing around between classes can get projects off the ground fast enough to trip over multiple good ideas. Ankit offers a precise, memorable definition: a project is when two people build something that was not assigned to them and get someone to use it — and notes you can complete an entire CS education and career without ever doing one. Do projects, especially in college, and startup ideas are 'guaranteed.'
Second-mover advantage in the agent era
The panel highlights how many of the day's success stories were not first movers — third, fourth, or forty-second to their space. Varun from Giga closed DoorDash as a contract while competing against companies with several hundred employees; Giga had eight. The mechanism: coding agents let a team with product clarity make ideas real extremely fast and with high precision, so unless the incumbent has genuine network effects (few do), a better product simply wins. The 'AI slop' criticism gets dismissed as a dated worldview — slop means you're using the tools badly; skilled users produce sophisticated code faster than ever. Emergent is cited as a company scaling rapidly on exactly this play: find something kind of working, do it better, beat them.
Token-maxing: what the frontier actually feels like
Ankit describes his four-month rabbit hole (shared with Gary Tan, whose token bills run thousands of dollars per day): he didn't realize how good things had gotten until Opus 4.5 over Christmas break pushed him to pay for the $200/month max plan — and his key realization is that below that level of usage 'you are not anywhere close to the frontier.' Letting tokens rip means writing 10,000 unit tests instead of 20, exhaustive docs, every corner case — yielding incredibly effective code fast. His email-client side project, built for fun, generated several startup ideas purely from pushing models to their limits; e.g., Gmail's auto-reply is bad, but at $5 of inference per email it's really good. The credits given to attendees exist precisely so they can feel what not being capital-constrained is like; he asks whoever ships a product on those credits fastest to email him.
For those who can't afford it, the panel points to open-source models (minimax and others are 'pretty good and really cheap'), the YC company OpenCode built atop them, and Vidit of Meesho's plan to bring the next billion people online with voice AI — which requires price points only open models can hit. Arnav's advice: if budget-constrained, go work at companies that give employees unlimited token budgets, and build for where models will be in 6-12 months.
What YC actually looks for
Clearing up application misconceptions: clarity beats impressiveness — if partners can't understand what you're building, nothing else matters. YC isn't investing in ideas (most successful founders pivoted, some after reaching millions of ARR); it's investing in founders, evaluated on taste and agency. Taste isn't visual polish but intention: design choices backed by customer insights and the pace of reaching those insights. Agency is PG's 'relentlessly resourceful' (an essay everyone was told to read): do you let the world's conditions happen to you, or do you exact your will on the world?
Jared, ten years at YC, says surprisingly little has changed — the fundamentals of great founders go back to Thomas Edison, who would have sounded exactly like today's speakers: obsessed with customers, tinkering, and building at the technological edge. What has changed is leverage: AI democratized building, which is why YC can fund 18-year-olds like the Giga or Zepto founders who then build epic companies at speed — shipping 'more like tonight, not tomorrow.' All six presenting companies are actively hiring engineers, and working at an exceptional startup is pitched as the best founder training.
Visuals
What the wave rewards × Where network effects live
Mobile wave (2010s)
- Labor tokenized by phones
- Hyper-local network effects
- Local GTM mastery wins
- Swiggy / Zepto / SuperDaily era
AI wave (now)
- Global from day one
- No local moat
- Living at the technology's edge wins
- India's technical talent advantage
Old US-entry playbook
- Warm intros required
- Years living in SF
- Budget-relationship sales
New US-entry playbook
- Cold email works
- Meritocratic, outcome-driven buyers
- IIT student sold to US insurers
Takeaways
- Build global from day one: the AI wave has no hyper-local network effects, so the old requirement of US networks and warm intros is gone — cold outreach with a superior product works.
- Don't take AI-era advice from non-AI-native people; deliberately surround yourself with people at the cutting edge to develop an independent point of view.
- Do 'projects' — something two people build that wasn't assigned and that someone actually uses. That single habit reliably surfaces startup ideas.
- Being second (or forty-second) is fine: without network effects, superior product velocity from coding agents beats incumbency.
- Spend enough on tokens to actually touch the frontier (max-plan level or above); build for where models will be in 6-12 months, not where they are today.
- In a YC application, clarity beats cleverness; the durable founder signals are taste (intention backed by customer insight) and agency (relentless resourcefulness).
Notable quotes
This wave is more about are you living at the edge of the technology and not as much about do you understand the right go-to-market... and I think nobody does that better than in India.— Puneet
That safe path might actually now be the risky path.— Arnav
You are no longer limited by your ability to build, you're limited by the pace with which you can learn.— John (YC partner)
A project is when two people build something that was not assigned to them and get someone to use it.— Ankit (YC partner)
Unless you are paying for at least that level of usage, you actually are not anywhere close to the frontier right now.— Ankit (YC partner)
More like tonight, not tomorrow.— Panel (on how fast you can now ship)