India Can Create The Largest AI Companies

ELI5 / TLDR

A closing panel at a Y Combinator event in India. The argument: the last Indian startup wave (Swiggy, Zepto, Meesho) was about hyper-local delivery — copy a US model, win your own geography. AI is different. It’s global from day one, and it rewards whoever understands the technology best rather than whoever has the best US Rolodex. India has the deepest technical talent, so for the first time the panel thinks genuinely global, very large companies can be built from India. The supporting advice: tinker, build projects nobody assigned you, spend more on AI compute than feels reasonable, and stop waiting for permission.

The Full Story

The format is a wrap-up panel after six founders spoke. The panelists are YC’s own partners plus two friends of the firm: Puneet (built grocery-delivery company SuperDaily to $100M revenue with, famously, one engineer besides himself, then sold to Swiggy; later a VC at Nexus) and Arnav (ex-YC, now at Peak XV). The crowd is young — the host mentions meeting fifteen-year-olds.

Why this wave is different from the last one

Puneet’s core claim is the spine of the whole talk. The mobile wave “tokenized labor” — anyone with a phone and a free hour could deliver for Swiggy or Zepto. That created hyper-local network effects, which meant a separate set of winners in every region: India, Latin America, the US. The opportunity was local, so the companies stayed local.

“The AI revolution is not local… I think the AI change is global and this is our time to shine.”

And the thing that wins is different. The old game needed go-to-market savvy, the right business model, a warm introduction. The new game, in his framing, is narrower and more technical:

“This is about do you understand this technology 10x better than everyone else and I think nobody does that better than in India.”

Selling globally without ever leaving India

The obvious objection: a founder in India who’s never been to the US, doesn’t know the market, has no network. Puneet’s answer is that AI flattened the barrier because everyone everywhere woke up to AI at the same moment, so buyers are unusually open. His proof point is almost absurd — a third-year IIT student cold-emailed US insurance companies (a notoriously hard sell) and closed them. The takeaway: drop the assumption that you need a warm intro. A genuinely better product gets a hearing on merit. He does, predictably, route this through “apply to YC” as the conduit to going global.

The Indian education system is now the risky path

Arnav’s section is the most pointed. The traditional advice — become a banker, consultant, engineer, doctor — points at safe, prestigious, high-paying jobs. His argument is that many of those jobs may not exist or will change drastically in ten years, so “that safe path might actually now be the risky path.” The people most insulated, he thinks, are owners and entrepreneurs.

He’s honest about the asymmetry, which is the most credible part of the talk. An average young Indian cannot take the risks a Silicon Valley kid can — the social safety net isn’t there. If you come from a humble background and land a stable, prestigious job, you’re already in the country’s top 10% and that’s a real win. He’s not dismissing it. He’s speaking to the subset who already feel the itch and want permission to ignore the cookie-cutter advice.

The mechanism for developing an independent point of view is, in his telling, mostly about who you stand next to:

“It’s very dangerous to follow the advice of people who are not AI native because they’re just not in the game with you.”

In an education system where being openly ambitious isn’t cool, surrounding yourself with people who expect ambition is the unlock.

Tinkering, pivots, and second-mover advantage

The YC partners add the builder-mechanics. Founders are getting younger not by design but because AI changed the binding constraint: you’re no longer limited by your ability to build, you’re limited by how fast you can learn — and young people learn fast. The best young founders “tinker,” follow curiosity, and work on things that are “barely good enough for the models to do,” because that’s exactly where the bottlenecks (and the good ideas) hide.

A recurring observation from the day: almost every founder’s winning idea was not their first idea, and many weren’t first movers in their category — they were the third or forty-second entrant who simply built better. With coding agents, anyone with product clarity can make ideas real fast, which the panel argues creates a structural second-mover edge:

“Find something that’s kind of working and then do it better than them and then beat them. And unless that first thing had incredible network effects, which few things do, you actually can just beat them from a better product."

"Let it rip”

The most concrete segment is Ankit’s pitch to stop being stingy with compute. His own four-month rabbit hole: he didn’t realize how good the tools had gotten until he paid for the $200/month max plan and discovered that below that level “you actually are not anywhere close to the frontier.” Gary Tan, he says, spends thousands of dollars a day on tokens — an admitted privilege, but a window into where costs are heading. The technique is to do things no human would bother with: not 20 unit tests but 10,000, not a few docs but tons, every corner case covered. He found new product ideas just by pushing an email-client side project to the limit — noting that Gmail’s auto-reply is bad, but “if you’re willing to spend like $5 per email, it’s actually really good.”

The counterweight, raised by the host: what about people who can’t afford that? The panel’s answer is partly open-source models (they name minimax and a YC company, Open Code, as cheap and surprisingly good), partly “go work for a company that gives you an unlimited token budget.” There’s a real-world point underneath — Meesho’s founder wants voice AI to bring the next billion shoppers online, which only works at a price point those users can reach, which likely means open-source models. The honest reading is that the frontier is great for coding, but a lot of tasks don’t need it.

What YC actually looks for

Clarity above everything — if they can’t understand what you’re building, it fails. They invest in the founder, not the idea, because ideas get pivoted repeatedly. The two qualities they name are taste (intention and customer-grounded design choices, not just things that look good) and agency (the disposition to make things happen rather than let the world happen to you — they cite Paul Graham’s essay “Relentlessly Resourceful”). One useful, specific definition lands here:

“A project is when two people build something that was not assigned to them and get someone to use it.”

You can have a full CS education and a successful career and never once do a project by that definition. Do projects, especially young, and you’re “guaranteed” to find startup ideas.

The closer, asked what’s changed in 20 years: surprisingly little. Interview Thomas Edison and he’d sound like today’s founders — obsessed with customers, tinkering, building at the edge. What AI changed is leverage and speed. Ship not next year, not tomorrow — “more like tonight.”

Key Takeaways

The previous Indian startup wave won via hyper-local network effects (tokenized labor); AI removes the local moat, so the addressable market is global from day one.
The binding constraint shifted from “can you build it” to “how fast can you learn” — which structurally favors young, technically deep founders.
Distribution barriers fell because buyers everywhere are simultaneously open to AI; merit and product quality now get a cold-outreach hearing that used to require warm intros.
Second-mover advantage is real right now: with coding agents, a superior product can overtake an incumbent unless that incumbent has genuine network effects (rare).
Spending on compute is under-rated — below roughly $200/month you’re not at the frontier; the willingness to “let the tokens rip” (10,000 tests, exhaustive docs) produces better output, not slop.
For cost-constrained builders: open-source models (minimax, Open Code) are cheap and increasingly capable, and the next-billion use cases will likely be built on them.
The “safe” prestigious career may now be the risky one if AI reshapes those professions — though the panel honestly concedes Indian founders lack the safety net that makes US risk-taking easier.
YC selects for the founder, not the idea: clarity, taste (customer-grounded intention), and agency (“relentlessly resourceful”). Ideas get pivoted; people don’t.

Claude’s Take

This is a pep talk, and it should be read as one. It’s the closing act of a recruiting event — half the runtime is YC selling YC, and nearly every thread routes back to “apply to us,” “use the credits we’re giving you,” “go work for the companies that presented.” The incentives are not hidden, which is to their credit, but they’re loud.

The central claim — AI is global, so Indian technical depth can finally build very large global companies — is plausible and genuinely interesting, but it’s asserted more than argued. The evidence is two YC companies (Giga, Emergent) and one viral anecdote about an IIT student cold-emailing insurers. That’s a thesis with promising data points, not a proven trend. “Some of the largest companies in the world will come out of India” is a forecast dressed as an observation.

What raises it above the usual founder-bait is the honesty in two places. Arnav’s caveat about the missing Indian safety net is the most useful sentence in the talk — it quietly admits that “just take the risk” is cheap advice imported from a context with cushions that don’t exist here. And the “let it rip” segment is concretely actionable rather than inspirational fog: there’s a real, checkable claim that frontier-level compute meaningfully changes output quality, with specific techniques.

The weak spots are the usual survivorship lens (you hear from the six who made it, not the cohort that didn’t) and a faith that token costs always fall and buyers stay open — both true so far, neither guaranteed. Six out of ten: a clear, well-delivered articulation of a real shift, worth the half hour for the framing of why this wave differs from the delivery-app wave, but it’s a motivational keynote, not analysis, and the conflicts of interest are doing visible work.