One in four operator-led Indian startups founded in 2025 is an AI company. AI ranks fourth by dollars raised.
Sit with that gap. Of the 189 operator-founded startups RTP Global and Tracxn studied across 2023 to 2025, 51 are building AI, and 30 of those are vertical AI—domain-specific tools for a single industry. That makes AI the single largest destination for these founders for the first time. Yet the capital hasn't followed. Investors are still writing their biggest cheques into fintech, retail, and mobility.
The founders sprinted ahead of their backers. Which raises the question this report tries to file under "early innings": when a VC funds one of these operators, what are they actually buying?
When a VC funds one of these operators, what are they actually buying? The cheque is underwriting the past, not the paradigm.
This week the story moves down a layer, from thesis to selection: who actually gets founded, who gets funded, and why "seasoning" quietly became the screen.
The seasoning case is genuinely strong on paper. Operator-founders—former CXOs and senior operators from the Flipkart, Swiggy, Razorpay, Zomato, Freshworks and InMobi lineage—are 16 times more likely to reach Series A than the broader ecosystem: 4.8% versus 0.3%.
They raise faster and larger. Average seed rounds of $2.4 million against $1.4 million for everyone else. Average Series A of $10.5 million against $8.4 million.
And the valuations are climbing. Average Series A valuation for this cohort hit $53.3 million, up from $38.5 million in the report's first edition. Less than 1% of all startups founded in the period, but 11% of 2025's funding—more than double their 5.8% share in 2023.
In a year when broader startup formation contracted and capital concentrated into fewer names—1,100 startups raised roughly $11 billion, down from 1,680 deals a year prior—operator-led formation actually rebounded 35%. This is the safe harbor of a selective market.
RTP's Nishit Garg names the appeal directly: "pragmatism, execution focus and capital discipline." As AI changes how products get built, he argues, those operator qualities "become even more valuable."
That's the thesis in one sentence—and it's worth taking seriously, because the domain-moat version of it is concrete and compelling.
Consider Vignesh Ethiraj, co-founder and CTO of NetoAI, who came up through network engineering—"watching networks break at 3am from a NOC chair"—before founding the company in 2024.
His pitch is that telecom is not a workflow problem an LLM can wing. "Telecom problems are physics problems," he writes—RF propagation, photonics, electromagnetism—and general-purpose AI "tops out around 75% on real telecom tasks."
So NetoAI built TSLAM, a telecom-specific model trained on 100 billion-plus tokens of real telecom data, claiming roughly 95% accuracy on those same tasks. In production, one satellite operator reportedly cut fault-troubleshooting time by around 70%.
That 75-to-95 gap is the domain moat made numerical. It's exactly what a VC screening for seasoning wants to see: a founder whose scars translate into a defensible accuracy delta a generalist can't touch.
Or take 30 Sundays, which just raised ₹61 crore—about $6.7 million—in a Bessemer-led Series A. Co-founded by an ex-BCG consultant and an ex-Apple engineer, it rebuilt outbound holiday planning around a conversational AI interface, scaling to a ₹200 crore annualised gross booking run rate in two years while staying profitable.
Bessemer's Anant Vidur Puri frames it as workflow surgery: "30 Sundays has rebuilt the journey using AI while retaining human intervention where it matters." Domain knowledge plus AI, embedded across acquisition, itineraries, follow-ups and reservations. Exactly the operator-plus-AI recipe the report celebrates.
Here's the tension. Domain depth is a static asset, and the paradigm shift underneath it is not.
The same AI wave that de-risks these deals is the one flattening legacy SaaS—the bulldozer that made investors wary in the first place. And a bulldozer doesn't check your CXO title before it levels the ground.
The uncomfortable logic: an AI-native competitor can often learn a workflow faster than an incumbent operator can bolt AI onto the workflow they mastered the old way. The operator's edge is knowing how telecom or travel works today. The disruptor's edge is not being anchored to how it works today.
NetoAI's 95% is real and defensible now. But domain data moats erode as foundation models improve and as synthetic and open telecom data proliferate—TSLAM's own open variants already sit at 26,000-plus downloads on Hugging Face, which is both distribution and a reminder that domain models are becoming commodities faster than most theses assume.
Even the report's authors seem to sense the mismatch. AI is the number-one destination by count and only number four by dollars—which, read uncharitably, means investors like the founders more than they've committed to the AI theses those founders are chasing.
That's the tell. If VCs had conviction in the AI bets themselves, capital would track founder movement into the sector. Instead it lags, pooling in the sectors where these operators have the longest track records—fintech, retail, mobility—the very places their resumes read most reassuringly.
So the cheque is underwriting the past, not the paradigm. It's buying execution discipline and de-risked stage progression, then pointing that discipline at a technology whose defining trait is that the rules keep changing.
That can absolutely work. Execution discipline against a real problem is not nothing, and operator-founders are demonstrably better at converting a first cheque into a Series A. Sixteen times better.
But there's a selection cost hiding in the 4.8%-versus-0.3% number that Fund IX diligence should stare at directly. Screening hard for seasoning structurally filters out the founders least burdened by how things used to be done—the naive reinventors who tend to author paradigm shifts precisely because they never learned the incumbent's constraints.
The RTP report is honest that this is "an early-innings story, with capital yet to follow the founder movement fully." The generous reading is that investors are patient and will catch up. The skeptical reading is that seasoning-as-screen is a comfort blanket dressed as edge—a way to feel prudent in a market that punished imprudence, while quietly betting against the disruption you claim to be funding.
Both readings can be tested. If the operator-AI cohort's accuracy moats hold and their Series A valuations keep compounding past $53.3 million, the seasoning screen was genuine edge. If AI-native upstarts out-learn them on the workflows the operators supposedly own, the resumes will have been a hedge that hedged the wrong risk.
There's a related diligence question worth raising—who owns the exhaust, the proprietary data and workflow traces that a domain model like TSLAM accumulates in production. If that exhaust compounds into a real moat, the operators win. If foundation models close the gap faster than the exhaust accrues, the moat was rented.
For now, the cleanest fact in the dataset is also the most unsettling one for anyone writing these cheques. India's most experienced founders have decided AI is where to build. Their most experienced backers have not quite decided to believe them.
They're buying the founder. The thesis is still on layaway.