Season 07 · Episode 06 · Venture Capital · 1h 14 min
Are We in an AI Bubble? A VC's Honest Take
Rahul Agarwalla, Managing Partner at SenseAI Ventures, explains why proprietary data—not model access—is the real moat for AI startups, and why the application layer is only getting started.
August 28, 2026·With Rahul Agarwalla
§01Chapters
Introduction & Guest Welcome
From Entrepreneur to AI Investor: The Japanese Big Data Story
Why He Chose Investing Over Building Himself
A VC's Role: Guiding Founders, Not Deciding for Them
Building a Vision Without a Crystal Ball
From CRM Pain to Lead Generation
Evaluating Startups Now That Building Is Cheap
Debunking “AI Will Build Every Business”
Lessons From the Internet Era: Google, Yahoo & Infrastructure
Why Data Beats Algorithms Long-Term
When AI Slop Fails Customers
Where Value Shifts: Oracle and the “Cement vs Paint” Analogy
Applied AI vs Frontier Models: Where Value Gets Captured
Sovereign AI and Strategic Independence
Electricity, Internet, and Where Margins Actually Go
Falling Price Per Token and Rising Margins
Does FOMO Ever Win in AI Investing?
Separating Price From the Investment Decision
How a VC Fund Actually Makes Money
Why Not Every Good Business Is Venture-Fundable
Putting Founders First, Even in Rejection
The Power Dynamics Between Founders and VCs
What Makes Being a VC Rewarding
SenseAI's Moat in a Crowded Market
Rapid Fire Round
Final Thoughts & Closing
§02Show Notes
Rahul's conviction in machine learning began with a Japanese text-analysis problem: finding a computational linguist whose technology could parse a language written without spaces helped turn Honda, Toyota and Canon into customers of his previous company.
He describes an investor as a parent rather than a decision-maker. Founders make the decisions; the fund supplies pattern recognition, guidance and support without taking control of the company they are building.
His first-meeting filter starts with three questions: what problem are you solving, what product are you building, and how does the world change because you exist?
Founders do not need a 4K picture of the future. They need enough clarity to move, plus the judgement to pivot as the picture sharpens and the original problem proves insufficiently valuable.
The wrapper critique misses the limits of even the largest platform companies. OpenAI and Anthropic cannot build every application, just as Google could not turn every adjacent product bet into a lasting winner.
The durable moat is proprietary data. Public foundation models train on only a fraction of the world's information; the rest remains private, specialised or undigitised. In Rahul's formulation, great data beats a great algorithm every time.
SenseAI saw 1,263 AI startups last year, met roughly 600 and invested in six. Of the companies it has backed, 92 per cent have gone on to raise a follow-on round.
The fund separates the decision to invest from the negotiation over price, a discipline designed to keep conviction about a founder from turning into valuation-driven FOMO.
Roughly 75 per cent of SenseAI's portfolio uses OpenAI or Anthropic somewhere in its stack. Using a frontier model is not the same as lacking defensibility when the application owns the workflow, customer relationship and data.
Rahul expects capital to keep moving from infrastructure and token production toward the application layer, as falling token prices improve the economics of the businesses turning models into outcomes customers will pay for.
“Great data beats great algorithm every time.”
Referenced in this episode
§ Keep listening
