FPV Ventures' Nikunj Kothari on startup lessons, AI app wave, and why data ownership determines winners

Summary

Nikunj Kothari, a former operator at Meter and Opendoor who recently joined FPV Ventures as an investor, makes the case that the application layer is now where durable AI value gets built. FPV closed its latest fund at $525 million and manages approximately $1 billion in assets across a generalist mandate spanning biotech, consumer fintech, and infrastructure.

Foundation Models Are Largely Played Out

Kothari's view on foundation model investing is direct: unless a startup brings a genuinely new architecture, the window has closed. Capital requirements make it nearly impossible to compete, and transformer-based approaches offer little differentiation. The more interesting bet is proprietary data enabling specialized models. He points to Meter, his former employer, which is building a networking model trained on packet-level data unavailable to any external party, developed in close partnership with Microsoft. That kind of data moat is where he sees remaining upside in the model layer.

The App-Builder Convergence Is Real, but Driven by Two Separate Motivations

The surge of prompt-based app-building tools from Replit, Bolt, Figma, Lovable, Airtable, Retool, and Canva (an FPV portfolio company) reflects both defensive and offensive positioning. Defensive players are reacting to internal-tool builders eroding their core use cases. Offensive players are using app creation as a wedge to pull in broader workflow context, increase retention, and consolidate spend. The bundling thesis drives both: enterprises increasingly want consolidated solutions rather than five separate subscriptions.

Kothari attributes the pile-on partly to the revenue momentum at Replit and Lovable, noting that Cursor has reached $500 million ARR and Replit scaled from $10 million to $100 million ARR in months. That velocity is pulling incumbents into the category whether or not the strategic fit is clean.

OpenAI Cannot Own the Whole Stack

On the threat from OpenAI expanding into meeting notes, context aggregation, and productivity tools, Kothari argues the horizontal-plus-vertical execution problem is underappreciated. Building broad platform features while going deep in individual verticals simultaneously produces cluttered interfaces and diluted focus. Claude already competes meaningfully on API quality for code generation, and no single company can absorb every vertical's workflow complexity. Thin-wedge startups built around features rather than deep workflow ownership are the ones most at risk of being displaced by incumbents.

Go-to-Market Is Bifurcating, and PLG Is Not a Universal Law

Kothari draws a clear line between two growth strategies emerging in AI products. The first is building quietly and letting organic product growth become the story, the Cursor model. The second is timeline-first virality, building audience ahead of revenue and converting it into subscriptions. He is skeptical that Twitter-native launch videos remain effective given saturation, and pushes founders toward multi-channel thinking, including local communities and events. On the enterprise side, he stresses the basics: ICP clarity, org-chart navigation, and identifying whether a problem is genuinely budget-prioritized. Cursor's path from bottoms-up viral adoption to enterprises requesting enterprise plans is treated as an exception, not a template.

The Traditional PM Role Is Functionally Obsolete

Kothari is unambiguous that product managers who operate as pure project coordinators are on a short timeline. No startup at FPV is requesting a classic PM hire. Founders want engineers, marketers, and salespeople with deep functional capability. The PM role that survives is the orchestrator who spans sales, marketing, prototyping, and people management simultaneously. He frames this as taking general manager P&L responsibility rather than product ownership, a model he applied at both Meter and Opendoor. The lag to large companies will slow the transition, but the direction is set.

Valuation Lifts for Non-AI Companies Depend Entirely on Data Position

The AI valuation premium does not flow automatically to non-native AI businesses. The determining variable is whether a company holds proprietary data that becomes an accelerant when AI is layered on top. Casetext is the reference case: eight years building a curated legal database meant GPT-3.5 became a step-function growth catalyst rather than a threat. Robotics, bio, manufacturing, and supply chain are where physical-world alpha remains concentrated as software gets commoditized. The friction point for non-AI companies seeking venture backing is the inflection requirement. Without a visible growth inflection, even strong businesses struggle to attract follow-on capital, a dynamic he ties directly to the current cohort of what he calls zombie unicorns from prior vintage years.