Joe Lonsdale on AI talent wars, incubating in Texas, and why the real economy is the most underrated opportunity

Summary

Joe Lonsdale argues that talent density, not revenue metrics, is the defining filter for AI-era investing. Pointing to Cognition's Scott Wu — a three-time global programming champion he first recruited to Palantir as one of eight gold-medal-level hires in a single year — Lonsdale frames elite technical pedigree as the clearest leading indicator of outsized outcomes. He concedes the window for arbitraging that signal is narrowing: where Palantir once had a near-monopoly on recruiting from the top 20 universities before Meta and Google began offering $100,000 signing bonuses around 2008–2009, frontier AI labs are now paying roughly 100 times that.

The Bar for Startups Has Moved Sharply

Lonsdale describes reviewing a company tracking from zero to $5M in revenue this year, believing it could reach $15–20M or more in 2025, but passing because the team was "a B, B-plus." He expects that company to raise at "a couple hundred million" regardless. The episode illustrates a broader dynamic: revenue ramp no longer compensates for talent shortfalls, and the power law is steepening. Hamad at General Catalyst's view that triple-triple-double-double-double growth is no longer interesting resonates with Lonsdale, though he locates the real differentiator in people and culture rather than any specific metric threshold.

His own miss rate is notable. He passed on the founding team that pivoted into Cursor after dismissing their original idea in 2022, when they were raising $7M at a $40M post-money valuation. That team is now reportedly raising at a $5B post-money valuation. A similar pass on another MIT-linked team compounded the lesson: with sufficiently strong founders, backing the idea is almost beside the point.

SaaS Business Model Transition

On public SaaS, Lonsdale sees a bifurcating outcome. Companies that own critical workflow infrastructure and retain genuine technical culture can use AI to accelerate meaningfully — he cites Atapar, which he founded fifteen years ago and which now holds over $8 trillion on platform, with its core infrastructure SaaS growing at 25%-plus. His target is getting that business to 40–50% growth by layering AI-driven workflows on top of existing customer relationships. Companies that fail to build that AI capability risk losing even their core value proposition to well-funded startups counter-positioning against them.

The broader point is that owning the customer relationship and the system of record provides durable advantage. He highlights Socotra in insurance as an example of a system of record with a credible path to capturing AI value before new entrants can replicate the underlying data and workflow depth.

Texas as an Industrial, Not AI-Model, Hub

Lonsdale is direct that founders pursuing frontier AI model work should still be in San Francisco, where network effects in talent remain dominant. Texas serves a different thesis. He frames Austin and the broader state as the optimal base for advanced manufacturing, defense technology, robotics, and healthcare — pointing to SpaceX and Boring Company alumni concentrations, MD Anderson as a leading cancer research center, and multiple multi-billion-dollar healthcare services companies already operating there. One portfolio company staffed by ex-Waymo engineers is automating construction equipment — excavators and quarry operations — in Texas, targeting what Lonsdale describes as a $100B-plus addressable market in quarrying alone.

On the regulatory dimension, Lonsdale contrasts Texas favorably with California, noting that state government engagement is substantive and commercially oriented regardless of party affiliation — a meaningful operational advantage for companies navigating permitting, infrastructure buildout, and defense procurement.

The Real Economy Is Underpriced

Lonsdale's highest-conviction underappreciated theme is AI applied to physical-world industries. He maps AI investment across six layers — energy infrastructure, chips, data centers, models, software infrastructure, and apps and services — and operates primarily at layers four and five. The massive capital flowing into the lower layers, including the gigawatt-scale data center builds from OpenAI (10 gigawatts cited by Sam Altman), Meta (1 gigawatt), and xAI's Colossus 2 (1 gigawatt across three states), creates compounding infrastructure that makes application-layer and real-economy AI deployment progressively cheaper and faster.

His argument is that the real economy represents approximately 85% of total capital, is early in its AI transformation, and will require substantial credit and industrial capacity through the 2030s. Most investors, he suggests, are over-indexed toward trillion-dollar platform plays and underweighting the hundreds of companies likely to reach $5–50B in value across niche verticals. He flags Candid, a healthcare billing company growing toward billions in revenue in a $280B market, as a prototype — potentially a $100B company in the early 2030s that almost no one outside the sector will have heard of.

Incubation Strategy and Portfolio Construction

Frustrated by valuation levels since 2021, Lonsdale has shifted toward incubating companies outright, taking the full first institutional round within his fund before selectively allowing co-investors into the second. He notes that in hot AI Series A rounds today he is satisfied getting 15–16% ownership, and has seen situations where he receives only 5–6%. By contrast, incubation preserves economics that conventional venture rounds have made structurally difficult to replicate. He cites Cyronic — a defense-tech company built around ex-Palantir talent serving the U.S. Navy — as a case where early, concentrated ownership was the correct call, acknowledging he was too generous letting network co-investors participate in early rounds.