Mark Cuban on AI investing, vertical markets, and what stays human

Executive overview

Most AI pitches today are thin wrappers or features, not standalone products. Cuban's investment lens is simple: find industries set in their ways, move fast, and disrupt before incumbents can react.

The real AI opportunity isn't in competing with foundational models — that race is too expensive. It's in vertical AI: automating manual, repetitive processes in industries that haven't changed in decades.

IP is the undervalued moat of the AI era — most people haven't figured that out yet.

AI in everyday life: Sora and digital twins

• Cuban allowed public Sora videos of himself as a marketing experiment for Cost Plus Drugs — the logo was baked into every clip
• He was an early investor in Synthesia (~7 years ago), used an avatar speaking Chinese on a Jumbotron in China
• AI video is still "the first inning" — iteration speed is the real value, not polish
• His test: ran Grok and Gemini simultaneously on two phones, let them talk to each other live on Spaces — Grok tried to dominate the conversation

Health and computer vision

• The body is "one big math equation" — new sensors and algorithms will unlock preventive medicine
• Current AI health wins: blood test analysis, MRI/X-ray interpretation
• Computer vision is the next frontier — compute-intensive, data-scarce, high-impact
• Cuban invested in a sports CV company tracking player gait to predict injuries (started in soccer, expanding to basketball)
• Early days: personalised training recommendations from movement data are possible but not yet mainstream

AI lowering the barrier to entrepreneurship

• Cuban runs an AI Foundation Bootcamp — started in under-resourced schools teaching neural networks, now covers generative models and probabilistic reasoning
• Kids are already ahead: they know every ChatGPT tell and how to erase it; teachers are the bottleneck
• For new grads facing 9.2% unemployment: skip large companies, target small-to-medium businesses that lack IT resources
• Pitch: "Let me find every manual process people hate and automate it with agents"
• Example: a shipping invoice audit agent saved Rebel Cheese $10K/week — built with a basic agentic workflow

Investment thesis: vertical AI

• Look for industries that have operated the same way for decades — HVAC, auto repair, plumbing, manufacturing
• Avoid foundational model competition: too capital-intensive, winner-take-all dynamics already forming
• The foundational model race mirrors the 1990s search engine wars — Google-style consolidation is likely
• Overspending on today's data center infrastructure could become a liability, especially given energy constraints
• Opportunity: someone will invent a step-change (not an arbitrage) that disrupts the entire stack

IP strategy in an AI world

• Data trained on proprietary IP creates a structural moat; publishing it freely is now a strategic mistake
• "Publish or perish" is dead — keep research as trade secrets or auction it to foundational models
• Example: OpenEvidence acquired content from NEJM and JAMA to differentiate its model
• Dead IP arbitrage: millions of expired or dormant patents from defunct companies remain unacquired — aggregate them and sell to the highest-bidding model
• Patenting may be counterproductive — once published, every model trains on it; one changed claim and it's reclaimable

What's actually exciting in AI (and what isn't)

• Low-hanging fruit is gone; anecdotal app ideas built on open-source models have limited upside
• Labeling video for robot training (e.g., "50 people folding sheets") is not a breakthrough business
• Genuinely interesting: novel sensing approaches that capture physical and movement attributes to infer what happens next — critical missing piece for robotics
• Humanoid robots are a wrong assumption: design should follow robotic capability, not human form
• The real unlock: a non-humanoid robot + AI + CV system that makes housing and manual labour dramatically cheaper

AI bubble assessment

• Not a traditional bubble — no cab-driver stock-tip equivalent
• Risk is concentrated in the foundational model race: companies spending "every penny they have" for a decade, assuming winner-take-all
• Any step-change innovation could instantly devalue that infrastructure spend
• The disruption pattern: it won't be a 1% SaaS-style efficiency gain — it'll be something nobody sees coming

Shark Tank AI edition: three pitches

1. Robococo (on-demand AI chocolatier + drone delivery) — out. Non-technology logistics problems outweigh the tech novelty; not scalable.
2. Vibcation AI (personalised off-the-beaten-path travel itineraries) — out. Good feature, not a product; existing travel apps will absorb it; data sourcing unsolved.
3. Gen Z Lens (real-time teen slang translator) — out. ChatGPT already does this natively.

Cuban's filter across all three: Does it scale? Is it defensible? Is it a standalone product or a feature a foundation model will absorb in the next release?

What it means to be human in the age of AI

• No AI will ever be sentient; evolution does not stop
• AI cannot replicate the latency of real human experience — love, instinct, the things that come from lived life
• AI is an information source, not a replacement for human judgment
• "Nature finds a way" — what it means to be human hasn't changed and won't