Mark Cuban on the AI Bubble: Who Actually Gets Wiped Out?
with Mark Cuban
20 Jul 20264 min read30m
TL;DR
Mark Cuban argues the AI bubble won't hurt everyday investors the way dot-com did — it'll wipe out VCs and private funds who deployed at peak valuations into companies pricing perfection. Meanwhile, he's bullish on AI for entrepreneurs (pointing to Lovable generating 770,000 apps a week) but skeptical of the '50% of white collar jobs gone' narrative, calling enterprise AI implementation far harder than anyone expected.
Key Moments
Mark Cuban
“if AI, we're talking about, you know, AGI and we're talking about taking over the world. If you can't you if you need to have forward deployed engineers that tells you all you need to know about AI because by definition you should just be able to ask AI to do what I need you to do”
Cuban using the fact that Microsoft, Anthropic, and OpenAI all need forward-deployed engineers as proof AI isn't as capable as claimed
“agents get bored, right? and they drift because as the underlying um large language model starts to change, the way it was originally programmed doesn't match what the large language model turned into”
Cuban warning that AI agents become brittle over time as the underlying models update, creating ongoing maintenance challenges
“if you show AI a video of a two-year-old on a high chair with a sippy cup, the two-year-old knows if you push the sippy cup over the edge, mom's going to come running and the kid's going to start laughing. AI's got no clue what's going to happen.”
Cuban illustrating the gap between AI and basic human common sense, arguing we have a long way to go
“it's not a bubble that's going to impact most people in the room, right? Or most people um across the US, but it could just destroy a lot of VCs and a lot of funds and a lot of PE, right? Because they're going all in.”
Cuban distinguishing the AI bubble from the dot-com bubble in terms of who gets hurt
▶ 0:00
About Mark Cuban
›
Mark Cuban is a serial entrepreneur and investor best known for selling Broadcast.com to Yahoo for $5.7 billion in 1999 and owning the Dallas Mavericks NBA franchise. He was a longtime investor on ABC's Shark Tank and has backed hundreds of startups across tech, AI, and healthcare. More recently, Cuban has been vocal about AI's real limitations at the enterprise level and has invested in companies like Lovable, Open Evidence, and Synthesia.
Takeaways
1
Lovable is creating 770,000 apps a week — 80% non-engineers Cuban's investment in Lovable is producing a striking stat: 770,000 applications built per week, with only 20% of users being engineers and 70% of usage outside the US. This is the clearest current evidence that AI is democratizing software creation for non-technical entrepreneurs globally.
2
Go public now to use stock as acquisition currency Cuban's contrarian advice to AI-era startups: IPO early — even at $50-100M — so you have stock as currency to acquire legacy businesses or data-rich companies before competitors do. Staying private means raising expensive cash every time you want to make a move, which is a structural disadvantage in a fast-moving disruption cycle.
3
AI agents drift and break as models update Cuban flags an underappreciated operational risk: agents built on top of LLMs become brittle when the underlying model changes, because the original prompts and logic no longer match the new model's behavior. This creates ongoing maintenance overhead and a real business opportunity for anyone who can manage it.
4
Forward-deployed engineers prove AI's limits Microsoft hiring 6,000 people, and Anthropic and OpenAI deploying engineers directly into enterprises, is Cuban's tell that AI can't yet self-implement. If it could truly reason and plan, you'd just ask it how to deploy itself — the need for human hand-holding reveals the gap between AGI hype and reality.
5
AI bubble hits VCs, not retail investors Unlike dot-com, AI valuations are concentrated in private markets — Anthropic, OpenAI, SpaceX aren't publicly traded. The pain when corrections come will fall on VC funds, PE, and private credit, not everyday people watching stock tickers.
6
LLMs may reduce political polarization where social media failed Social media optimizes for engagement and reinforces whatever you already believe. LLMs are structurally incentivized to be accurate — their business model depends on trust, not time-on-site. Cuban predicts people will increasingly turn to LLMs for political questions and get more balanced, factual answers than any algorithm-driven feed would serve them.
7
Data center overbuilding mirrors the dark fiber era Cuban draws a direct parallel to the 1990s fiber overbuild: bandwidth scarcity drove massive infrastructure investment, then efficiency breakthroughs made most of it redundant. He expects AI compute efficiency gains to similarly strand a lot of data center capex — unless video and world models consume enough tokens to fill the capacity.