Mark Cuban on AI agents replacing SaaS, humanoid robot skepticism, and Cost Plus Drugs scaling

Summary

Mark Cuban sees AI agents replacing SaaS as the dominant startup pattern in 2026. Dozens of founders are building vertical-specific agent businesses targeting marketing teams, customer service, and operations automation, positioned as monthly subscriptions that eliminate the need to hire. Nearly all of them are still hunting for traction. Cuban notes that founders with real revenue growth aren't coming to him yet. The barrier to entry is so low and iteration so fast that the real battle now is customer acquisition and proving unit economics work at scale.

On robotics, Cuban is contrarian. Humanoid robots have a five to ten year lifespan before they fail. The assumption that houses need humanoid robots because humans live in them is backwards. Instead, houses will be redesigned to fit optimal robot form factors such as spider-like or ant-like designs with specialized appendages, while robots coevolve to match the environment. Stairs aren't a blocker; homes will get mechanized dumbwaiters. The real constraint is bandwidth. As AI shifts from text-heavy LLMs to video-based world models, the infrastructure demand for video ingestion will become prohibitive. Self-driving cars scare him less than adversarial graffiti. He tested Tesla's full self-driving at highway speeds and found it terrifying. Someone is already trying to fool vision systems with malicious patterns painted on medians and stop signs.

Cuban is not worried about AI doom. LLMs today can't model physical consequences. A two-year-old with a sippy cup understands cause and effect: knock it over and mom runs. Large language models never learn that. To build real-world reasoning, AI would need to capture physics, manage latency, and process massive video streams. That processing and bandwidth load makes a Terminator scenario implausible. What worries him more is information fragmentation. In 2002, media gatekeepers controlled the narrative. Now algorithms create personalized realities. He references his Fallen Patriot Fund, created after the 2003 Iraq invasion to support injured soldiers' families, as a moment when there was at least a shared information foundation. Today, no two people consume the same algorithmic feed. That makes it almost impossible to make informed decisions about anything.

Cost Plus Drugs is scaling rapidly. The consumer-facing play at costplusdrugs.com shows the actual cost of medications plus a 15% markup and $5 shipping. For most drugs, that undercuts insurance copays. The B2B side is where the real thesis lives. Cuban approached hospitals and self-insured companies with a straightforward pitch: he pays on time at Medicare reference pricing, roughly 100% of Medicare rates, with no clawbacks or delayed payments. In exchange, he gets better pricing and the right to publish the contract on costpluswellness.com so any company can reach out to that hospital and negotiate the same deal. The model has grown to more than 9,000 providers. The strategy teaches self-insured companies they can escape dependence on insurance carriers, who Cuban argues steal through underpayment and denial games. Hospitals have been turned into subprime lenders because 40% of Americans can't cover a $400 emergency, so they end up financing patient debt. On why he chose healthcare over easier exits, Cuban is direct: healthcare is broken and nobody says the economic side is great. If you're going to disrupt, go big or go home.

The most underrated business he has seen is streaming. In 1995, he and a friend started AudioNet to broadcast Indiana basketball into Dallas over the nascent internet. He registered audionet.com for free because nobody wanted it. By 1997, when he expanded into video, he found broadcast.com and paid the owner $8,000. That owner later regretted the deal, but by then Cuban had already glommed dozens of high-traffic domain names such as final4.com, baseball.com, and sandwich.com to drive traffic back to broadcast.com. This was pre-Google, when people typed domains directly into browsers like portals. He still owns domains like misterpresident.com and democracy.com for the same reason.

Cuban is skeptical of the peptide boom, the unregulated supplement versions coming off the boat. He puts peptide queries into LLMs and asks for trial data and research. The results don't convince him. He draws a clear line between that and FDA-regulated GLP-1s like Ozempic and Mounjaro, which he respects. Eli Lilly and Novo Nordisk got smart by working around PBMs and selling direct to patient and direct to company, which will eventually pressure prices down.

On Amazon's widget ecosystem, Cuban identifies a structural unfairness. American sellers can operate only one company on Amazon and must pay taxes and establish nexus. Chinese and Vietnamese sellers face no such limits, don't post bonds, don't pay taxes, and Amazon deposits money directly into foreign bank accounts. This cost advantage makes knockoffs trivial. He proposed to legislators that foreign sellers post a $25,000 bond before listing, with a 90-day challenge window for original creators. That single rule change would revive American manufacturing and shift the economics entirely. The cost of protecting IP goes beyond lost sales to lawyers, legal notices, and DMCA takedowns, all preventable with a bond requirement.

On using his likeness in AI-generated content, Cuban notes he was an early Synthesia investor, putting in substantial capital ten to twelve years ago. He played with Sora and other tools to learn them. With Synthesia he set conditions: any video using his likeness must end with the Cost Plus Drugs logo. That single constraint has generated hundreds of thousands of impressions and a measurable bump in signups. He's aware that unregulated versions generate deepfakes of him doing lines of coke, which violates the platforms' terms of service but happens anyway.