Mark Cuban on Why AI Could Turn Solo Founders Into Trillionaires

Summary

• Cost Plus Drugs is using public prices and consumer trust to introduce competition into an opaque market. Cuban says the business has grown from 111 generic medications in January 2022 to roughly 4,000 SKUs, with a set 15% markup and no advertising spending. Julie Yoo says Cost Plus’s mere presence forced incumbents to change where decades of policy and other pressures had not. Its public pricing API also gives large buyers negotiating leverage: “Trust is still our primary product.”
• Cuban argues vertical integration lets large carriers raise costs and steer people toward affiliated businesses while obscuring what providers are paid. He cites UnitedHealth’s 3,000-plus subsidiaries and $150 billion in annual intercompany transfers. Yoo sees AI enabling new health plans; Cuban’s counter is that better technology cannot undo unfavorable contracts or eliminate the question of who takes risk.
• His immediate AI prescription is to audit benefits contracts before signing them. Dallas’s PBM contract runs 1,335 pages, illustrating why manual review struggles. For organizations with more than 1,500 employees or members, he says savings could reach millions, depending on existing overcharging; his suggested prompt is: “Where are they fucking me over, and what should I do about it?”
• Manufacturing pods are Cuban’s largest commercial ambition, with aggressive volume targets and substantial regulatory costs. At roughly $20 million each, he projects a couple hundred annual pod sales within two years and a couple thousand within five. He says domestic generic production could undercut his Indian suppliers, but puts each generic’s ANDA application at $365,000 plus other fees—$365 million for 1,000 applications, before manufacturing investment.
• Personalized medicine could become an AI subscription feature, but Cuban leaves the timing wide open. Whether in five, ten or twenty years, he expects personalized drug-design costs to approach zero and suggests including the service in $20 or $200 subscriptions. Yoo expects better drugs to change care delivery; both see substantial demand for doctors, with Cuban’s formulation: “Novel data equals person.”
• Cuban’s consumer agenda combines access to existing hospital assistance with a proposed alternative to insurance premiums. He says some nonprofit assistance programs extend to 600% of federal poverty levels, potentially above $100,000 for some families in some states. His proposed system would retain premium-equivalent deposits in individually owned accounts, backed by government-guaranteed loans and stop-loss coverage; unused balances and interest would become available at 65.
• He treats AI adoption as a management and distribution challenge as much as a technical one. Employers need executives who understand the tools and entry-level employees who accumulate organizational context. Frontier labs, meanwhile, need to reach personalized social feeds: Cuban says position papers are “useless” without that distribution, while entrepreneurs should use AI to learn rather than avoid learning.

Deep dive

1. Transparent prices give Cost Plus leverage beyond its own sales

• Cuban dates the launch to January 19, 2022, with 111 generic medications; the portfolio now totals roughly 4,000 SKUs. Growing volume attracted additional manufacturers and products. He says the business spends nothing on marketing or advertising, relying on savings to generate referrals. Even if someone offers “cost plus 1%,” his defense is that “Trust is still our primary product,” supported by transparency and a set 15% markup.
• Local access comes through teamcubancard.com and approximately 17,000 affiliated pharmacies. Its $12 dispensing payment supports independent pharmacies but makes a $5 statin uncompetitive. For expensive specialty generics, biosimilars and insulin, Cuban says local pickup can be cheaper, particularly when it avoids cold shipping. Separately, Cost Plus Marketplace sells hospital injectables at “cost plus ten or cost plus twelve”; purchasing departments sometimes resist despite CEO support.
• Yoo says Cost Plus’s presence forced incumbents to change, despite policy and other forces having failed to change them for decades. Cuban sees its public API becoming a negotiating benchmark: large buyers can demand, “You have to beat this price.” He also says at least one company told him a big-three carrier had claimed Cost Plus was in its network, although Cost Plus has no agreement with any of the big three.

2. Incumbent control limits what better technology can fix

• Yoo describes a potential replacement cycle in employer health plans, with some startups facing 60% annual premium increases. Cuban says vertical integration compounds the problem: UnitedHealth has 3,000-plus subsidiaries and $150 billion in annual intercompany transfers, by his account. Employers can face higher premiums and deductibles while patients are steered toward affiliated businesses whose ownership they do not recognize.
• Cuban says 60% of people receiving employer healthcare are covered by self-insured employers. Carrier-owned administrators can obscure the difference between employer charges and hospital receipts. Employers may lack their own claims or the 835 and 837 files needed to reconcile payments. His contention is that medical spread pricing is worse than pharmacy spread pricing, across a much larger spending pool.
• Yoo sees AI making full-stack health plans easier to build. Cuban’s pushback: “This all comes down to who takes the risk.” He expects payer-provider coding “bot wars” eventually to net out, although early adopters can gain temporarily. Acquisition-created data silos could nevertheless make the large carriers slow to become AI-native.
• Asked about EHR disruption, Cuban answers, “Honestly, I don’t know.” Epic dominates, and he focuses on partnering to show drug prices when clinicians prescribe. Yoo welcomes the announcement but objects that availability waits until “May of next year.” Cuban agrees the delay is typical of healthcare.

3. Contract scrutiny comes before operational automation

• Cuban calls contractual analysis AI’s most important near-term healthcare application. Dallas’s PBM contract runs 1,335 pages: “There’s no group of humans with enough coffee on the planet” to catch everything. Once unfavorable terms are signed, he argues, operational AI cannot protect the employer. Review must happen before commitment.
• His agents find public-records opportunities, locate contracts and RFPs, and draft requests that he personalizes and sends. Returned documents enter Google Drive, where several models identify favorable terms, problems, remedies and a plain-language explanation. He also checks one model’s conclusions against another: “One of your peers came out and said this about this contract. What do you think?”
• Cuban gives the condition as “that 67% of employers representing 67% of the 160 million-plus Americans”—an ambiguous formulation—and says the system would collapse if they reviewed and negotiated contracts this way. He recommends a dedicated healthcare CEO or CFO. For organizations above 1,500 employees or members, savings could reach millions, depending on overcharging. Some public companies have acknowledged problems privately but, he says, fear discussing them on earnings calls.
• His direct-contract offer exchanges administrative simplicity for better prices and publication rights. Generally trust the doctor, avoid prior authorization and denials except extraordinary cases, pay the contracted amount on time, and assume patient-payment responsibility or credit risk. Providers then avoid becoming “subprime lenders.” Contracts published on costpluswellness.com become reusable agreements or negotiating templates for other employers.

4. Manufacturing pods carry the larger revenue ambition

• Co-founder Dr. Alex Oshmyansky originally proposed manufacturing generics on the FDA shortage list, including hospital and pediatric cancer medications. Cuban says AI and robotics compressed a football-field-scale operation into 5,000 square feet and permit switching drugs every four hours, producing a six-month supply before moving to another product.
• That equipment has been reduced further into tractor-trailer-sized manufacturing “pods,” configurable for different drugs and deployable to disaster or military settings. Pods can connect together, including separate sterile and tablet operations. Cuban says the government was the first customer, the first pod was shipped recently, and a second was ordered. He defers technical detail to Oshmyansky, but emphasizes research-hospital configurations for N-of-one therapies; Dallas is already making ASO drugs.
• At about $20 million per pod, Cuban says margins are attractive. His forecasts are a couple hundred sold annually within two years and a couple thousand within five. At $20 million each, 1,000 pods sold per year would represent $20 billion in annual sales: “If we’re selling $20 billion worth of pods a year, watch out, Elon.”
• Cuban says pods could manufacture generics in Dallas for less than Cost Plus pays its Indian manufacturers. The constraint is approval economics: he puts each ANDA application at $365,000 plus other fees, making 1,000 applications $365 million before pod costs. An American flag identifying its own US-made drugs on the retail site is “probably two years away” because of FDA work.

5. AI adoption still depends on people who understand the business

• Cuban rejects eliminating entry-level hiring. Models lag new events, so employees must supply context, interpret consequences and accumulate implementation history. His employment framing is categorical: “People who know how to use AI replace the jobs of people who don’t know how to use AI.” Executives must understand models and agents before enabling their organizations.
• His example is a consumer-food executive discussing GLP-1-driven acquisitions. Cuban asked ChatGPT to design a cheaper, more GLP-1-friendly competitor to the company’s leading cookie, including a full bill of materials, sources and manufacturing. In roughly three minutes, the executive responded, “Oh, shit.” Cuban’s point is that management cannot enable useful experimentation without knowing such questions are possible.
• Agent commerce also needs information agents can access. Cuban says they struggle to consume social posts and ads. His experimental agentfeeds.dev, “not truly active,” turns post transcripts into feeds people can give their agents. He expects many such services, while acknowledging unresolved trust problems.

6. Consumer agents could reveal financial assistance already available

• Cuban cites the claim that 40% of Americans cannot afford a $400 emergency bill. His suggested first step is to ask AI for nearby nonprofit hospitals and investigate their financial assistance programs before an emergency. He says some programs reach 600% of federal poverty levels—potentially over $100,000 for some families in some states.
• His conditional example is someone who thoroughly understands a local hospital’s assistance program considering whether to accumulate premium-equivalent money in a money market fund. He explicitly qualifies this: “I don’t want to make the decisions for people.” The opportunity he emphasizes is helping families understand eligibility, available options and how to obtain assistance.
• If starting a company here, Cuban would make nonprofit hospitals’ assistance programs accessible through consumer agents. Hospitals, he says, “suck at marketing” and depend excessively on insurer networks to deliver patients. Better contracts and cheaper networks still face adoption barriers until patients understand how to use them. Reducing fear of a medical financial crisis is his proposed route to AI acceptance.

7. Personalized drugs could become subscriptions without replacing doctors

• Asked about frontier labs’ ultimate impact, Cuban predicts that personalized drug-design costs will approach zero as algorithms improve. His timing remains “five years, ten years, twenty years.” He starts with over-the-counter medicine: instead of experimenting with how standard ibuprofen affects each person, models could tailor doses or mixtures, then progress to other personalized drugs.
• He has told two frontier models to include this service in $20 or $200 subscriptions once available. DNA collection might require an upfront payment, with participation contingent on willingness to share sensitive information. He acknowledges that initial costs would be enormous before falling. “I’m selling my own book here”: Cost Plus wants its pods positioned to manufacture those individualized treatments.
• Yoo suggests personalization could shift doctors’ work away from medication troubleshooting and titration, while increasing demand for actual care and moving physicians toward “physician-scientists.” Cuban agrees: “Novel data equals person.” If a personalized medicine makes him vomit, he wants a human observing with all five senses and incorporating family context. He rejects replacing doctors; their training would instead change, and he anticipates work for “AI translators.”
• Intellectual property complicates the abundance thesis. A clinician could restrict a useful procedure to a proprietary model and charge “$9,999 for every access.” Cuban argues that selecting the cheapest model is insufficient if essential knowledge is unavailable for training, post-training or retrieval. “The information may want to be free, but the person who owns the IP feels differently.”

8. Cuban would redirect premiums into savings and guaranteed credit

• His policy agenda starts with publishing all taxpayer-funded contracts without redactions, exposing nonprofit hospitals’ detailed general ledgers and procurement records, and crediting cheaper purchases toward insurance deductibles. He says eight states already allow the last measure at some level. He also says 72% of large carriers’ revenue comes from taxpayers and argues that they offload risk. Although he supports universal healthcare, he argues that reliable economic data must come first.
• His alternative to conventional premiums uses mandatory deposits into individually owned accounts. Illustratively, a Texas family of five would deposit $2,100 at the specified income level. A bank would lend when balances fall short, backed by government guarantees; part of deposits would fund stop-loss coverage, perhaps above $40,000. Payroll deductions repay loans. He argues government already guarantees education, housing and business credit.
• At 65, participants moving to Medicare would receive unused savings and accrued interest. Spending restrictions would prevent misuse. Separately, Erik Torenberg asks whether healthcare could still consume roughly a fifth of GDP in ten years despite better technology. Cuban says yes: the gain should be healthier, more productive people and better value from spending, rather than necessarily a smaller economic share.

9. AI’s public case must reach individual feeds

• Cuban’s advice to frontier labs is to use their own tools to understand social algorithms, segment audiences and enter personalized feeds. Conventional ads and position papers miss how people form opinions. His instruction is blunt: “Do it like the Russians do it.” He argues that people no longer share the same information environment, so broad messaging cannot assume a common audience.
• He describes data-center hostility as a proxy for poor understanding of AI, while also criticizing communities for unfavorable negotiated deals. The connection to healthcare is contractual: communities, like employers, can accept terms they do not understand. He says legislators also underestimate the consequences of everyone receiving a different social feed.
• Asked for neglected entrepreneurial opportunities, Cuban answers, “All of them.” He distinguishes people using AI “to learn anything” from those using it “so they don’t have to learn anything.” His closing advice is to experiment, verify outputs and tolerate awkwardness: “That uncomfortableness is probably the path to something that is groundbreaking and game-changing.”