Mark Cuban on Healthcare Prices, AI and the Case for Transparency

Mark Cuban argues that the US healthcare market is inefficient and its economics are broken. Julie Yoo says Cost Plus Drugs has put pressure on incumbents in an industry she believes has long lacked competitive dynamics. Cuban's businesses are part of the market he discusses, and his claims throughout the conversation reflect that perspective.

Erik Torenberg asks Yoo what drew her to Cuban's work. She credits him with entering a complex industry and using his public platform to challenge established practices. In her view, the presence of Cost Plus Drugs has prompted changes that other forces had not achieved.

Yoo says buyers are beginning to ask whether they can access Cost Plus products, even without direct contracts. Cuban says the company's public prices and API give large buyers a reference point in negotiations. He also describes an alleged instance in which a company was told it could access Cost Plus prices through a large intermediary despite no agreement with that intermediary.

Cuban says Cost Plus Drugs launched in January 2022 with 111 generic medications and grew to roughly 4,000 stock-keeping units. As volume grew, he says, manufacturers expanded their offerings through the business. He describes keeping up with demand as a continuing challenge.

To offer local pickup alongside mail order, the company created the Team Cuban Card and affiliated with pharmacies. Cuban says participating pharmacies receive a $12 dispensing fee. That can make very cheap drugs cost more than elsewhere, he acknowledges, while he says some specialty generics, biosimilars and refrigerated products can be cheaper to collect locally.

Cuban turns to manufacturing, an idea he attributes to a cold email from co-founder Dr. Alex Oshmyansky. The original goal was to make generic drugs on the FDA shortage list, including products used in hospitals. Cuban says robotics and AI helped compress a much larger manufacturing process into a 5,000-square-foot facility, with production lines that can be changed over between drugs.

The company has also developed modular manufacturing units, or pods, roughly the size of tractor trailers. Cuban says they can be shipped and configured for different products or linked together, and that the government was its first customer. He describes potential uses in disaster response and manufacturing sterile injectables or generic tablets.

Cuban says a roughly $20 million pod could be sent to a research hospital and configured for individualized, or N-of-one, therapies. He says the Dallas facility is already making ASO drugs. He presents these as company capabilities and plans, rather than a claim that individualized treatment is broadly available today.

Cost Plus Marketplace sells sterile injectables made by the company or sourced from other manufacturers to hospitals, clinics and other providers. Cuban says its pricing can save hospitals substantial sums, although persuading purchasing departments remains difficult even when executives are interested.

Cuban describes a separate direct-contracting approach informed by his experience as a self-insured employer and conversations with hospital leaders. He says hospitals dislike delayed payments, denials, prior authorizations and high deductibles that leave providers trying to collect from patients.

His proposed exchange is to trust doctors' decisions except in unusual cases, pay contracted prices on time and take responsibility for patients' out-of-pocket amounts or credit risk. In return, he seeks a lower price and permission to publish the contract on Cost Plus Wellness. He wants employers and providers to use published agreements as examples or negotiation templates.

Yoo links this proposal to employer-sponsored health benefits. She says large national carriers' entrenched relationships and the cost of assembling networks have historically made the market difficult for startups to enter. She describes sharp premium increases as one reason employers are seeking alternatives.

Cuban argues that vertical integration compounds the problem. He claims large health companies can raise premiums and deductibles while directing patients to subsidiaries whose ownership is not obvious. These are his allegations about the industry; he has a competing business interest.

For self-insured employers, he says third-party administrators process medical claims, while carrier-owned administrators may operate under administrative-services-only contracts. Cuban argues employers often cannot reconcile what they paid with hospital charges and payments, and says access to claims data can be restricted. He sees bypassing intermediaries as a central opportunity.

Yoo says startups now have an opportunity to build health plans with AI reducing administrative work. Cuban agrees AI can help clinically, operationally and with contracts, but argues that adding it to a broken financial system will not by itself fix incentives or create transparent prices.

Cuban criticizes the length and complexity of benefits contracts, citing a Dallas pharmacy-benefit contract he says ran to 1,335 pages. He urges employers to use AI to review agreements before signing, arguing that a model cannot undo unfavorable terms after a contract is signed. His assessment is not a substitute for legal review.

Cuban says competitors face difficulty building networks while large incumbents have extensive subsidiary relationships. He supports legislation aimed at breaking up large healthcare companies because he wants a more open market for entrepreneurs. Yoo says disputes between payer and provider bots expose problems in the underlying billing system rather than originating with AI itself.

Cuban sees AI-driven battles over claim coding as a limited, short-term contest: each side can adopt similar tools. He thinks data siloed across acquired subsidiaries could make it hard for large carriers to become AI-native. His main recommendation remains reviewing employers' contracts before they sign.

He suggests feeding existing benefits, broker, consultant, administrator and pharmacy-benefit contracts into AI and asking where their terms disadvantage the employer and what could be done. He describes this as a way to surface questions quickly, though the discussion does not establish that model findings are accurate without further review.

Cuban describes using agents to find public entities with open-records processes, locate contracts or requests for proposals and draft records requests for his review. He keeps the resulting material in Google Drive and asks different language models to summarize favorable terms, unfavorable terms, possible actions and a plain-language version for the organization concerned.

He says this process exposes problems in contracts used by schools, cities, unions and other employers. Cuban predicts that widespread review and negotiation could sharply lower healthcare costs, and recommends that employers appoint someone responsible for managing benefits contracts. His claimed savings and market-wide effects are forecasts, not results established in this interview.

Asked about his long-term business vision, Cuban says the mission is to make healthcare affordable and that reliable data is necessary for decisions about a universal system. He argues that trust is the primary product of Cost Plus Drugs: public pricing and a stated 15% markup are meant to make its promise understandable to customers.

Cuban also sees the manufacturing pods as a commercial growth opportunity. He forecasts hundreds of sales a year within two years and potentially thousands within five, while acknowledging the cost and approval work involved. He says each generic drug requires an application and fees, which makes a large domestic manufacturing expansion expensive.

Technology alone will not transform the system, Cuban argues. He describes what he calls scale capture and says large integrated companies emphasize shareholder returns rather than lower costs when discussing technology. Because of that, he says his company avoids direct dealings with the major intermediaries he criticizes.

Yoo asks what healthcare employment might look like in five years. Cuban says organizations should continue hiring entry-level workers: people need to understand changing conditions, supply context, oversee models and accumulate implementation experience. He also thinks employees at every level should learn how to use AI.

Cuban wants executives to understand agents, model choices and the limitations of information available to a model. He distinguishes book knowledge from practical judgment, saying people still need to guide systems and understand consequences. In his view, experimentation can reveal customer benefits and new business opportunities.

As an example, he describes asking ChatGPT to propose a lower-priced cookie competitor tailored to people using GLP-1 drugs, with a bill of materials and sourcing information. He says the executive he showed was surprised by the result. Cuban uses the anecdote to illustrate a possible question to ask a model, not evidence that the proposed product was validated or manufactured.

Cuban predicts that individuals using agent swarms could build unusually valuable businesses. He also argues that agent-driven commerce will require changes in how information reaches agents, citing difficulty accessing social-media material and advertisements. He mentions an inactive experiment, agentfeeds.dev, intended to make post transcripts available to agents, while acknowledging unresolved trust issues.

Cuban sees a major opportunity for AI to help patients understand healthcare costs and documents such as explanations of benefits. He argues that families facing financial uncertainty need better information about available assistance and that hospitals should make their programs easier to find and use.

He suggests asking about nonprofit hospitals' financial assistance programs before a crisis. He also speculates that some people might compare such assistance with saving money they would otherwise pay in premiums. Eligibility and coverage vary, and this is his personal speculation, not a recommendation to forgo insurance or a guarantee that assistance will cover care.

Cuban imagines a service that helps people find and ask hospitals about assistance programs. He argues hospitals rely too heavily on insurer networks to bring them patients and could communicate directly about available options. Better technology and contracts, he says, matter less if patients cannot understand and use them.

He compares the current learning curve for AI with the early days of spreadsheets. In his view, showing people concrete benefits and ways to reduce uncertainty can help them become more comfortable with the tools.

Torenberg asks how frontier AI labs might change healthcare despite siloed information. Cuban predicts that, as models and algorithms improve, the cost of designing treatments tailored to individual biology could fall dramatically. He explicitly acknowledges that his own company has a commercial interest in this vision.

Cuban imagines starting with individualized over-the-counter medicines and later moving to other drugs. He suggests models might eventually use genetic information and other data to tailor treatments. This is a speculative future scenario; the interview does not establish clinical safety, effectiveness or regulatory approval for such a service.

He proposes that AI providers might eventually include personalized medicine in a subscription as computing costs fall. He also raises genetic-data privacy and says Cost Plus wants its manufacturing pods to produce individualized medicines. His projected costs, capabilities and timeline are not demonstrated results.

Yoo suggests that better-tailored drugs could change care delivery by reducing some trial and error around treatment. Cuban responds that people would still need doctors to observe symptoms, ask questions and interpret context, especially when a new treatment has unexpected effects. He does not expect AI to replace physicians.

Cuban anticipates that medical education could change: doctors may need to work with evidence tools, interpret model outputs and explain them to patients. He also raises a potential intellectual-property problem. Owners of valuable medical knowledge, he argues, may charge for access rather than make it freely available for model training or retrieval.

Yoo agrees that doctors will remain important and thinks more accessible advice could increase demand for actual care. She imagines a greater role for physicians as scientists who gather new information from individual patients. The discussion also turns to wearables and more continuous measures, such as blood pressure, as a possible source of longitudinal data.

Cuban envisions people acting as translators between AI and individuals: identifying what is unusual about a person's circumstances, providing context to a system and explaining its output. He doubts that a humanoid robot will necessarily be a comfortable substitute for a person in that role.

Asked about electronic health records, Cuban says he does not know enough to predict how the market will change. He notes that his company partners with Epic and supports giving doctors pharmaceutical price information when prescribing, but avoids claiming expertise about how the EHR business might be disrupted.

The hosts ask whether healthcare will continue to consume a large share of economic output. Cuban expects it may, but argues the value could improve if spending helps keep people healthy rather than only treating illness. His expectation of greater productivity is a forecast.

Torenberg asks hypothetically what Cuban would recommend to a president on healthcare. Cuban's first proposals are to make contracts involving taxpayer money public without redactions, and to let patients count a lower-priced purchase toward their deductible. He also wants nonprofit hospitals to disclose detailed financial and procurement records.

Cuban proposes a different financing model: people would deposit an amount comparable to an insurance premium into an account they own. If they need care and the balance is insufficient, a bank would lend up to a threshold, backed by a government guarantee, with stop-loss protection above that level. Repayment would come from payroll; unused savings and interest would remain with the person at retirement.

He compares the proposed guarantee with government-backed loans in other sectors. Cuban believes this could change the role of insurers and give people more control over spending, subject to limits on eligible care. It is a hypothetical policy proposal; he does not establish its feasibility, cost, distributional effects or protection for every patient.

Asked how AI companies should address public skepticism, Cuban criticizes the industry's communication. He argues that social media algorithms produce different information environments for different people, and that traditional advertising and position papers are insufficient to reach younger audiences.

Cuban says frontier companies should use AI to understand audiences and develop promotion suited to those feeds. He links opposition to data centers partly to uncertainty about AI, while also criticizing communities' negotiations over data-center deals. These are his interpretations of public attitudes and his proposed communications strategy.

He calls for better public and policymaker understanding of social media's influence, arguing that people no longer necessarily read the same news or see the same feed. He describes looking at his son's feed as an illustration of how individual interests shape the material presented.

Yoo asks where entrepreneurs still have overlooked opportunities. Cuban answers that the opportunity extends across industries. He divides AI users into those who use it to learn and those who use it to avoid learning, and encourages founders of all ages to experiment with questions and problems.

Cuban thinks many entrepreneurs remain uncomfortable talking to AI, whether by voice or text. He compares this hesitation to early spreadsheet users checking calculations by hand. He says people should still check outputs, and describes comparing several models' responses when reviewing contracts.

Cuban predicts unfamiliar businesses will emerge as people gain confidence with AI. He thinks Silicon Valley's culture may make some founders more comfortable trying ideas, but expects opportunities far beyond the region. The challenge, he says, is being willing to ask questions and act on what one learns.

His closing advice is to ask a model about many kinds of problems and experiment even when the interaction feels strange. He describes putting two phones next to each other to hear systems converse. Cuban sees discomfort as a possible signal of a new idea, and the hosts thank him for the discussion.