In 30 seconds
Chamath Palihapitiya expects AI to create a handful of companies spanning applications, models, data centers and power, and argues that incumbents should invest or acquire rather than rely on buybacks. David Friedberg sees open models as a check on concentrated control, while Jason Calacanis expects wider use of AI agents to expand demand. The hosts disagree about markets: Palihapitiya sees inflation and uncertainty as reasons for caution, while Calacanis points to potential productivity gains.
- 1AI may reward companies that own infrastructure through applicationsPalihapitiya expects a small number of integrated companies to emerge and sees land and power as critical constraints.
- 2Cash generation may matter more than the software or services labelPalihapitiya argues investors should focus on margins and cash flow as AI changes business models.
- 3A strong AI business can still face a weak marketPalihapitiya sees inflation, long-term interest rates and uncertainty as risks to valuations; Calacanis expects rising AI use to expand demand.
- 4Friedberg sees the biggest opportunity in problems not yet solvedHe says scientific and engineering advances could be more valuable than using AI only to reduce existing costs.
- 5A proposed disclosure rule sets off a debate over open modelsThe hosts oppose restrictions on Chinese open models, arguing that local use and competition matter for company control and the broader market.
- 6The hosts disagree over how much can be inferred about NvidiaPalihapitiya and Friedberg say too many facts remain unresolved; Calacanis offers his own speculation.
Takeaways
- Palihapitiya expects eight to ten AI conglomerates to integrate infrastructure through applications.
- Friedberg argues that open models let businesses retain control and help prevent AI from being concentrated in a few hands.
- Palihapitiya thinks strong technology adoption does not by itself establish that stock prices have bottomed.
- Friedberg sees potentially greater AI value in new scientific and engineering solutions than in cost savings alone.
Read the edited text by chapter
The case for integrated AI conglomerates
Palihapitiya predicts a handful of companies will span applications, models, computing and power. He argues that incumbents must invest or acquire to compete.
Chamath Palihapitiya expects AI to produce eight to ten large conglomerates over the next decade. In his model, each would span applications, the models that power them, data centers, land and energy. He sees Palo Alto Networks' purchase of an open-model team as an example of companies reaching beyond their existing businesses.
Palihapitiya says the market rewarded a security company for buying an AI business, suggesting investors want established firms to make bigger bets. He expects Anthropic and other major players to move both toward infrastructure and toward applications, rather than remain in a single layer of the market.
He thinks demand for different kinds of computing hardware will weaken Nvidia's exceptional pricing power. Land and electricity, in his view, will be important constraints on the companies trying to build this vertically integrated stack.
Palihapitiya says the stock market is effectively deciding which established companies can become integrated competitors. Firms that do not make the transition, he thinks, could see capital move elsewhere even if their existing businesses remain profitable.
Companies with ample cash flow that neither invest in models nor build infrastructure risk losing investors to those taking greater risks, he argues. Jason Calacanis asks what a software product means when agents can create features on demand.
What Anthropic’s offering means for business strategy
The hosts discuss the reported offering and how AI may change business categories. They disagree about when companies should buy back shares rather than invest.
The hosts discuss Anthropic's public offering, which Calacanis says drew demand at a $2 trillion valuation. He welcomes having public financial information on a major model provider alongside data from cloud and chip companies.
Palihapitiya argues that labels such as software and services will matter less than financial performance as AI changes business models. He says he would prefer a services business with exceptionally high margins and cash generation over a software company with weaker economics.
Calacanis asks Friedberg whether a company spanning all these layers can adjust where it earns its profits, giving software away while making money on infrastructure. He compares the argument to the way several large technology groups combine businesses.
Calacanis suggests conglomerates can shift profits among hardware, infrastructure and applications. David Friedberg compares the AI transition with the internet forcing businesses to rethink how they operate, and expects industry boundaries and supply chains to change as digital intelligence spreads.
Friedberg says relatively few people and companies yet use the tools heavily, leaving room for adoption. Looking at agriculture, where he works, he sees processes that could be redesigned from first principles rather than merely digitized.
Palihapitiya criticizes stock buybacks in a period of technological upheaval: he thinks executives should be using cash to build or acquire new capabilities. Calacanis says buybacks have long supported earnings per share and share prices, while acknowledging the appeal of infrastructure investment.
Calacanis says political conditions may affect the ability of big companies to make acquisitions. Friedberg responds that any proposed deal must be compared with the value of buying the company's own shares.
Friedberg says buying stock can make sense if it is undervalued, and an acquisition should be judged against the value a company can create with that capital. Palihapitiya argues the more important question is whether a deal opens a strategic business line over several years, not only whether it adds earnings immediately.
Palihapitiya says some shareholders still focus too much on whether an acquisition immediately increases earnings. He wants managers to consider whether it creates a position in a new business over the years ahead.
Palihapitiya doubts incumbents fully understand what the next generation of applications should look like. He suggests buying smaller teams that do understand it, and says he would favor a CEO who explicitly commits to using cash flow to adapt to disruption.
Calacanis agrees that building data centers and the ability to create intelligence can be a sensible use of capital. Palihapitiya responds that firms may still need to buy their way into applications created by smaller teams.
Growth potential does not settle the valuation question
Friedberg sees room for Anthropic to grow while Palihapitiya worries about macroeconomic conditions. Calacanis expects far greater use of agents but is challenged on the timing of a market recovery.
Asked about Anthropic's valuation, Friedberg says he has heard its growth has exceeded internal expectations and sees substantial room to reach more business customers. He considers it strong in enterprise products, while treating open models as a possible competitive risk.
Friedberg speculates that models may be improving faster than humans can evaluate their safety. If a provider can repeatedly reduce the cost of a given level of intelligence, he says, it might build a hard-to-match advantage. He treats that as a potential outcome, not a proven result.
Friedberg says the potential for a model to improve its efficiency has been part of the appeal to investors. He still identifies open models as a competitive threat and expects major providers to continue competing for advanced uses.
Friedberg also distinguishes the business from its stock: the supply of shares available for trading, the demand that remains and investors' holding periods can all affect the price. He sees potential over a long holding period, while acknowledging that the market will establish the price.
He says an investor needs to distinguish long-term demand for shares from first-day buyers who may sell quickly. Even when a business is growing rapidly, trading supply and demand will affect where the market values it.
Palihapitiya agrees Anthropic could become one of the large integrated companies he predicts, and says new financing reduces its balance-sheet risk. But he does not claim to know the right valuation or to infer much from the first day's trading.
His near-term concern is the broader market. He cites persistent inflation, higher long-term interest rates, conflict and political uncertainty as reasons investors might reduce risk, even when a company reports strong earnings.
Calacanis lists debt, inflation, political turnover and possible regulation among the obstacles facing the market. His counterargument is that technology adoption in workplaces remains early enough to grow substantially.
Calacanis offers a more optimistic technology case. He expects use of AI to spread from startups to large organizations and says people who now spend minutes with the tools could eventually work with agents for many hours a day.
Calacanis forecasts a large increase in token use as workers coordinate multiple agents. He says he sees heavy use in startups he invests in, while Palihapitiya challenges whether faster adoption means stock prices have already bottomed.
Palihapitiya accepts that current AI usage may be far below its potential, but says that does not prove the market has reached a low. He expects higher rates to have mathematical consequences for the value of equities.
To explain his optimism, Calacanis imagines an employer spending far more on an employee's AI subscription if that worker becomes much more productive. He expects revenue per employee to rise; Palihapitiya says the market still wants evidence for those results.
Calacanis expects knowledge-work businesses to generate more revenue with each employee as AI spending increases. He describes this as his conviction based on observing startups, rather than a mathematical certainty.
New problems may matter more than cost cutting
Friedberg expects the greatest gains from solving previously intractable scientific and engineering problems. He also recognizes that uneven participation can create a political backlash.
Friedberg sees the largest opportunity beyond cutting costs in existing processes: applying AI to scientific and engineering problems people could not previously solve. He offers new medicines and materials as examples of potential value rather than completed breakthroughs.
Friedberg says calculating the return on using AI to produce the same output can obscure its potential for creating something entirely new. He contrasts incremental savings with pursuing a new drug or material.
Friedberg argues that new industries can draw workers and capital before older work disappears, so he is skeptical of a simple mass-unemployment narrative. He also acknowledges an unequal economy in which people who do not benefit from the new frontier may oppose it.
The hosts consider how an uneven distribution of AI's benefits might feed political opposition. Friedberg says people who see a frontier forming without sharing in its gains may prefer to stop it.
What disclosure rules could mean for open models
The hosts discuss a reported proposal to disclose use of Chinese models. They argue over access, security and the incentives behind domestic open-model investment.
Calacanis turns to a reported White House discussion of requiring companies to disclose their use of Chinese open models. He says no decision had been announced and mentions congressional letters to companies including Airbnb, DoorDash and Cursor.
Calacanis describes the proposal as a possible requirement to put disclosures on company websites. He asks why such an idea would surface at this moment, but says the proposal had not been decided.
Palihapitiya doubts a disclosure label would deter companies if the models reduce costs and let them control their own technology. He argues that policymakers need technical expertise to understand the implications of running an open model locally.
Palihapitiya argues businesses want to keep costs manageable and retain control over their information. In his view, refusing cheaper alternatives could hurt a company's competitive position more than a website disclosure would.
Friedberg opposes restrictions on open models, arguing they could concentrate control in a few companies and government agencies. He compares the choice with earlier open software projects that allowed other developers to build alternatives to closed systems.
Friedberg says open software can be copied and adapted, allowing developers to build competing products. He views open AI as a similar choice about whether individuals and businesses can use technology outside a small set of providers.
Friedberg says a model's geographic origin can be hard to determine when weights are posted anonymously. He distinguishes downloadable model weights from the software that runs them and argues companies should test the tools they incorporate in their products.
He says model weights are data used within a separate environment and maintains that users can inspect and test what they deploy. He opposes treating every model of a given geographic origin as inherently opaque.
Calacanis suggests a connection between the reported disclosure discussion, Nvidia's leadership dispute and competition with closed-model developers. Palihapitiya offers a different interpretation: arguments over China involve power and ideology, and the outcome depends on whether policymakers think open models advantage China.
Calacanis sees a possible political fight around open and proprietary models. Palihapitiya instead suggests that policymakers differ over whether access to open models helps or harms China, and emphasizes the role of ideology.
Palihapitiya asks why well-funded American technology companies did not establish a stronger domestic open-model alternative sooner. Calacanis points to incentives that rewarded researchers at proprietary labs; Friedberg adds that private companies promising closed products could attract capital more easily.
The hosts debate incentives for researchers. Calacanis points to exceptionally large compensation packages at frontier labs, while Friedberg thinks it has been easier to raise private capital for a closed model with a clear commercial plan.
Friedberg suggests companies behind in model development could have an incentive to make models broadly available and earn money elsewhere in the stack. Palihapitiya agrees that the strategy sounds plausible but asks why earlier investment did not yield stronger results.
Palihapitiya asks why large American companies with money and technical staff did not spend more on open alternatives earlier. The others suggest some tried, but the work has proved difficult and takes time.
Palihapitiya wonders whether researchers who have already earned substantial money at frontier labs might leave to work on alternatives with a more nonprofit orientation. Calacanis says that is possible, while neither treats it as a certain forecast.
Calacanis suggests that well-funded companies seeking to catch up could back open models while earning revenue from chips or applications. Palihapitiya says the logic seems clear in hindsight, yet he still questions why the outcome has fallen short.
Nvidia’s leadership dispute and the limits of speculation
Calacanis describes the company’s recent situation and a public call for employee action. Palihapitiya and Friedberg say important facts are still unresolved, while the hosts remain divided over market risk.
On Nvidia, Calacanis describes a share-price decline, strong reported earnings and uncertainty following a leadership change. He also says the company faces legal scrutiny and employee concern. The hosts differ on whether outsiders have enough information to explain what happened.
Calacanis says Nvidia's acting leadership has emphasized more selective financing and describes changes to its approach to China. He also discusses reported legal investigations and employee reactions; the other hosts urge restraint about the reasons behind the leadership change.
Calacanis plays an excerpt in which former Uber chief Travis Kalanick says a petition is useful but suggests an employee walkout as a stronger way to support Jensen Huang. Calacanis presents it as one response from people in the industry.
Palihapitiya calls Huang an exceptional founder but argues a public company's board cannot simply restore him while serious questions remain unresolved. He says upcoming congressional hearings are important and criticizes speculation based on incomplete information.
Palihapitiya says the board's position also affects the company and its shareholders. He expects coming hearings to matter and says a public campaign cannot resolve unanswered questions of governance and possible legal exposure.
Friedberg says the people and institutions directly involved have not supplied enough information for him to determine why Huang left. Calacanis wants to form a view from reporting, but Friedberg refuses to treat those reports as a substitute for first-hand evidence.
Friedberg rejects drawing conclusions from media reports when the people directly involved have not supplied the evidence he wants to see. Calacanis responds that discussing possibilities is part of a podcast, even when the outcome remains uncertain.
Calacanis speculates about the possibility of intermediaries reselling computing equipment, while explicitly offering it as a guess. Palihapitiya says the lack of established facts is precisely why he is withholding a conclusion.
Friedberg returns to his concern that AI could become controlled by a small number of companies. Whatever the cause of Nvidia's situation, he argues that people across political affiliations should favor competition and access to open models.
Palihapitiya points to the day's gains in Anthropic and Palo Alto Networks as market signals, while remaining cautious about the wider market. He reiterates that investors are rewarding companies making large bets on the technologies they expect to matter.
Calacanis and Palihapitiya distinguish their different approaches: Calacanis is optimistic about products and long-term adoption, while Palihapitiya says he is reducing risk despite recognizing that the market rewards some large AI investments.
He identifies unresolved economic pressures, the potential effects of Nvidia changing its financing practices, and other disruptions as reasons for caution. Calacanis counters that his longer-term assessment rests on products, management and market demand. The hosts close without resolving their disagreement about the market's direction.
More from this channel
1:02:06EN
1:24:55EN
1:33:03EN
1:25:46EN
55:39EN
1:43:22EN