Why Could Open AI Models Beat Closed AI Systems?

TL;DR

Open AI models could gain ground because companies may not trust frontier providers with proprietary data when those providers also enter the application market. Jason Calacanis argues that falling AI prices, shrinking differences between models, and local deployment will reduce the model layer’s value, shifting competitive advantage toward applications, operating systems, robotics, and founders willing to experiment.

Transcript

GPI. Next topic. So, we are joined by Jason Kakamis, uh, angel investor from Silicon Valley and the co-host of all podcast. >> Um, >> number one podcast in the world. >> Yeah. You said that in the beginning of show, right? >> Yes. I was manifesting. >> Yeah. And you have >> had a dream. >> Had a dream. >> Had a dream. It'll be number one >> this we... Read More

Key Insights

• Open models are presented as strategically important because companies in education, life sciences, biotechnology, and finance may expose valuable internal intelligence when they send proprietary data to frontier-model providers.
• Trust is the central concern for application companies because large AI providers are spending heavily on infrastructure, losing or investing tens of billions of dollars, and may eventually need to compete in the application layer.
• Anthropic’s relationship with Cursor illustrates the perceived platform risk: Calacanis recounts that Cursor used Anthropic’s software while Anthropic developed an internal coding agent and later released Claude Code.
• Closed-model providers are expanding beyond foundational models into applications, with Claude Design, voice capabilities, website design, and coding agents identified as areas where existing application companies may face direct competition.
• Application companies are reportedly preparing to move toward open-source systems because they want greater control over their technology and data. Calacanis says these projects are not publicly acknowledged but are already being developed.
• AI economics are expected to change rapidly, with the program’s chapters forecasting annual price declines of 90 percent and diminishing differences among models, developments that would reduce the standalone value of model access.
• Local AI deployment could weaken dependence on cloud infrastructure because the discussion anticipates a period when AI can run at home, making data centers less central to how individuals and businesses use intelligent systems.
• Japan’s opportunity is linked to robotics and entrepreneurial risk-taking, including whether the country can control the intelligence inside robots, produce globally competitive companies, and treat a startup’s first year as an experiment.

Questions & Answers

Q: Why could open AI models beat closed AI systems?

Open models could win because businesses may prefer technology they can control, fork, and operate without giving proprietary information to a powerful external provider. Calacanis argues that closed-model companies face pressure to move into applications after investing heavily in infrastructure. That creates a conflict when the same provider supplies intelligence to an application company and later competes with it.

Q: Why does Jason Calacanis distrust OpenAI and Anthropic?

Calacanis says large AI companies cannot automatically be trusted because customers provide them with valuable data and operational intelligence while using their models. He argues that these providers are valued extremely highly, spend hundreds of millions of dollars on infrastructure, and are losing or investing tens of billions. Those economics may push them to compete directly with their own application customers.

Q: How does proprietary company data create AI platform risk?

The risk arises when a company in education, life sciences, biotechnology, finance, or another specialized field sends internal data to a frontier model. Calacanis says the model provider learns as customers use its system, potentially gaining intelligence about their businesses. If that provider later enters the same application market, the customer may face a competitor that supplied its underlying technology.

Q: What does the Cursor and Claude Code example show?

Calacanis uses Cursor to illustrate how an application company can become vulnerable to its model supplier. According to the account presented, Cursor was working with Anthropic and using its software while Anthropic had a coding agent under development internally. Anthropic then released Claude Code, leaving Cursor to compete with a company that had previously served as a foundational technology provider.

Q: Why are AI companies moving into the application layer?

Calacanis links application expansion to the enormous cost of building frontier AI. He says the major providers are spending hundreds of millions on infrastructure and losing or investing tens of billions of dollars, with hundreds of billions potentially required later. To support those commitments, they may pursue products such as coding agents, design tools, voice systems, and website-building applications.

Q: How could falling AI prices change the model business?

The program forecasts that AI prices could decline by 90 percent each year while differences among models become smaller. Under that scenario, access to a particular model would carry less standalone value because comparable intelligence would be widely available at much lower cost. Competitive advantage would consequently move toward useful applications, customer relationships, controlled data, and broader operating platforms.

Q: When could businesses run AI without the cloud?

The discussion anticipates an era when AI can operate at home, reducing dependence on cloud services and making centralized data centers less dominant. The provided material does not specify a date or technical threshold for that transition. Its central claim is that cheaper, increasingly similar models and local execution could change where intelligence runs and who controls the associated data.

Q: What should Japan focus on in the AI era?

The interview frames Japan’s opportunity around controlling the intelligence used in robots, creating companies capable of competing globally, and becoming more willing to make consequential bets. It also presents entrepreneurship as increasingly accessible, with companies potentially created in only days. Founders are encouraged to regard the first year as an experiment centered on progress, learning, and repeated execution.

Summary & Key Takeaways

• Jason Calacanis argues that open models will eventually prevail because businesses cannot safely assume that large frontier-model providers will remain neutral infrastructure partners. Companies using closed models may provide valuable data and market knowledge to vendors that are spending heavily and therefore have strong incentives to compete directly at the application layer.

• The interview connects this conflict to products including Claude Code, Claude Design, voice systems, website builders, and coding agents. Calacanis says application companies are quietly developing projects based on open-source models that they can fork and control. Nvidia’s NeMo is cited as another example of an available open-source model project.

• The broader forecast is that AI prices could fall 90 percent annually, models could become less differentiated, and capable systems could run at home without cloud dependence. Calacanis also discusses employment disruption, robot intelligence, rapid company creation, first-year experimentation, difficult founders, Japan’s opportunity, and the importance of making ambitious bets.