The Frontier

Dario Amodei published an open letter calling for frontier AI labs to pace their public releases to prioritize safety. Brett explains that closed-lab capabilities are surging due to massive cost declines, creating liabilities if unchecked agents disrupt infrastructure.

David Sachs criticizes top-down AI regulation, arguing developers should bear liability for their models rather than seeking regulatory capture. Dario Amodei sparked controversy by suggesting a waiver of the Sherman Act to allow frontier coordination on releases.

Brett suggests closed-weight labs want open-weight models to cause infrastructure disruption first. This scenario allows firms like OpenAI and Anthropic to sell cloud-based, highly monitored AI defenders to desperate enterprises, securing both revenue and control.

Brett dismisses claims that AI will eliminate massive swaths of the workforce. He notes that net employment in AI-impacted industries has grown by one million jobs relative to baseline expectations since the launch of ChatGPT.

Brett projects that a ten-times productivity increase for knowledge workers could expand the software market dramatically. He estimates that enterprises would pay ten percent of this value to AI vendors, yielding thirty trillion dollars in potential revenue.

Meta launched Muse, a fully agentic consumer application that integrates with banking, shopping, and email services to perform end-to-end tasks. Nick used the app to scan state databases, successfully recovering unclaimed funds.

Meta plans to monetize Muse by taking a transaction fee on commerce instead of charging for subscriptions. To address privacy, Muse routes browser activity through cloud-hosted virtual machines, isolating sensitive data like banking credentials connected via Plaid.

Brett claims SpaceX's $100 billion Louisiana investment is one of history's largest infrastructure projects. The site is geographically critical because launching AI satellites requires flying due south, and the rockets require local natural gas to produce methane fuel.

Brett models a 100% annualized return on capital for SpaceX's tenth Starlink launch. Returns on its hundredth AI satellite launch will drop to the high 20s as SpaceX prioritizes internal research and development over early commercialization.

Sam notes that California's high-speed rail project will cost an estimated $125 billion for a quarter-mile of track. Brett argues that high yields from AI infrastructure will dry up financing for these government white elephants as capital flows to higher returns.

Brett estimates that SpaceX can achieve a 75% internal rate of return on a $30 billion investment in terrestrial data centers. This high rate of return will force other corporate and government borrowers to raise their interest rates to compete for capital.

Brett states that Anthropic and SpaceX estimate the AI enterprise knowledge work market at $30 trillion. This assumes AI tools 10x knowledge worker productivity, and enterprises pay software providers 10% of that value while retaining 90% of the surplus.

Sam and Brett point out that AI model costs are declining by over 99% annually. When developer platform OpenRouter discounted specific model prices by 50%, throughput volumes immediately surged 14-fold, demonstrating highly elastic demand for AI tokens.

Brett warns that massive AI infrastructure spending could crowd out traditional borrowers in the $10 trillion to $15 trillion corporate bond market. Standard businesses will face higher refinancing rates as BAA corporate bond yields break out past 6%.

Elon Musk plans to invest over $100 billion in a chip fabrication facility called Terra Fab. Musk claims that scaling the plant to a capacity of one terawatt annually will eventually require $1 trillion in total capital.

Figure AI launched a headset device to pay users to record daily household tasks. Brett notes this addresses a critical data shortage in humanoid robotics, although Sam estimates generalizable robot-as-a-service commercialization is still 10 years away.

Sam argues that robo-taxis will serve as a vital political buffer for AI data centers. Because the average American adult spends an hour a day driving, the immediate lifestyle and safety improvements will win public support for broader technology installations.