DOGE kills its first bill, Zuck vs OpenAI, Google's AI comeback

Chamath Palihapitiya said SpaceX already uses stablecoins to move Starlink revenue out of smaller foreign markets without taking wire or currency-exchange friction. He argued that the same rails could eventually challenge card networks charging roughly 3%, making payments—not speculative tokens—the clearest near-term crypto use case.

The episode framed DOGE’s first real source of power as public pressure rather than formal government authority. Aaron Levie said Musk and Ramaswamy demonstrated that exposing disputed provisions online could make congressional support uncomfortable, while AI tools now let ordinary readers digest enormous bills far faster than before.

Levie described a government rule that forced one unnamed agency to use contractors costing roughly 2.5 times what comparable employees would cost. He said the arrangement also reduced accountability, offering a concrete example of how procurement rules can increase spending while producing worse operational control.

OpenAI’s latest financing created a major structural deadline: the episode says its $6.6 billion round at a $157 billion valuation requires a for-profit conversion within two years or investors can seek their money back. Musk is challenging the conversion in court, while Meta separately asked California’s attorney general to oppose it.

Palihapitiya and Levie expect foundation models to become increasingly interchangeable, with companies routing different tasks across several providers instead of committing to one. Levie argued that research spreads rapidly and open-source models cap pricing, eventually pushing token economics toward compute cost plus a modest margin.

Their deeper disagreement was over whether AI destroys the existing software market or expands it into work previously performed by people. Palihapitiya predicted today’s roughly $5 trillion software-and-IT economy could shrink toward $500 billion; Levie countered that AI agents could absorb service budgets and create entirely new software categories.

Friedberg said Ohalo already trained non-programmers with Cursor and ChatGPT to build internal software, but deploying it safely remains the harder problem. The group agreed that production integration, permissions, security, testing and regulatory approval—not generating interfaces or code—remain major barriers to companies replacing established enterprise systems themselves.

Google emerged as the episode’s strongest competitive threat to OpenAI because its AI progress now combines models, infrastructure, browser control and visual data. Friedberg argued Google’s long-built infrastructure and data advantages are compounding, while Levie said the company has shifted from protecting its existing business to launching aggressively because losing the AI race became the greater risk.