Reading the Anthropic S-1, with Zvi Mowshowitz

Patrick McKenzie and Zvi Mowshowitz use Anthropic’s IPO filing to examine how the company presents its governance, mission, risks, and growth. Its structure gives a single Class F share 50.1% of voting power, preserving founder control even as public ownership expands. The filing discusses existential risk cautiously and indirectly, while also describing rapid frontier progress and “closing the loop” - using AI to accelerate AI development - as central to the business. The speakers see tension between safety commitments and competitive pressure, as well as between candid risk disclosure and lawyers’ incentives to limit corporate liability. They also highlight an uncapped 3-to-1 employee donation match, an unusually costly benefit.

The discussion weighs Anthropic’s growth prospects against the scale and uncertainty of its investments. Slower third-quarter revenue is interpreted as evidence that the company lacks GPUs, not customers; its forward compute commitments approach $900 billion. The speakers consider whether token demand can sustain that spending, whether AI services will become commoditized, and whether the industry’s financing resembles a bubble or dot-com-era circular deals. They discuss shifting compute among training, inference, and subleasing, and how much future capability depends on scaling versus broader adoption. The conversation also explores AI-assisted cybercrime, including a reported $5 billion Binance theft, and asks where responsibility lies when AI tools enable attackers. Overall, the filing offers a lens on the business, governance, and safety trade-offs facing frontier AI companies.