Why Chamath Thinks Nvidia's Peak Earnings Era Is Already Over

As seen on the 24/7 Wall St. homepage on November 20, 2026.

Anthropic Goes Public, Nvidia in Crisis, Markets Tank, Open Source Crackdown?

  • Chamath: 9-10 vertically integrated AI mega-corps will own everything from data center land to apps
  • Palo Alto Networks' Applied Compute deal cited as proof security firms are buying their way into models
  • Anthropic priced at a $2 trillion valuation, raising over $107 billion, topping SpaceX's $85 billion deal
  • Nvidia revenue and net profit doubled to $116B and $65B, yet the stock fell after Jensen Huang's ouster

Chamath argues Nvidia's over-earning era is ending for reasons unrelated to Huang's exit, as compute spreads across formats and land and power become the real constraint. He also says cash-rich incumbents that refuse to enter the model or data center game get sold to fund those that do. Jason counters that token demand is still early.

Continue ReadingShow less

Chamath's core argument on the All-In podcast is structural: as compute spreads across different hardware formats, the conditions that allowed Nvidia to capture outsized profits are fading. Land and power are becoming the binding constraints on AI infrastructure buildout, which shifts pricing leverage away from GPU makers toward whoever controls those physical resources.

That thesis gets sharper when you look at what Nvidia actually reported. Revenue and net profit both doubled, yet the stock still fell. For Chamath, a stock declining on numbers that strong is itself evidence that the market is beginning to price in the end of the over-earning window, whatever the reason given in headlines.

The broader framework Chamath is building around this is a prediction that 9 to 10 vertically integrated AI mega-corporations will come to control the full stack from data center land all the way up to consumer applications. Palo Alto Networks' acquisition of Applied Compute is the kind of move he points to as confirmation: a cash-rich incumbent buying its way into the model and infrastructure layer rather than being left behind.

Sponsored

Anthropic's IPO at a $2 trillion valuation, surpassing SpaceX's raise, sits at the other end of this argument. It shows how much capital is chasing the companies that are already inside that vertical stack. Chamath's read is that incumbents sitting on cash but refusing to participate in models or data centers will eventually be sold to fund those that do.

Jason pushes back with the simplest counterargument: token demand is still early, meaning the total volume of AI inference the world will eventually consume is large enough to keep feeding Nvidia's revenue even as competition in compute formats grows. That tension, between a maturing GPU cycle and an AI demand curve that has not peaked, is the core debate the episode is working through.

Mentioned: AMD, AAPL, MSFT, NVDA, META, METAM, PANW, GOOGL, GOOG, GGLBP, GOOGN, GOOGM, GGLAP, CRSR