How does Benchmark rationalize not being in any of the core model providers?
"It f**king sucks. It's a complete and utter failure on our part.
You can't be in a situation where you have a chance to make a 30x on scaled capital in four or five years and you don't do that, that's always a failure.
It stings especially when all your friends are sending you their implied look-through ownership of Anthropic, OpenAI and SpaceX, eye-watering numbers." @EverettRandle
”This is a Rubicon moment in the history of AI.
It's the first time that the US has regulated an AI model based on its capabilities.”
@EverettRandle joins Harry and @rodriscoll (in person for the first time ever) to shoot the s*** on:
- SpaceX Soars to $2.7T - Anthropic's Fable Banned by US Government - Wix and Adobe Hit All-Time Lows - Mistral Raising at $20B and The Case for Sovereign Models - Fin Acquired by Salesforce for $3.6B
My notes below:
1. Everyone Misunderstands the SpaceX Valuation
SpaceX’s $2 trillion-plus valuation does not make sense on the P&L alone, but that is the wrong lens. Elon’s playbook is selling long-dated call options to the market: full self-driving, Starship, orbital data centers, Mars. He has delivered enough of them that the market keeps buying in.
2. Elon’s AI Strategy Has Been the Best of Anyone
While everyone else spent two years debating whether to “do something in AI,” Elon built two data centers, bought Cursor to fill the Colossus revenue gap, and signed $2 billion per month in compute contracts with Google and Anthropic. By September, xAI’s AI revenue run rate will exceed SpaceX’s entire launch business.
3. Why Elon’s Biggest Advantage Is Cost of Capital
No bank would lend $24 billion with no contracts to build Colossus in under a year. Elon could do it because his cost of capital is uniquely low, earned through decades of delivering on impossible bets. That lets him make asymmetric wagers no one else can, reinforcing the track record that keeps his cost of capital low.
4. Why Anthropic Creates a Sovereignty Problem
The US restricting Claude’s export based on capabilities is a Rubicon moment. If that logic extends to OpenAI and Google, governments may begin gating access to intelligence itself. Who gets access to ASI when it arrives: NATO allies, select citizens, approved institutions?
5. Any Liquidity for Pre-AI SaaS Companies Is Top-Decile Performance
The Intercom/Fin $3.6B sale to Salesforce for is a masterclass in turning a zombie asset into real capital. Growing at 7% with no AI story, the equity was essentially worthless. If your board is not making this the top agenda item and asking, “Is this relevant for us?” that is a failure of governance.
6. We Have a Crisis of Open-Source Models in the Western World
Outside China, there are essentially no world-class open-source models. The talent and focus required to be best-in-class across pre-training, mid-training, and post-training are scarcer than people think. For many countries, the realistic path may be fine-tuning a Chinese open-weights model.
7. How Does Benchmark Rationalize Not Being in Any of the Core Model Providers?
They do not. Ev was blunt: “It f**king sucks.” Missing a 30x on scaled capital in four to five years is a failure, full stop, regardless of how well the rest of the fund performed. It is one of the most honest VC admissions you will ever hear.
(links below)
