“It felt like the Nasdaq was selling off. And at the same time, what was happening in AI was, I think, the most extraordinary moment in the history of capitalism”
AI compute demand is an unprecedented exponential, and AI stocks got historically cheap, so lean in.
“Anthropic clearly has a dramatically lower cost per token than OpenAI. They just do. And you could just see that in the amount of money that they have burned…”
Anthropic is dramatically more capital-efficient than OpenAI, and compute-constrained, so its real run-rate is far higher than its mark implies.
“I think the watts shortage will probably begin to alleviate '27, '28. And then I think orbital compute will really solve that.”
SpaceX-style racks in space turn orbital compute into a real inference business that bypasses terrestrial energy and zoning limits.
“The TeraFab, it's a SpaceX... joint venture to build the world's largest fab here in America. And I think they're going to be successful. One, they have a…”
TeraFab will succeed as a US foundry via Intel knowledge, semi-cap A-teams, and Elon's talent magnet.
“AI is just shifting from all-you-can-eat to pay-by-the-drink... people really like to use AI, and particularly now that one person can have 100 agents working.”
Frontier-token economics plus closed APIs let Nvidia and labs push pricing, driving OpenAI and Anthropic past $200B ARR this year.
“today I think probably Trainium is doing the best. Nobody's a better GPU, but Trainium is, I think they're, you know, they're tugging on Superman's cape.”
Trainium 3's switch scale-up network makes it the best non-GPU accelerator, tugging on Nvidia's lead in MoE inference.
“My firm was a venture investor in Cerebras. What Cerebras has done is something hard and fundamentally different. Wafer-scale computing... that architectural…”
Cerebras built a hard, differentiated wafer-scale architecture that nobody can copy, a strong venture position into its IPO.
“I don't know anyone like me who's not really bullish on DRAM.”
DRAM is mispriced cheap versus semi-cap equipment; the cross-sectional valuation gap can't hold.
“something like NVIDIA, which is still... in early April is essentially as cheap as it gets relative to the market, like in the last 10 or 12 years... It's…”
Nvidia stays dominant: TSMC supply discipline, deep startup ties, and a valuation that prices in unfathomable share loss.
“Astera is a stock I've been close to a long time. There's a lot of bears on that. I love that. Great. You know, I first invested in the Series C.”
Astera is miscategorized as a copper loser; its switch business sits on both sides of every connection.
“Google... they had that TPU advantage, which is now gone. The reason I think they're still in a great position is just they have the most compute of…”
Google's position stays strong on the largest compute installed base and unique data despite losing its TPU cost edge.
“You got to give Zuckerberg immense credit, what he's done in terms of making Meta an AI-first company internally. And I do think he is the only one of those…”
Meta is now a genuine AI-first company; Muse's strong debut and aggressive talent buys improved its position.
“Amazon, I think, is in a really strong position because of Trainium. You're going to see, like, real P&L efficiencies from robotics over the next 18 months in…”
Amazon is well-positioned: Trainium scaling, robotics P&L gains, and deep startup engagement.
“the decision Satya is making, which the market has punished him for, but I think is the right decision, is we are going to use our compute... to make our own…”
Microsoft's choice to use its own GPUs for products over reselling to OpenAI is right but suppresses the stock near-term.