@berryxia: I just finished listening to Patrick O'Shaughnessy's latest podcast—absolutely mind-blowing! The guest was Anthropic's CFO, Krishna Rao, in his first long public conversation. When he joined two years ago, the company's annualized revenue run-rate was only $250 million...

I just finished listening to Patrick O'Shaughnessy's latest podcast—absolutely mind-blowing! The guest was Anthropic's CFO, Krishna Rao, in his first long public conversation. When he joined two years ago, the company's annualized revenue run-rate was only $250 million. Today, it has skyrocketed to $30 billion. Along the way, he also led the fundraising of nearly $75 billion. But what really gave me chills is the real power he holds: the procurement, allocation, and dynamic scheduling of all of Anthropic's compute. How to split Trainium, TPUs, and GPUs, whether to prioritize training or inference, which project gets today's resources first—all of it is ultimately his call. He has already signed over $100 billion in AI computing procurement commitments. Suddenly, I realized the most brutal core thesis: In the top AI companies of 2026, the person truly sitting in the driver's seat and controlling the whole show might not be the chief scientist. It could be the CFO. Because model capabilities are rapidly converging, and compute is the true bottleneck and scarce resource. Whoever gets more compute and uses it more ruthlessly wins. This podcast is packed from start to finish—from decision-making under the "Cone of Uncertainty," to why returns on frontier intelligence keep increasing, to how Anthropic internally uses Claude to write code—there's an insane amount of valuable content. If you want to truly understand how AI companies fight internally, you have to listen to this episode in full. What do you think will be the biggest variable determining the survival of AI companies in the next phase: model capability, or the ability to acquire and allocate compute?