The Bet Behind the Bet: What Leopold Aschenbrenner’s Portfolio Tells Engineers About AI’s Real Constraints Most people covering AI are watching the model benchmarks. Aschenbrenner watched the power grid. That difference in focus is what turned $225 million into roughly $5.5 billion by the end of 2025, and then into $13.67 billion by Q1 2026….
Jensen Huang Ate Noodles in a Hutong. That Tells You Everything. There’s a moment from the Trump trade delegation visit to Beijing in May 2026 that I keep coming back to. While diplomats and executives performed the usual theater of geopolitical trade talks, Jensen Huang, the CEO of the company that makes the chips powering…
Jensen Huang Ate Noodles While the Trade War Raged. He Was Right To. There’s a photo circulating from Beijing this week. Jensen Huang, CEO of the world’s most valuable chip company, sitting in a hutong alleyway eating zhajiangmian from a place called Fangzhuan workshop. Not in a conference room. Not on a delegation panel. Not…
China Just Bet $8.4 Billion That the Future of AI Compute Is in Orbit Most of the AI compute conversation right now is about H100 allocations, inference costs per token, and who has the largest training cluster. That’s a real fight. But China just signaled that it’s also playing a completely different game, one that…
Inference Cost Is an Architecture Problem Most AI engineers I know have never seriously thought about inference cost until it destroyed their unit economics in production. I’ve watched it happen more times than I’d like to admit. The pattern is always the same: weeks of benchmarking, heated Slack debates about GPT-4o versus Claude versus Gemini,…
Terafab Project: Four Words, Seven Days, and What Musk Might Actually Be Building On Saturday, March 14th, Elon Musk posted five words to X: “Terafab Project launches in 7 days.” No thread. No context. No explainer post. Just a timestamp and a name that, if you’ve been paying attention to where xAI has been heading,…
Vera Rubin Changes the Math, Not Just the Hardware I’ve been sitting with Nvidia’s Vera Rubin announcement for a few days now, and the number I can’t stop thinking about isn’t the one everyone leads with. It’s not the 10x performance-per-watt improvement over Blackwell. It’s the inference token cost. Because that one doesn’t just improve…
