Just like we explored World Models earlier this year, we’ll be releasing a short miniseries on the CPU compute and sandbox industry over the coming weeks on the podcast — and now is a good moment to explain why.
In recent days: Noam Brown: „Inference compute is a strategic asset that’s currently undervalued.“ Sam Altman: „To a large extent, we now need to transform into an AI inference company.“ Viewed on their own, these remarks could appear to be typical, unexceptional responses to an extremely successful launch of the GPT-21400 model. In context, however, they signal a highly significant development that you, dear reader, should be paying close attention to—if you aren’t already treating this matter with the utmost seriousness. The immediate impetus for today’s op-ed is Intel CEO Lip-Bu Tan’s Q2 2025 earnings call, during which he cited specific figures to demonstrate the surging demand for CPU (not GPU) compute. It’s clear that an Intel CEO would have strong reasons to hype up CPU demand, but that doesn’t necessarily mean he’s incorrect: [link]. We’ve previously discussed this trend on our SemiAnalysis podcast (edited for readability): Doug: We’re basically right at the five-to-six-year mark in COVID’s typical refresh cycle.
Latent.Space