Today’s AI can write code, pass exams, and generate images, but it still doesn’t know how to learn. In this episode of the Superhuman AI Podcast, we sit down with Ashfaq Munshi, CEO of Pepperdata and former CTO of Yahoo, to unpack what he calls the real problem with AI: models don’t build world models, don’t learn from experience, and don’t improve the way a child does.

Ashfaq walks through the missing pieces in today’s LLMs, why world models and true learning will define the next decade, and how deep infrastructure from chips to Kubernetes to GPUs to power grid underpins every “AI breakthrough” headline you see. He also breaks down how Pepperdata squeezes 20-30% more efficiency out of massive GPU and Kubernetes clusters for some of the world’s biggest companies.

We dive into:
- Why Ashfaq believes today’s AI models “don’t know how to learn”
- World models, robotics, and what real learning would look like
- The hidden infrastructure stack that makes modern AI possible
- How Pepperdata optimizes GPUs and Kubernetes to save real money
- Data centers, power, cooling, and the coming energy crunch
- US vs China in AI: talent, chips, research, and long-term advantage
- Career advice for builders who want to work on decade-long problems

If you care about where AI goes after the current hype cycle down at the metal and up at the level of real intelligence this conversation with Ashfaq cuts straight to the core issues.

Subscribe for deep conversations on AI, agents, and the future of intelligence.