20VC20VC: Why Robotics Will be 100x Bigger than Coding Agents | Inference Will Be a $10TRN Market: What No One Sees | Why Software Investors are F***** Today with Naveen Rao, Unconventional
24 Oct 2026 · 69 min · Episode page ↗The true bottleneck to AI's boundless demand is energy, not compute, and solving this requires a fundamental shift to energy-first co-design across the entire intelligence stack, from models to physical hardware.
Summary
Naveen Rao, a serial entrepreneur and deep thinker in AI, argues that the exponential demand for intelligence is limitless, making energy the critical constraint for future AI growth. He challenges conventional thinking about AI investment, asserting that the future lies in hardware co-designed with models for energy efficiency, predicting a post-Transformer world and a massive, specialized robotics market that dwarfs current coding agent applications. This conversation offers a contrarian view on where true value and defensibility will reside in the evolving AI landscape.
Interesting ideas
Energy, Not Compute, is the True AI Bottleneck. Rao contends that the rapid growth of AI will soon exhaust existing energy capacity. Unlike semiconductor fabs, new gigawatts of power are incredibly difficult and expensive to build, putting us significantly behind. This implies that the 'alpha' in AI investment will increasingly shift from optimizing pure compute speed to securing and efficiently utilizing energy, leading to a two-tiered energy market where data centers pay a premium.
The Shift from Speed-First to Energy-First Design. Historically, computers were designed for speed. However, with physical efficiency gains (Moore's Law) largely plateauing, the next frontier for optimization is co-designing models and hardware with an 'energy-first' mindset. This means treating the model as a malleable component within a holistic 'intelligence machine' to achieve massive gains, challenging the traditional separation of software and hardware development.
Hardware is the Future; Software Investors are 'F***ed'. Rao provocatively states that pure software investors will struggle, as the era of easy, frictionless SaaS margins is ending. He argues that intelligence will have infinite demand, and the conversion of energy to intelligence is hardware. The market will reward those who can most efficiently convert energy to intelligence, making the hardware layer (chips, systems, cloud, networks) the primary accumulator of margin over the next decade, with models themselves becoming more commoditized.
Robotics Will Eclipse Coding Agents by 100x. While coding agents are currently overhyped, Rao predicts that robotics, particularly in physical problem-solving, will become a 100-times larger market. This is because intelligence needs to extend into the physical world to address complex, dynamic challenges, leading to a future with billions of robots solving problems that were previously unapproachable. The value will reside in end-to-end devices where the robot and its specialized model are inseparable.
AI Elevates Humanity by Tackling 'Unapproachable' Problems. Rao dismisses AI doomerism as 'idiotic,' instead viewing AI as an unparalleled tool to advance scientific understanding. By dramatically expanding the surface area of solvable problems, AI enables humanity to ask and answer much bigger questions about the universe and fundamental physics that were previously beyond human capacity, leading to a new era of scientific discovery and human understanding.
Questions worth thinking about
If energy becomes the primary bottleneck, what are the most overlooked, unconventional energy sources or storage solutions that AI's demand might accelerate?
How does a 'co-design' philosophy, where the model and hardware are inseparable, fundamentally alter the competitive landscape for large AI labs versus specialized startups?
If the inference market becomes $10 trillion and hardware captures most of the margin, what new business models emerge for software companies that provide critical, non-commoditized layers within that hardware-centric ecosystem?
Written by AI from the episode's full transcript. It can get things wrong: if something here looks off, tell us at hello@podpings.com.