Descriptions:
A16z sits down with Naveen Ralph, co-founder and CEO of Unconventional AI, for a technically grounded conversation about the fundamental limits of digital computing for AI and the case for rethinking chip architecture from first principles. Ralph brings rare depth to the discussion: he co-founded Nirvana, an AI chip accelerator company later acquired by Intel, built Mosaic (a cloud computing startup), and served as head of AI at Databricks before launching Unconventional AI.
Ralph’s central argument is that the digital computing paradigm—largely unchanged since the 1940s—is fundamentally mismatched with many AI workloads. Drawing on a PhD in neuroscience, he explains how biological neural systems operate at roughly 20 watts while performing computations that dwarf today’s silicon architectures in efficiency. He identifies workloads expressible as dynamical systems—those with an inherent time dimension—as particularly suited to analog or neuromorphic approaches, contrasting these with the numerical simulation that digital systems must perform.
The interview also addresses the broader infrastructure challenge: with big tech planning to add hundreds of gigawatts of data center capacity, Ralph notes that aging power transmission grids represent a serious structural bottleneck. He positions Unconventional AI as pursuing a fundamentally different path from Nvidia’s CUDA ecosystem, while acknowledging TSMC as a likely partner and leaving open the question of whether Nvidia will ultimately be a collaborator or a competitor as the compute landscape evolves.
📺 Source: a16z · Published December 08, 2025
🏷️ Format: Interview
