Summary
Former Intel CEO Pat Gelsinger, now a general partner at Playground Global, joins a16z for a wide-ranging conversation on why hardware may be at its most exciting moment in decades. Drawing on a career that began at Intel at 18, included work on the 286, 386 and 486 chips, and later led VMware and Intel, he offers an insider view of the AI infrastructure buildout.
The discussion centers on constraints that now shape AI: power availability, with Gelsinger arguing that energy capacity equals economic capacity and warning of defaults on data center projects that cannot secure power. He describes how the unit of design has moved from the chip to the rack, and why bottlenecks shift whenever one problem becomes easy. He also covers the lack of major new memory technologies in 30 years, the limits of HBM, and the prospects for memory innovation.
The conversation also addresses the crowded field of AI inference accelerator startups, why Gelsinger expects consolidation, and how large players such as Nvidia, with its Groq deal, absorb heterogeneity into unified architectures. It is a valuable listen for anyone following AI chips, data center economics and hardware investing.
📺 Source: a16z · Published October 09, 2026
🏷️ Format: Interview
