What will computer chips look like in 2045?

The episode examines a central tension in the future of computing: how much chip technology can advance when AI demand is growing faster than manufacturing capacity can readily expand. Host Dwarkesh Patel spent two days touring Intel’s most advanced fab and speaking with its technical leadership; the show notes do not name individual guests. Rather than offering a definitive picture of chips in 2045, the episode explores the transistor, wiring, packaging and production changes that could shape them.

The technical discussion spans gate-all-around transistors, backside power, stacked transistors known as CFET, and hybrid bonding, supported by animated explainers. It also examines bringing up a new process node, lessons from Intel’s 10nm experience, replacing copper wires with ruthenium, and the chip-to-chip bottleneck. Together, these topics frame progress as a problem extending beyond individual transistors to connections within and between chips, as well as the manufacturing processes needed to deliver those changes.

The immediate context is the scale of AI’s demand for compute. The notes state that AI compute has grown threefold annually over the last decade, while attributing to Intel’s CTO the assessment that today’s global stock of chipmaking tools would allow Intel to only double leading-edge wafer production. Those figures concern different measures—compute growth and wafer-production capacity—but they motivate the episode’s question about how foundries can delay demand substantially outstripping supply. The discussion also raises company-specific trade-offs, including why High-NA EUV pays off for Intel but not TSMC, and why stacked transistors take a decade to ship; the notes do not provide the detailed arguments behind either topic.

Several questions remain open in the supplied material: which technical changes can most effectively ease supply constraints, whether Intel can catch up to TSMC, and how fundamentally fabs could be redesigned. The closing discussion covers Terafab and rethinking the fab from scratch. The notes establish these as subjects of inquiry, not settled conclusions, and provide no firm deployment roadmap through 2045.