Bloomberg Television39 min
The AI Chip Race with Apple’s Former Chief Chip Architect | Power On with Mark Gurman
Mark Gurman, Gerard Williams III, Alexis
Interview, Fireside Chat
- NuvaCore's development journey began in early April, with full silicon production estimated to span two to three years from the current date, potentially compressing the typical two-and-a-half to three-year schedule into a two-year design point or a one-year timeline for Generations 1 and 2.
- The company plans to utilize industry processes from TSMC, Samsung, or Intel to create CPU architecture compatible with x86, ARM, or RISC-V instruction sets, targeting end products ranging from chiplet-based silicon co-integrated with accelerators to full-scale blade-based compute devices for hyperscalers.
- AI tools have reduced the CPU performance development phase from seven to twelve months to a two-to-three-month window, while the overall ambition is to deliver high-performance and power-efficient products that enable data centers to fit two to three times the number of cores in the same physical space.
- Future roadmaps indicate that bringing CPU technology to market may occur within two to five years, though achieving a clean-sheet-to-endpoint timeline of one year remains uncertain.
- The strategic outlook anticipates AI models serving as extensions for design teams rather than replacements, requiring experienced engineers to vet capabilities, while end users eventually scale to exploit very large models hosted in the cloud.
- Current investor base consists of strategic players with active involvement in silicon who recognize the project as a long-term endeavor.