PERSON DIRECTORY
Sonya Huang
Host of Training Data. Sonya Huang appears in 2 indexed conversations across Sequoia Capital, Training Data. This directory brings every appearance, source, TL;DR, digest, and transcript into one searchable feed.
Why Hardware-Software Co-Design Is AI's Real 100x: Dylan Patel of SemiAnalysis
Dylan PatelShaun MaguireSonya Huang
Hardware-software-model co-design can turn three 2x gains into 100x, creating model-silicon lock-in and shifting CUDA’s moat toward ecosystem gravity.Demand is outrunning compute supply: Anthropic’s Opus 4.8 API margins exceed 80%, but leveraged buildouts remain exposed if useful model work stops expanding faster than capacity.
Nvidia CTO Michael Kagan: Scaling Beyond Moore's Law to Million-GPU Clusters
Michael KaganSonya HuangPat Grady
AI’s scaling curve is shifting NVIDIA’s optimization target from chips to rack-scale systems, with models doubling every three months and demanding roughly 10X–16X annual performance growth.Mellanox’s low-jitter networking, NVLink/CUDA integration and fault-tolerant architecture support scaling toward 100,000 GPUs, but energy and heat constrain 100–150-megawatt deployments as Kagan says he does not know how long the 10X annual slope will last.