Jensen Huang: LPX chips will stay niche, GPU remains king

NVIDIA founder Jensen Huang said on an earnings call that SRAM-based AI inference accelerator chips like LPX will stay a niche market for a long time, with GPUs remaining dominant.

LPX is designed for low latency and high token speed, but has lower throughput and capacity. It works well for software coding tasks but isn't great for agent-based tasks. Its use cases are limited, mostly high-priced AI services needing fast token speeds.

Huang noted these high-priced services currently make up far less than 20% of the AI market and may reach 20% in the future, meaning a relatively small customer base.