AI industry shifts focus from building larger models to making them efficient enough to deploy at scale

Sara Hooker, CEO of AI lab Adaption, argued that today's AI systems are "monolithic"—once trained, their knowledge and capabilities are fixed—and said models need to evolve continuously to avoid massive inefficiencies.

Rodrigo Liang, CEO of chip company SambaNova, said trillion-parameter models remain too expensive and power-hungry, and described SambaNova's strategy as delivering faster inference with lower power consumption through hardware designed for large-model workloads—claiming "two to 3x better" performance than Nvidia Blackwell GPUs on the same models.

Hooker noted that today's enterprises deploying AI agents at scale often pay repeatedly for the same computational errors because agents aren't learning from mistakes, contributing to soaring API bills.