Aligning AI Speed with Infrastructure Reality. How Meta Builds for 12-Month Cycles in a 12-Day World
Автор: IgniteGTM
Загружено: 2026-01-13
Просмотров: 128
Описание:
📍 Recorded live at AI INFRA SUMMIT 4, Convene San Francisco
AI development moves in weeks. Infrastructure planning moves in years. In this mainstage session, Elisa Chen (Meta) breaks down one of the most difficult challenges in modern AI: aligning long procurement cycles with the rapid pace of model innovation, shifting workloads, and unpredictable demand.
Elisa shares how large-scale infrastructure teams forecast capacity, prioritize scarce resources, manage GPU shortages, and prevent both underutilization and overinvestment. She outlines the operating principles Meta uses to keep research velocity high while balancing budget, power, space, and operational realities.
Highlights from the session:
Why AI teams iterate in weeks while infrastructure teams plan in 12+ month cycles
The core drivers of AI capacity demand, from organic traffic to R&D volatility
How Meta approaches forecasting, prioritization, headroom planning, and scenario modeling
Strategies for allocating scarce GPU resources, managing elastic pools, and reclaiming idle capacity
The future of AI infra: specialized hardware, declining costs, and autonomous capacity management
📣 Super early bird available — sign up for the next AI INFRA SUMMIT → https://luma.com/aiinfra5
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
