Sam Khosroshahi, Lambda | NVIDIA GTC '26
summary
Lambda is building gigawatt-scale campuses with liquid‑to‑chip cooling, co‑packaged optics and network acceleration so GPUs never sit idle, arguing every idle GPU is lost value. Claims include a 40‑million‑times compute growth over a decade and tiered CPU/ASIC use for inference; Lambda publishes MFU benchmarking for FLOP utilization and reports early deployments of Vera Rubin and co‑packaged optical stacks. Regional capacity bookings run across North America under the Reserve Your AI Factory program.
AI-written summary · about
