Keynote Panel: Hardwares & Accelerators
summary
Panel argues big deals close by repeatedly delivering scale and reliability, not magic: negotiating aligns three trains—data center & energy, hardware delivery, and customer timing—and large contracts trigger deeper scrutiny. Demand skews to inference and usage; neo‑clouds win by designing for failure, using telemetry and 20% spare pool plus automated detection and hot replacement with burn‑in. Specifics: an open source inference‑max benchmark compares cost per million tokens, Nvidia GPUs sell at ~4.5x markup, GB200 ships to neo‑clouds earlier than hyperscalers.
AI-written summary · about
chapters (6)
- 0:00Welcome
- 7:25NeoClouds disrupts AI cloud
- 17:10Memory needs with vendors
- 21:39Nvidia vs AMD GPU pricing
- 23:03Data center locations
- 29:17Closing
