00:00 – Intro
03:18 – 02:40 – Guest intro (Mohamed Awad, EVP, Cloud AI Business Unit at Arm)
04:02 – Internal and external drivers behind Arm’s decision to move from IP #licensing business to products – Meeting customers where they are
05:55 – Specific design choices made to build AGI CPU for Agentic AI – #Performance, #Efficiency, #Scalability
08:36 – Customization vs off-the-shelf – Democratizing off-the-shelf high-performance
13:05 – Focus of AGI CPU market: Hyperscalers w/o custom silicon (e.g. Meta Oracle ), #Neoclouds, moving from Arm in the #cloud to #onprem (e.g. @SAP), #X86 replacement (e.g. F5)
16:50 – AGI CPU fully built on Arm IP availability to licensees, no system or secret sauce. Licensees can develop AGI #CPU clones, if they like
21:00 – Can Hyperscalers using Arm IP for custom silicon learn from AGI CPU?
23:00 – Organizational challenges of changing from IP licensing business to product business (Fab and Memory supply constraints) – Same partners but from different relationship
26:03 – Contributing AGI CPU innovation back to #OCP (Open Compute Project) ecosystem
28:10 – Evolution of AGI CPU – Maximizing memory and I/O bandwidth per core to maximize CPU utilization; Equal access to Arm IP innovations for AGI CPU team and licensee
30:32 – Is making own silicon limited to the #AI infra market, or will it extend to others? – Meeting customers wherever they want (i.e., if market opportunity presents itself)
31:20 – Benefits of AGi CPU to the Arm ecosystem: 1)Hyperscalers both with custom silicon (Amazon Web Services (AWS), Google, Microsoft, NVIDIA ) and without; 2) Players such as Broadcom, Marvell Technology who help them; 3) Arm IP licensees
34:45 – Interesting landscape of AI infra CPU – X86 duopoly (Intel and @AMD) vs. wide variety of Arm ecosystem
37:49 – Dynamics of roadmap between IP, Compute Subsystem, and AGI CPU
39:25 – What does the success of Arm’s new pivot to making products look in the short and long term
41:18 – Closing
