You are viewing a single comment's thread from:

RE: LeoThread 2025-10-13 17-37

in LeoFinance7 days ago

Part 3/8:

This necessity reflects a broader understanding that off-the-shelf hardware isn't enough for the future of AI. Custom chips, optimized for workload-specific performance, are becoming essential rather than optional.

Strategic Collaboration: Co-Development of Chips and Networks

Charlie Cowas, President of Broadcom, provided insights into the innovative nature of this collaboration, which he describes as a "trifecta" of technological integration:

  • Custom Accelerated Chips: Building specialized AI accelerators, GPUs, or XPUs tailored for OpenAI’s workloads.

  • Open Ethernet Networks: Linking these accelerators through industry-standard Ethernet protocols to ensure scalable, flexible data movement.