The demand for AI is about to transform an often-ignored part of the tech infrastructure stack: networking.
AI’s growth puts tremendous bandwidth performance requirements on the network and the networking industry is gearing up to address this demand.
Hosted by the Linux Foundation, Arista Networks, Cisco, HPE Aruba Networking, Intel Corporation, AMD, Broadcom, Meta, and Microsoft are forming Ultra Ethernet Consortium (UEC), to redefine Ethernet for the next-generation AI infrastructures.
Here’s what’s on the horizon for UEC…
1. Addressing Market Impact: 20% of Ethernet data center switch ports are expected to connect to accelerated servers supporting AI workloads by 2027, fueling a data center switch market projected to exceed $100 billion in the next five years!
2. Supersizing Ethernet: Meeting the high-performance needs of AI with increased scale, stability, and reliability.
3. Ultra high-bandwidth Switches: We’re about to witness record shipments in high bandwidth switches, even 800Gbps switches, bringing a new era of speed and capability, addressing AI/ML demands.
4. Multi-pathing & Congestion Control: Advanced techniques for optimal AI traffic handling, ensuring AI workflows have simultaneous access to destinations and even spreading loads.
5. Enhanced Security: Incorporating modern encryption methods for highly secure, efficient AI operations.
This isn’t just about tuning Ethernet; it’s about crafting a modern transport protocol that can keep up with the accelerating pace of AI and redefine our connected world.
What are your thoughts about AI-driven transformation of the networking industry?
#networking #ethernet #generativeai #infrastructure #security
Data: Network World, Arista Blog and Whitepaper, SerDes Speed
