AI: Devouring the World One Byte at a Time – 2024 Year in Review
In my 2023 year-end note I said 2023 would be the “Year of Artificial Intelligence”. Looking back at all that’s happened this year, that prediction has been thoroughly validated.
Consider Nvidia’s stock price, which has almost tripled this year, propelled by an insatiable appetite for AI-focused products. Hyperscalers are projected to spend some $180 billion on data center capital expenses, largely for accelerated GenAI servers. The emergence of AI Neoclouds offering GenAI Infrastructure-As-A-Service (IaaS) is helping to address mounting capacity needs. Meanwhile, power constraints are leading to new deals – like Microsoft’s 20-year contract to help restart the Three Mile Island nuclear plant. It’s proof that AI has become an unstoppable force.
So, what does this mean for Ethernet? As we heard during our recent Ethernet in the Age of AI Technology Exploration Forum (TEF 2024), Ethernet faces immense pressure to support both the Scale-Up and Scale-Out networks that are essential for optimizing accelerated server productivity. The good news is that new industry groups like the Ultra Ethernet Consortium and UALink Consortium are forming to take these challenges on. There was a heightened amount of discussion around these topics at the 2024 OCP Global Summit, which will be reflected in the special 10th Anniversary edition of the Ethernet Alliance Ethernet Roadmap coming in 2025.
While the IEEE P802.3dj project is working toward defining 200G per lane for Ethernet by late 2026, the industry is (loudly) asking for 400G per lane yesterday, if not sooner. The Ethernet Alliance has a critical role in facilitating the industry-wide collaboration needed to meet this exceptionally tough challenge. Remember, it’s incumbent that we deliver solutions maintaining Ethernet’s brand of “It just works!” and we can’t afford to take our time in doing so. The clock is ticking and the pressure is on.
To all of this I say, “Never bet against Ethernet!”
Throughout my career, I’ve watched Ethernet evolve from a basic wired interconnect, to the wired interconnect, and finally, the backbone of today’s connected world. The key task now is focusing on what not to do so that we can narrow the scope of the problem. Often, when one technology – say, Ethernet – replaces another – for example, Sonet/SDH – the temptation is for the replacement technology to attempt to do everything that the previous technology did, and more. But over time we refine, separating the wheat from the chaff, and letting what’s unnecessary fall away.
As RFC 1925, Rule 12 states: “…perfection has been reached not when there is nothing left to add, but when there is nothing left to take away.”
Here’s to 2025.
Peter Jones
Chair, Ethernet Alliance
