Beyond LLMs: How AI Networking Is Being Rebuilt from the Ground Up – Tech Field Day Takeaways

I'm Tom Hollingsworth, event lead for networking here at Tech Field Day, and here are my takeaways from Networking Field Day 40. Hello, everyone. We just had an amazing Networking Field Day event, Networking Field Day 40, that happened April 8th, 9th, and 10th.

It was a jam-packed event full of some great information about the current state of the networking industry, and there's a lot more technology innovation going on that you need to know about, and that's one of the reasons why I wanted to give you my three big takeaways from this event. Takeaway one is that scale-across is a new pillar. Scale-across networking is a vital new pillar in building AI networks.

Scale-across involves building out a network that spans multiple data centers and buildings. That's because modern data centers often have a power cap that exceeds the requirements of a cluster of GPUs, which often number in the tens of thousands. This leads to a need for geographic scale, either on a single site or across multiple sites.

Scale-out also increases resilience and resource utilization, even at the cost of increased network complexity. Scale-across is much more sensitive to latency and jitter and buffering than systems that are co-located in the same rack or even the same server room. Scale-across is set to thrive with the introduction of new optical technologies and advanced transport protocols from the Ultra Ethernet Consortium and others.

Companies like Cisco, Nokia, and Arista are leading the way with scale-across technology, and up-and-coming companies like Upscale AI are ready to take advantage of the strong market for novel solutions. My second takeaway from Networking Field Day 40 is that optical technology must evolve. The density of the modern AI data center means that optics are working harder than ever and creating lots of challenges.

Power consumption is at an all-time high, and the heat generation from these optics is causing premature equipment failure. New technologies such as linear programmable optics, or LPO, and Arista's extra dense pluggable optics, or XPO, are helping to ease the transition to more technology radical forms of optical interconnects, such as co-package optics and silicon photonics. LPO can reduce power consumption by almost 50% and increase thermal dissipation by the same amount.

LPO optics also have lower latencies and lower per-unit costs. XPO offers even more performance, as well as the ability to integrate liquid cooling. This is something that's seen as vital as a component for many high-performance modern data centers, and integrating it into the optic itself means that it will run at a cooler temperature and reduce overall likelihood of failure.

In the next three years, we can expect optics to grow in speed and complexity, and the time is right to ensure that the optics that we're deploying today are ready for what comes down the road. My third big takeaway is that the future of AI networking isn't LLMs. When you look at the number of companies that have been building out successful AI-focused solutions, while LLMs may have dominated the conversation a couple of years ago, the place where we're at right now relies on much more than simple predictive text generators.

Companies like Aviz Networks are leveraging agentic AI to help build out a full AI network operations center. This means that you can use their copilot platform to diagnose system problems and effect repairs quickly. Other companies, like Selector AI, are working closely with solutions like the MCP Server to help integrate what agents are doing more broadly across the network.

The data that they're collecting allows you to see trends in your devices in order to ensure that the problems that are happening now don't become full-blown crises down the road. Agents can report and take action on your behalf. And even when it comes to agents, they may not be the final form of the network.

When we talk to companies like netAI, and we see what the work that they're doing around graph neural networks, it opens up entirely new possibilities and gives us a glimpse into what could be coming down the road. And when you look at what companies like Netris are building on the back end with the solutions that they have been building upon for cloud networking and many others, you have a sense that there's more to this technology than simply writing someone's essay for them. This event showcased how quickly networking can change.

Years ago, we were talking about the advent of software-defined networking and hoping that we were going to keep up into the cloud era. And today, the network has become a vital component of ensuring that expensive GPU build-outs have the bandwidth and latency necessary to accomplish their tasks without costing the company millions of dollars in lost revenueI want to say a special thank you to the companies that presented at this event for the first time, netAI, Lightyear, and Upscale AI. Each of them had great presentations and introduced a very important topic in the space of networking.

I also want to say a special thank you to our new delegates who joined us for the first time, Jeff Elo, Vincent Salendro, and Mike Witte. And if you check out the Networking Field Day event page, you'll even see that Mike Witte gave a special Tech Field Day community session where he talked about the importance of GPU clusters and interconnecting them with networks. Make sure that you head over to Networking Field Day's event page to see all of the videos that were posted from this event, as well as to see all of the coverage from the delegates who participated.

They would love to share their thoughts about my takeaways and many more. Thank you for watching this episode of the "Tech Field Day Takeaways" series on the Tech Field Day Plus YouTube channel. If you enjoyed it, be sure to leave a like, subscribe, or share your thoughts in the comments.

Also, make sure to follow Tech Field Day on X, Twitter, Bluesky, and Mastodon for updates, and check out all of the presentation videos on the Tech Field Day website, as well as the Tech Field Day YouTube channel. com. You can tune into all of our events live on LinkedIn, on our website, and now on the main Tech Field Day YouTube channel.

Thank you for watching, and we'll see you at the next event.