AvidThink 2025 March Update

We held back our March update by a few days to include Roy’s takeaways from KubeCon EU 2025, which ran from March 31 to April 4 (including the co-located LF Open Networking and Edge Summit, plus the Cloud-Native Telco Summit). This is in addition to his notes from the MPLS World Congress (technically MPLS SRv6 AI SDNet) in Paris and NVIDIA’s GTC in San Jose.

Q2 2025 kicks off with massive uncertainty and volatility in the worldwide market. Under the umbrella of a global trade war, it’s unclear if the AI and data center boom can sustain tech optimism. We’re seeing signs that huge data center commitments from the end of 2024 and early 2025 may be wavering, and given the herd mentality of investors, we expect to see pauses or delays in data center projects while the global tariff war plays out.

Even if semiconductors are currently untouched by tariffs, global supply chains will reconfigure across the short, medium, and long term, with businesses focusing even more on resilience in the face of different scenarios that will play out over the next decade. We’re hearing from EMEA, LATAM, and APAC firms about their desire to choose suppliers domiciled in diverse geographic locations and more mentions of sovereignty (data, management, control).

US-based suppliers and cloud providers (especially the hyperscalers) may have to put more effort into allaying the concerns of global and regional telcos and other infrastructure players; meanwhile, this might be an opportunity for other players to grow their footprint.

Speaking of data centers, you may be one of the few who have yet to download our 2025 Data Center Networking Report, which covers the impact of AI and cloud workloads on data centers.

Here’s what the rest of our newsletter has in store for you:

• Roy’s Travelogue — NVIDIA GTC, MPLS World Congress, KubeCon EU (LF ONE Summit)
• Latest Reports — Data Center Networking for AI and Cloud Workloads
• Recent Articles— Insights from MEF’s GNE 2024 plus 2025 look ahead
• NextGenInfra — OFC 2025, MPLS WC 2025, MWC 2025, Data Center Networking
• February News Roundup — 5G SA momentum, SoftBank buys Ampere, AI, AI, AI-RAN and more
• Other Content — AvidThink in the News, Webinars, Meet Us Live!

Roy’s Travelogue

I barely had a week’s break from MWC Barcelona before NVIDIA’s GTC showed up on our analyst circuit. The AI gold rush continues, with in-person attendance of over 25,000 from all over the world gathered to check out the latest from the hottest company in technology. I attended the MPLS World Congress in Paris the week following GTC, followed by KubeCon EU in London.

NVIDIA GTC – San Jose

Fortunately, NVIDIA GTC was at the San Jose Convention Center, which required no flying, only fighting local traffic and a dearth of parking spots. By now, you’ve seen wrap-ups from all the major mainstream newspapers (that’s how important NVIDIA is today) as well as the trade rags, so I won’t rehash the obvious announcements around Blackwell Ultra, Vera Rubin, Rubin Ultra, NVL576 Kyber Rack at up to 600 KW/rack! A couple of things that caught my attention included:

• NVIDIA Dynamo, the open-source orchestration system for scaling AI inferencing, is just as important as the hardware announcements.

• Physical AI and world models, with Isaac GROOT N1 foundation model for humanoid robots and Newton Physics Engine.

• Aerial Omniverse Digital Twin (AODT), that can used to simulate real-world environments, such as cell towers and UE traffic patterns.

• Co-packaged Optics (CPO) with Spectrum-X Photonics is a bold move to go all in on CPO with attendant benefits of reduced power consumption while supporting higher throughputs (check out our Data Center Networking report for more context on this).

Two final takeaways — (1) Jensen touting the annual cadence of hardware release and how value in existing GPUs is destroyed as a result, and forward selling upcoming chips. Pre-selling typically stalls existing sales (unless there’s a shortage of existing GPUs, which counteracts this). Given that I’m hearing of and seeing more and more idle H100s in the market (aggregation marketplaces for excess GPUs show a potential glut of stranded one-off H100s out there), what Jensen said might come true, and value/revenue could be destroyed in existing GPU with no uptake for new GPUs. (2) It’s amazing how long well-paid engineers/executives will stand in line for $0.10 worth of candies from Microsoft’s onsite candy store with a large array of cool candy dispensers… the opportunity cost, the sugar highs, the carbs!

MPLS World Congress (SRv6, SDN, AI and Quantum Networking) – Paris

Upperside Conferences held its 26th annual conference which no longer discusses MPLS but focuses on SRv6, network automation, and other pertinent topics for layers 1, 2, and 3 connectivity. The show felt busier than previous years (1,000+ attendees) and was still packed on the morning of the 3rd day (which is usually relatively quiet). Co-locating the SD-WAN/SASE and the Quantum Networking forums was a good idea and allowed the crowds to mix and discuss physical networking (quantum key distribution) up the stack to layer 7 application connections. MEF contributed speakers to an enterprise-centric forum, with Accenture (last-minute cancellation due to travel issues), Bloomberg, Colt, and MEF represented on the panel – various aspects of NaaS, along with what enterprises expected from carriers.

Key takeaways for me from the conference: SRv6 is interoperable between all the major vendors and leading startups and will increasingly be used for policy-based networking and intelligent routing. Quantum computing-resistant encryption (post-quantum cryptography or PQC) is now on the horizon for many enterprises, carriers, and vendors that will need to provide usable solutions.

KubeCon EU – London

KubeCon EU welcomed more than 12,500 attendees (the highest ever and an increase over the 12,000 last year) to the Excel Center in London (it’s a big venue that takes 10 minutes to walk from one end to another, though it’s not quite the Fira in Barcelona). A dynamic crowd of developers and the ecosystem around them showed up to continue pushing the cloud-native agenda forward.

Co-located with KubeCon was the Linux Foundation Networking Open Networking and Edge Summit (called ONESummit previously but goes by its full name now) and cloud-native telco day. The core group of networking and telco folks met to discuss the state of networking, edge, and telco since the advent of SDN and NFV and the cloud journey we’ve been on. It was a significantly smaller subset of attendees, but the true believers were there.

Key takeaways:

• Many technologies and requirements the networking and telco community cared about have diffused and been rolled up into the broader ecosystem. Isolation and portability show up in the WASM efforts that cross industries; networking needs are present in more mainline efforts to accommodate SmartNICs, eBPF, Cilium, Calico, etc., and edge workloads are tending toward AI, meaning that edge inferencing is again cross-industry and less telco-specific.

• Remaining dedicated efforts, including API, telco cloud, and edge efforts that encompass CAMARA, Project Sylva, and EVE (which is more industry 4.0 and IoT than pure telco), will continue to drive interest. The absorption of the Open Infra Foundation (formerly OpenStack Foundation) under the Linux Foundation will strengthen the telco cloud efforts (especially given existing Broadcom customers looking for cheaper VMware ESXi and NSX alternatives).

• AI continues to drive interest across verticals, and newly announced Linux Foundation Networking Projects Salus and Essedum, which Infosys is helping kickstart with code donation (the Infosys Responsible AI Toolkit will power Salus).

• KubeCon EU had a couple of main themes: observability and unified observability. Did I say observability already? The remarkable number of vendors touting that on the show floors across the north and south halls makes it confusing to pick. In addition, multi-cluster management (KubeFleet), security (SLSA, SBOM), AI-assisted operations for K8S, cost-management (FinOps is still a thing but not as hot), developer portals (BackStage), platform engineering, and AI inferencing stacks on K8S were the other key messages across vendors and presentations.

I’m happy to chat more about any of the shows and our findings. Drop us a line at [email protected]. Up next for me are RSA 2025 (San Francisco) where our media partner Jim Carroll has cooked up a new “En route to RSA via Waymo” interview series (drop us a note if interested), FutureNet World back in London, and then AI Networking Summit (aka ONUG) in Dallas. Catch you then!

– Roy for the AvidThink Team