AvidThink 2026 March Recap

Barely a week after MWC in Barcelona came NVIDIA’s GTC and OFC (optical fiber communications conference), followed by the flagship security conference RSAC, KubeCon EU in Amsterdam, and the 27th edition of Upperside World Congress (formerly known as MPLS World Congress) in Paris, all in the same week! Roy couldn’t be in two places at once (quantum entanglement aside), so he headed to Paris while our media partner, Jim Carroll of Converge Digest, headed to RSAC.

You’ll find on our NextGenInfra.io site a collection of showcases, including MWC 2026 and OFC 2026, with videos from Upperside World Congress to be posted shortly.

Here's what's in store for this month's newsletter:

Looking ahead, keep your eye on NextGenInfra.io for updates on Data Center Networking for AI — coming soon!

—The AvidThink Team

March Madness — AI Infrastructure Edition

If end February/early March was the MWC marathon, the rest of March felt like a triathlon. I traveled to San Jose, Los Angeles, and Paris in just two weeks.

NVIDIA GTC (San Jose, March 15–19) — 30,000+ attendees

Over 30,000 (5,000 more than last year) descended on San Jose — the largest GTC ever. The networking content was impossible to miss, but speaking of connectivity, traffic into and out of the convention center was a nightmare. GTC outgrew San Jose last year, and it was even more evident this year. Regardless, the headline for us was NVIDIA's Vera Rubin platform with NVLink 6, delivering 3.6 TB/s per GPU and 260 TB/s aggregate across NVL72, plus the new Kyber rack architecture that scales Vera Rubin NVLink domains to NVL144 and NVL576. And then Feynman in 2028 with NVL1152. Cable-free, vertical-insertion, and designed to blur the line between scale-up and scale-out. NVIDIA surprised many by quickly bringing Groq into the fold, integrating Groq LPUs into the Vera Rubin platform for disaggregated inference — purpose-built silicon for prefill and decode — and claiming 35x throughput per watt over Blackwell NVL72. On the optics front, Spectrum-X CPO switches at 102.4 Tbps, underscoring NVIDIA's seriousness about owning the full-stack connectivity story. More on all this in our upcoming Data Center Networking report.

OFC 2026 (Los Angeles, March 16–20) — 18,000 attendees

The buzz at OFC confirmed that optics is having its moment. One of the announcements that caught our attention was the XPO Multi-Source Agreement, organized by Arista with 45 founding members. XPO defines a 12.8 Tbps liquid-cooled pluggable with 8x the bandwidth of today's 1.6T OSFP and bridging the gap between pluggables and CPO without sacrificing field serviceability. Nokia and Ciena both introduced multi-rail amplifier platforms (up to 160 and 128 fiber pairs per rack, respectively) for multi-campus AI deployments. Marvell confirmed its Ara 1.6T DSP is shipping in volume, and teamed with Lumentum to demo an integrated optical circuit switching system spanning scale-up, scale-out, and DCI (demonstrating that it might be possible to do optical circuit switching outside of Google). The path to 3.2T is real: Broadcom debuted Taurus, its first 400G/lane optical DSP, and Coherent showed 3.2T transceivers in silicon photonics. Again, more in our upcoming report.

Upperside World Congress (Paris, March 24–26) — 1,150 attendees

From photonics back to packets. I kicked off the Upperside World Congress with an opening keynote on "Connectivity in an AI Era." That was probably fitting, since I had just spent two weeks immersed in exactly that. The Congress, formerly the MPLS World Congress, continues to grow and remains the go-to gathering for the IP routing and transport community. This year's main-stage agenda leaned heavily into AI-driven DCI, SRv6 in AI infrastructure, IP/optical convergence with 800G-ZR/ZR+ pluggables, and what hyperscale routing looks like in the AI era. Vendors from Cisco, HPE/Juniper, Ciena, Nokia, and Huawei pitched their visions. The EANTC interop showcase kept everyone honest.

Three events, one takeaway: whether you're scaling up NVLink domains, lighting new fiber pairs across campuses, or engineering IP overlays for AI traffic — connectivity is part of the AI conversation. And I think March was just the warmup.

Hope to see you all at one of the upcoming events I’ll be attending: Wi-Fi Now (Mountain View, CA), FutureNet World (London), AI Networking Summit/ONUG (Dallas), OCP Canada Tech Day (Montreal), and TM Forum (Copenhagen).

– Roy for the AvidThink Team

Check out our most recent articles, including our hot-off-the-press take on OXIO’s acquisition of Movistar Mexico, our third post in our telco API series, and our views on NTN

Our OFC 2026 site is up alongside our 2026 MWC Barcelona site! Plenty of current and pertinent content covering the latest in mobile technologies, satellite, agentic Ai in telco, telco APIs, as well as fiber data center interconnects, scale-up, and scale-out optical connectivity. All those hot discussions from OFC are captured in our interview showcase featuring Broadcom, Ciena, Cisco, Lumentum, Marvell, Nokia, Viavi, and more!

Qualcomm Targets 3GPP Release 21 for 6G Standards by 2029
Qualcomm’s John Smee outlined a 2029 target for initial 6G standards (3GPP Release 21), emphasizing an “AI-native” air interface where AI optimizes signal transmission dynamically. A key innovation is integrated sensing and communications (ISAC), enabling centimeter-level positioning and environmental awareness for use cases like drone traffic management and smart infrastructure.

Experts Say ISAC Commercialization Likely Post-2030
ISAC aims to unify communication and sensing via shared RF waveforms across low-, mid-, high-band, and THz spectrum, enabling applications such as autonomous vehicles and industrial automation. However, IEEE and ETSI-aligned research and ABI Research forecasts suggest mainstream deployment will occur after 2030 due to infrastructure density and multi-sensor integration requirements.

SK Telecom and Ericsson Launch Multi-Year 6G Research Collaboration
SK Telecom and Ericsson will collaborate through 2031 on 6G R&D, focusing on AI-RAN, zero-trust security, and ISAC. The partnership aims to extend 5G capabilities while advancing energy efficiency and AI-driven network performance to position both firms in early 6G standardization efforts.

U.S. Pushes for 6G Device Readiness by 2028 Los Angeles Olympics
The U.S. government is pressuring industry to deliver early 6G devices by the 2028 Olympics, though timelines conflict with ITU IMT-2030 and 3GPP Release 21 standardization cycles (2029–2030). The Olympics are expected to serve as a pre-standard testbed despite uncertainty around commercial readiness.

China Operators See Slowing Growth as 5G Market Matures
China Mobile, China Telecom, and China Unicom reported flat or declining revenues in 2025 as 5G penetration exceeded 60%, with capex cuts of 8–14%. Despite this, subscriber growth continues, with China Mobile reaching 642M 5G users (64% penetration), signaling a shift from expansion to monetization.

AT&T and Ericsson Call for Zero Trust 5G Security Model
As 5G SA and Open RAN expand the attack surface, AT&T and Ericsson advocate replacing perimeter security with Zero Trust architectures requiring continuous verification. They urge industry-wide standards to address risks introduced by AI-driven and software-defined networks.

SoftBank Selects Ericsson for 5G SA Core Modernization
SoftBank signed a multi-year deal with Ericsson to deploy a cloud-native dual-mode 5G core supporting 4G/5G coexistence and transition to 5G SA. The platform includes automation, subscriber data management, and IMS systems, with a roadmap toward 5G-Advanced and AI-driven operations.

5G Core Spending Surges 83% YoY in Q4 2025
Omdia reports an 83% year-over-year increase in 5G core investments driven by SA deployments and digital transformation, with 88 operators launching SA globally. Growth is strongest in North America and EMEA, with monetization tied to slicing, URLLC, and enterprise services.

EE Prioritizes Consumer Use Cases in 5G SA Rollout
BT’s EE is targeting consumers rather than enterprise slicing with its 5G SA rollout, focusing on speed and throughput improvements. Coverage currently reaches 67% of the UK, with plans for 99% by 2030.

T-Mobile and Nvidia Advance AI-RAN with Distributed Compute Sites
T-Mobile is partnering with Nvidia and Nokia to transform cell sites into AI compute platforms using RTX Pro 6000 Blackwell systems and anyRAN software. The operator reports a 40% efficiency gain from “service-aware RAN” and is deploying AI for real-time network optimization and disaster response.

Early 5G-Advanced Deployments Expand Across Global Operators
At least 85 operators have launched 5G SA, enabling progression to 5G-Advanced, with ₁₂ operators already deploying it. Features such as RedCap and dynamic slicing are expected to unlock new monetization opportunities.

Vodafone Idea Expands 5G Footprint to 133 Cities
Vodafone Idea plans to expand from 43 to 133 cities, focusing on high-demand urban clusters and industrial corridors. The rollout leverages equipment from Nokia, Ericsson, and Samsung Networks to close gaps with competitors.

Huawei Growth Slows Despite Continued R&D Investment
Huawei reported 2.2% revenue growth (~$128B) and a 9% increase in net profit to $9.9B, alongside $28B in R&D spending (+7%). Workforce expansion to 213,000 reflects continued investment in AI, cloud, and automotive software despite slowing growth.

Rakuten Symphony Launches Japan Unit to Expand Enterprise Cloud Offerings
Rakuten Symphony established a Japan-based unit to deliver cloud-native OSS, edge, and automation solutions to enterprises. The strategy leverages Rakuten Mobile’s architecture to reduce infrastructure costs and accelerate digital transformation.

NTT East Demonstrates 98.5% Success Rate in Open RAN Interoperability Tests
NTT East validated multi-vendor Open RAN interoperability across 335 configurations with 26 partners, achieving 98.5% success. The trials highlighted RIC-driven automation benefits for power efficiency and performance in industrial use cases.

T-Mobile Deploys Private 5G Across MLB Stadiums for Automated Officiating
T-Mobile rolled out private 5G across 29 MLB stadiums to support real-time strike zone validation using Hawk-Eye cameras. The system delivers low-latency, isolated connectivity for officiating decisions during live games.

CBRS Drives 75% of U.S. Private 5G Deployments
Analysys Mason reports CBRS underpins 75% of U.S. private 5G networks, with over 2,500 manufacturing deployments expected by 2032. Regulatory changes to spectrum allocation could disrupt this ecosystem due to hardware redesign and interference risks.

4G LTE Still Dominates Private Networks Despite 5G Growth
GSA reports over 2,000 enterprise private networks globally, with LTE still comprising 51% of deployments while 5G leads new builds. Market growth continues across regions like Canada and Brazil despite economic and supply chain uncertainties.

AT&T Advances Cloud RAN Strategy with Intel While Maintaining Hardware Flexibility
AT&T is deploying cloud RAN using Intel Xeon 6 while maintaining flexibility across Nvidia, AMD, and TPU platforms. The software-centric architecture supports continuous upgrades and aligns with its $14B Ericsson modernization program.

Nokia and Ericsson Diverge on AI-RAN Hardware Strategy
Nokia is shifting toward Nvidia GPU-based RAN processing, while Ericsson continues investing in custom silicon for performance and efficiency. Ericsson also maintains a parallel hardware-agnostic vRAN strategy to mitigate supply chain risks.

ODC Raises $45M to Advance AI-Native Open RAN Platform
ODC secured $45M from Nvidia, Nokia, AT&T, and others to scale its Odyssey AI-native RAN platform integrating sensing and edge intelligence. The funding underscores industry momentum toward software-defined, AI-driven telecom infrastructure.

Virgin Media O2 Launches RFP for Next-Gen RAN Suppliers
Virgin Media O2 initiated a two-year RFP to modernize its UK RAN, prioritizing cost efficiency, capacity, and coverage. The process could introduce new vendors beyond Ericsson and Nokia amid evolving Open RAN dynamics.

Nvidia Introduces NemoClaw Enterprise Stack for Agentic AI
Nvidia unveiled NemoClaw, an enterprise framework built on OpenClaw with added security via OpenShell runtime, including policy enforcement and privacy routing. Positioned as foundational infrastructure, it enables safe deployment of autonomous AI agents in regulated environments like telecom.

AT&T Launches GenAI-Powered Unified App for Mobile and Home Services
AT&T’s redesigned app integrates mobile and home services with a GenAI assistant and agentic architecture decoupled from specific LLMs (e.g., Gemini, ChatGPT). The platform enables dynamic service management and personalized customer interactions with built-in guardrails.

Nvidia Invests $2B in Marvell to Expand AI Infrastructure Ecosystem
Nvidia’s $2B investment in Marvell targets integration of custom silicon, optical interconnects, and NVLink Fusion into scalable AI data center architectures. The partnership aims to unify compute, networking, and telecom infrastructure for next-generation AI workloads.

If you would like to meet with AvidThink at one of the following upcoming events, please click on the event below to request a meeting.

if this newsletter was forwarded to you, why not sign up for your own subscription?