AvidThink 2026 June Recap

June saw Roy visiting Montreal for OCP Canada Tech Day, Meta’s Menlo Park offices for the Meta@Scale AI and Data conference, and finally Copenhagen for TM Forum’s DTW Ignite 2026.

We continue to keep busy this summer while different team members take time off with their families. The team is busy with a few upcoming reports. Notably, the 3rd edition of our AI for Networking (AIOps and automation) report, which has been a crowd-pleaser (we also like to think that it provides deep insights to executives across enterprises and carriers), is set to launch in early Q4 2026. We’re also starting to plan our theses for our Enterprise Connectivity (SASE, SD-WAN, Campus) report, which is targeted for mid to late Q3 2026. In the meantime, we continue to engage the market and sponsors on two new reports: (1) Satellite/Non-Terrestrial Networks (NTN) and (2) Next-Gen Edge Computing, with a focus on AI inferencing workloads. If you’re interested in participating in any of these, drop us a note at [email protected].

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

Visit NextGenInfra.io to check out the latest video showcases and download your copy of our Data Center Networking Report.

—The AvidThink Team

Final round of travel before my summer no-travel break! So glad that I’m home for at least 6 weeks. In any case, here’s a quick rundown of places visited and events attended in June.

OCP Canada Tech Day - Montreal

I’ve been to Montreal before, but in Winter, when I saw the airport, the inside of a car, an office, a restaurant, a hotel room, and the airport again. So glad this visit happened as Spring turns to Summer. Had a wonderful time at the OCP Canada Tech Day, where the audience size basically doubled from the year before, and I got to learn more about the Canadian ecosystem for data centers, distributed AI, and quantum.

Thanks to the OCP team, I moderated a panel in the AI Computing Continuum (core-to-edge) track, discussing the prospects for edge workloads, AI inference, and the demands on the underlying computing architecture. Enjoyed my time with Jeff Wittich (Ampere), Nazym Paltachev (rivvor), Rob Wood (Tetrel Security), and Remi Duquette (Maya HTT). The on-demand recordings are available (see the link below in our webinar section). Looking forward to the OCP Global Summit in San Jose in the Fall, when many of these topics will be updated.

Meta@Scale - AI and Data - Menlo Park, CA

Next up was an opportunity to attend Meta@Scale in person. I had conflicts in previous years for their Networking and other @Scale, and was resigned to watching these videos online. I really appreciated being able to attend their AI and Data event at their Menlo Park offices. I’m glad that Meta is still open to sharing their learnings about running AI applications and infrastructure at scale. If you recall, the Open Compute Project (OCP) and Telecom Infra Project (TIP) were originally spawned by Meta/FaceBook. Both organizations have been instrumental in supporting open standards, architecture, and software and hardware across the data center and telecom industry.

You can watch the videos online when you get a chance. For me, it was helpful to understand the upper-layer challenges in AI and data and their impact on the underlying infrastructure. Likewise, seeing Boris Cherny, head of Claude Code at Anthropic, on stage discussing his current workflows and reliance on AI agentic capabilities gives me inspiration to up our game here at AvidThink. Of course, his tokens are “free” (the opportunity cost is not having them available to paying customers), which is why he’s consumed 8B tokens since March. And he hasn’t manually written a line of code since November of 2025. Today, the majority of coding he does is on his phone (via voice commands). Clearly, there’s an incentive for him to talk up his own product and company, and there’s likely hyperbole embedded in his answers. Nevertheless, it provides a view of one possible future for coders (and the rest of us).

Other takeaways for me were in the agent identity, guardrails and security, and the harnessing space — interesting take on AI guarding AI and some different ways of thinking about the problem. It was actually a great mental setup for me as I headed to DTW Ignite to face the next wave of agentic AI in telcos.

TM Forum’s DTW Ignite 2026 - Copenhagen, Denmark

DTW Ignite has the benefit of being held in Copenhagen in nice weather. Even with the rest of Europe burning, we experienced a peak of 28C, which was warmer than usual but quite tolerable. Unless your cheap hotel had no air conditioning, like mine.

At DTW, thanks to TM Forum and Netcracker, I had the chance to moderate a panel with Tamás Bányai, Chief Executive Officer, One Hungary; Christian Thrane, Chief Executive Officer, TDC Brands; John Porter, Chief Executive Officer, Telenet; and Sylvain Seignour, President, Netcracker & CSG. We touched on how telcos can and should embrace AI as part of their strategy to become AI-native (just as we sought to be cloud-native previously). And huge thanks to Efrat Nir-Berger for snapping a photo of the panel — yes, my arms aren’t that long, plus I was distracted by moderating the panel. I also got a selfie with her at the AWS booth, where she was demoing a very practical AI use case (Store Genie) for PLDT, powered by Amdocs and built on AWS, which is providing a good return on their investment.

As might be expected, there was plenty of agentic AI at DTW Ignite, including the telco variant of agent-to-agent (A2A), named A2A-T (no prizes for guessing what the “T” stands for), which Huawei open-sourced earlier in the year at MWC. Many of the Catalyst submissions were agentic-focused or agentic-adjacent, including the winner for Outstanding Catalyst for AI and Automation: A2A-T Runtime, Phase III, which involved more than 25 champions (carriers) and participating organizations (so many logos to fit on the screen).

Many of my discussions focused on agentic frameworks, agent harnesses and the management of telco context, which some refer to as “ontology.” Such context can include knowledge bases, methods of procedure (MOPS), equipment user guides, telco-specific workflow processes/nomenclature, as well as topological relationships and equipment state and configuration (sometimes captured as digital twins of telco networks). Likewise, there was debate about model routing, orchestration, small language models, fine-tuning, and other LLM-specific topics. There is still rapid development on this front (telco and non-telco), and I expect our current workflows will evolve again in 3-6 months. I intend to capture much of this in our upcoming AI for Networking report, which will span carriers, data centers, and enterprise networks — I am seeing many similar patterns across all domains.

And as I was striving to keep up with my June travel and meeting schedule, our media partner, Jim Carroll of Converge Digest, was hobnobbing with the Cisco and HPE executives at Cisco Live US and at HPE Discover. Jim’s videos and interviews from those events are available on his Converge Digest website. His video captures from those shows are also available on our joint @nextgeninfra.io channel on YouTube.

We’re busy revamping our NextGenInfra site and working on our reports in anticipation of a busy Fall. If you are interested in any of the reports we’re working on listed earlier, or if you would just like to talk and catch up, drop us a line. Until next month, stay cool (especially if you’re anywhere in the Northern Hemisphere)!

– Roy for the AvidThink Team

Check out our most recent articles on the AvidThink website:

If you haven’t yet downloaded your copy of the 2026 Edition of our Data Center Networking Report, grab your copy today!

GSMA: 5G Now Represents More Than Half of Mobile Connections in China
GSMA’s Mobile Economy China 2026 report found that 55% of China’s mobile connections were on 5G at year-end 2026, with China accounting for 40% of global 5G connections. China Mobile, China Telecom, and China Unicom are now shifting toward 5G-Advanced, already commercialized across more than 330 mainland cities and exceeding 30 million users, with early use cases in autonomous vehicles, drones, and XR.

Deutsche Telekom Expands 5G Coverage and Capacity Across Germany
Deutsche Telekom added 81 new mobile sites and upgraded capacity at 548 existing locations, targeting regions including Baden-Württemberg, North Rhine-Westphalia, and Bavaria. DT says its 5G network now reaches about 99% of German households, while 4G is near 100%, and it has phased out LTE Dynamic Spectrum Sharing in the 2.1GHz band as usage shifts to 5G.

Turkcell and Bango Launch 5G “Super Bundles” for Streaming and Data Plans
Turkcell partnered with subscription platform Bango to launch “super bundles” that combine streaming services with data plans as the operator rolls out 5G across Türkiye. Turkcell, which serves more than 43 million subscribers, says the bundles could save customers up to 40% compared with standalone subscriptions while helping drive customer acquisition and data usage.

Ericsson: Global 5G Subscriptions Surpass 3 Billion
Ericsson’s latest Mobility Report says global 5G subscriptions passed 3 billion in Q1 2026 after adding 162 million subscriptions during the quarter, with 5G expected to reach 6.4 billion subscriptions by the end of 2031. Ericsson projects 5G will account for 85% of mobile data traffic within five years, while uplink traffic is growing faster than downlink traffic, driven by collaboration apps, user-generated content, cloud storage, and future AI workloads.

Samsung Targets Late 2027 Commercial Rollout for Advanced 5G Uplink Technology
Samsung says its new 5G uplink technology could be commercially deployed by late 2027 through software upgrades to existing networks, supporting FWA and AI workloads without new cell sites or spectrum. In a joint trial with MediaTek, Samsung achieved 670 Mbps uplink throughput using MediaTek’s M90 modem, Samsung vRAN software, massive-MIMO radios, and a “3Tx 5-layer” uplink configuration that could double maximum uplink throughput.

Dish Wireless Files for Chapter 11 as EchoStar Winds Down the Business
Dish DBS Corp. and subsidiaries including Dish Wireless filed for Chapter 11 as part of a restructuring plan to wind down Dish Wireless following major spectrum sales. The restructuring support agreement has backing from holders of more than 88% of Dish DBS secured and unsecured notes, representing more than $8.7 billion in Dish Wireless debt, and EchoStar said the filing was triggered by delayed cash proceeds from AT&T’s spectrum deal.

6G

3GPP Sets Early 2029 Completion Date for First 6G Standards
3GPP formally agreed to a 6G roadmap that targets early 2029 for completion of the first 6G standards, keeping commercial deployment broadly aligned with a 2030 timeframe. Release 21 will include initial 6G specifications and 5G-Advanced work, with a March 2027 functional freeze, March 2028 80% checkpoint, June 2028 second freeze, December 2028 stage 3 freeze, and final code freeze in March 2029.

Verizon, AT&T, T-Mobile, and SpaceX Win Key AWS-3 Spectrum in $3.5B Auction
The FCC’s AWS-3 re-auction of certain EchoStar/Dish holdings ended after 72 rounds with more than $3.5 billion in winning bids, surpassing the $2.9 billion threshold needed to avoid federal penalties for EchoStar. Verizon won licenses in 82 markets for $3.16 billion, T-Mobile won licenses in 102 markets for $277.78 million, AT&T won ten markets for $120.77 million, and SpaceX won two markets for $8.49 million, likely to support uplink spectrum for direct-to-device satellite services.

Ericsson and Telia Open $31M Swedish Testbed for AI, 5G, and 6G
Ericsson, Telia, Lund University, KTH Royal Institute of Technology, and other partners launched Digital Arena Sweden, a $31.3 million pre-commercial test center for AI, 5G, and 6G. The initiative builds on the NorthStar 5G program and targets sectors including mining, transport, defense, and medical applications, with a focus on integrating intelligence directly into future network infrastructure.

China Telecom: Future Networks Must Be AI-Native
China Telecom Chief Technologist Yue Wang said future 6G networks must be AI-native because today’s telecom architectures rely on predefined interfaces, rules-based logic, and human-engineered control systems. Wang described an AI-native network across infrastructure, operational, and service layers, warning that current networks lack AI-ready data, APIs designed for closed-loop control, and network functions built for AI lifecycle management and real-time decision-making.

Open RAN

Samsung Takes a Flexible Approach to AI RAN and GPU Deployment
Samsung is positioning its AI RAN strategy around architectural flexibility rather than universal GPU deployment, arguing that GPUs make sense for high-demand sites but not across most of the network. Samsung has deployed tens of thousands of vRAN sites for Verizon and supports operators including Telus, KDDI, and Vodafone, while also benchmarking its vRAN software on Nvidia CPU-GPU combinations to test uplink and downlink performance gains.

AT&T Passes Halfway Point in Open RAN Modernization
AT&T says it is more than halfway through its five-year, multi-billion-dollar effort to replace legacy Nokia infrastructure with Ericsson-powered open and cloud-based RAN technology. The operator is targeting 70% of wireless traffic on open hardware interfaces by year-end, has already moved more than half of traffic to open-capable hardware, and is deploying third-party open RAN small cells from 1Finity in markets including Dallas, New York, and Phoenix.

Indosat Ooredoo Hutchison and Nokia Advance AI-RAN in Indonesia
Nokia is working with Indosat Ooredoo Hutchison to modernize Indonesia’s mobile network with low- and mid-band 5G RAN capabilities, including AI-RAN. The companies are collaborating with Nvidia on AI-RAN field trials scheduled to begin by year-end, with Nokia developing AI algorithms optimized for Nvidia AI-RAN platforms to improve spectral efficiency.

AI

GSMA and TM Forum Partner to Accelerate Telecom AI Adoption
The GSMA and TM Forum are collaborating to align telecom AI assets, standards, and interoperability frameworks for more autonomous, AI-native networks. GSMA Intelligence data shows only 16% of current AI deployments target network use cases, despite network operations representing up to 34% of operator opex, so GSMA will focus on models and agentic systems through Open Telco AI while TM Forum develops deployment frameworks and standards.

Telecom Operators Target Energy Management as a High-Value AI Use Case
Operators are increasingly applying AI to energy management, a major cost category that can represent 4% to 20% of annual telecom operating expenses. MTN, AT&T, and Deutsche Telekom have highlighted AI-driven efficiency gains, including smarter data processing, network foundation models, and algorithms that dynamically power network assets on or off based on demand.

IBM Study: Enterprises Face AI Vendor Lock-In and Sovereignty Risks
An IBM study found that 91% of executives lack a clear understanding of their dependencies on AI vendors, models, and infrastructure, while 71% said switching their primary AI vendor or model would be difficult. IBM also found that 28% of organizations use four or more AI vendors, often through ad hoc adoption, and that poor workload placement can drive token processing costs up to 2.8 times higher when AI workloads run too far from enterprise data.

Orange Appoints Usman Javaid as Chief AI Officer
Orange Group appointed Usman Javaid, currently Chief Product and Marketing Officer at Orange Business, as Chief AI Officer reporting to Group CTIO Bruno Zerbib. Javaid will lead standard and agentic AI adoption across networks, customer relations, and business services as Orange targets more than €600 million, or about $640 million, in AI-driven value by 2028.

Akamai Warns Public Cloud Latency Is Too High for AI Agents
Akamai says agentic AI workloads need lower-latency infrastructure than traditional public cloud architectures were designed to deliver. Its State of AI Inference 2026 report found that 82% of organizations require end-to-end response times of 500 milliseconds or less for critical AI use cases, while 64% target 250 milliseconds or less, and Akamai argues CPU-side controller loops can account for up to 90.6% of total latency.

AT&T Develops Custom Network Foundation Models to Automate Operations
AT&T has developed custom “network foundation models” trained on proprietary network data, configurations, and KPIs to automate infrastructure, improve energy efficiency, and respond to real-time service disruptions. Raj Savoor, AT&T’s Vice President of Network Analytics and Automation, said the models have about 10 billion parameters and were trained on roughly 110 billion tokens, enabling more dynamic optimization than legacy ML systems.

T-Mobile’s “Kinetic Tokens” Concept Sparks Debate Over Telco AI Monetization
T-Mobile US CTO John Saw proposed “kinetic tokens” as a metric for measuring the network and physical requirements of emerging physical AI systems, distinct from standard AI tokens used for compute and text generation. The concept drew skepticism because many see robotics and physical AI connectivity as conventional internet traffic, but it reflects a broader debate over whether operators can monetize edge infrastructure and avoid becoming commodity connectivity providers.

GSMA Launches Telco Common Corpus to Improve Telecom AI Models
The GSMA and French infrastructure organization Pleias launched the Telco Common Corpus, an open-source effort to improve telecom-specific AI benchmarking and model development. The project will provide verified technical literature, patents, open data, and web projects to address gaps in frontier models, especially for network management and other telecom-specific tasks.

Private Networks

Ericsson Demonstrates Private 5G for Industrial Robotics at Texas Smart Factory
Ericsson is using its 300,000-square-foot 5G USA Smart Factory near Dallas to showcase private 5G for industrial automation, including autonomous mobile robots. The facility previously used 28 Wi-Fi access points, but now relies on two 5G Radio Dots using millimeter-wave and C-band spectrum from an unnamed public operator to reduce handovers and improve real-time asset tracking.

SD-WAN

Dell’Oro: SASE Revenue Surpassed $3B in Q1 2026
Dell’Oro Group says the SASE market grew 21% year-over-year to more than $3 billion in Q1 2026, driven by cloud-delivered security, branch modernization, and AI-era policy controls. SSE revenue rose 22% year-over-year, while access router revenue fell 10% as enterprises move from standalone routers toward SD-WAN overlays, secure branch CPE, and SASE-aligned branch platforms from vendors including Cisco, Palo Alto Networks, Fortinet, Cato Networks, and Versa Networks.

CyberFox Acquires Timus Networks to Target SASE for MSPs and SMBs
CyberFox acquired Tampa-based SASE startup Timus Networks to deliver cloud-based security for remote and on-premises users in small and mid-size businesses. The acquisition is designed to help organizations replace legacy hardware appliances from vendors such as SonicWall or WatchGuard with software-based cloud security, targeting underserved segments including small manufacturers, auto dealerships, and local government agencies.

Cloud

Amazon Commits Additional $13B to India AI and Cloud Infrastructure
Amazon plans to invest an additional $13 billion in AI and cloud infrastructure in India by 2030, expanding AWS data center capacity in Mumbai and Hyderabad and supporting access to managed AI services, developer tools, and Trainium chips. The commitment raises Amazon’s planned India AI and cloud investment to more than $21 billion between 2026 and 2030 and cumulative India investment since 2010 to more than $88 billion.

Satellite

Rocket Lab to Acquire Iridium in $8B Satellite Connectivity Deal
Rocket Lab agreed to acquire Iridium Communications in an approximately $8 billion cash-and-stock deal, combining Rocket Lab’s launch, satellite manufacturing, and space systems business with Iridium’s global satellite connectivity, positioning, navigation, and timing network. Iridium brings global L-band spectrum, a LEO constellation of 66 active satellites and 14 in-orbit spares, and about 2.55 million subscribers across critical infrastructure, maritime, aviation, and government markets.

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?