AvidThink 2026 October Recap

October kept Roy busy with events and more. He moved from CoreWeave’s Fully Connected event in San Francisco to Network Field Day #41 in Santa Clara, then to OCP Global Summit in San Jose, followed by PyTorchCon (co-located with AGNTCon + MCPCon) in San Jose, then to Megaport Connect in Chicago, and wrapped up with Meta’s Product@Scale in Menlo Park. That gave him a breather before heading over to FutureNet Americas in New York this week.

Upcoming events for us in November include FutureNet Americas in New York, which Roy just attended, followed next week by KubeCon North America in Salt Lake City, Network Automation Forum AutoCon 6 in Tucson, then a break for the US Thanksgiving holiday, and then AWS re:Invent in Las Vegas in December.

Meanwhile, our team has been super busy. BTW, if you missed AI Infra Summit last month, check out the full showcase of 30+ videos. For this month, we’re pleased to announce our OCP Global Summit 2026 Showcase with another ~30 videos, featuring names you’re likely familiar with: Ayar Labs, Broadcom, Ciena, Cisco, Credo, d-Matrix, Flex, Keysight, Lightmatter, Lumentum, Marvell (notice how the two names appear next to each other), Molex, NVIDIA, Viavi, in addition to our analyst friends at 650 Group, and companies that deserve your attention: Upscale, Voyant, Xscape, Xsight, BE Networks, Solutional, Rebellions, Advanced Energy, Multilane.

In addition to our OCP Global Summit Showcase, we’re proud to launch our AI for Networking 2026 Showcase, with the third edition of our popular accompanying report. You’re among the first to know, so download your free copy of our ungated report and read it. This year, we re-examine the state of AI-driven operations (AIOps) across multiple networking domains, from enterprise to data center to telcos. We’ve spoken with numerous end customers, vendors, and system integrators and compiled our observations and analysis into a content-packed report. We look at how the industry can unlock the value of agents for network automation to improve network reliability, increase scale, and manage costs.

Download the report and watch the accompanying in-depth videos from companies like Blue Planet (Ciena), Extreme Networks, Gluware, HPE, Nokia, Ribbon/AWS, and more!

Next up are our Data Center Networking Report 2026 Fall Edition (Nov/Dec) and the Enterprise Connectivity Report we’re actively working on.

If you’re interested in participating in any of these (or the others we have in our pipeline covering Telco AI factories, NTN, and Physical AI), drop us a note at [email protected]. Or fill in the form at: https://nextgeninfra.io/contact/.

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

Don’t forget to visit NextGenInfra.io and check out the latest OCP Summit and AI for Networking videos!

—The AvidThink Team

Another busy month on the road for me. You may recall that when we last parted, I was wrapping up at the CoreWeave Fully Connected event in San Francisco. I got to experience the buzz around one of the most prominent neoclouds, and how their special relationship with NVIDIA let them reach the market early with a commercial Vera Rubin offering (which required some data center innovations on their part). From there, it was down to Santa Clara for Futurum Media’s Network Field Day #41 (part of the Tech Field Day series) that’s still helmed by Stephen Foskett and Tom Hollingsworth.

From CoreWeave Fully Connected (San Francisco) to Network Field Day 41 (Santa Clara)

At Network Field Day 41, it was non-stop presentations for four days by a list of established and startup networking companies: 7Signal, Arrcus, Cisco (DC, IT, and Security), Hedgehog, HPE Networking (Campus, Branch, DC), NetAlly, NetBeez, Nokia (DC/IT), OpsMill, VMware by Broadcom, Xsight Labs. My fellow delegates have done an awesome job writing up their observations; you can find them on the NFD 41 page. I am also grateful for the opportunity to participate in the TechField Day series and spend multiple days immersed in the latest in enterprise and data center networking and learning from presenters and fellow delegates. Many of my learnings have made their way into our recent AI for Networking report, and will also show up in our upcoming Data Center Networking Report (Fall Edition).

OCP Global Summit 2026 (15,000 + 3,000 wait list, San Jose, CA)

Huge congratulations to the Open Compute Project team: George, Cliff, Dirk, Steve, James, and the others, on a very successful event. Looking forward to a larger venue next year at Moscone Center in San Francisco, where more attendees can be accommodated.

The expo hall during the first two days was just like Mobile World Congress when it went past 100K attendees. You had to aggressively squeeze through the crowds to get anywhere. If you couldn't attend, check out the videos on our OCP Global Summit showcase to get a feel for the key discussions and topics on the ground. For me, in addition to the power (800V) and energy availability discussions, the angst over data center buildouts, cooling technologies (funny how cold plate liquid cooling is now normal with two phase and immersion on the horizon), the buzz in networking focused on both eliminating and prolonging the life of pluggables with the latest in CPO/NPO, XPO and USFP (the new pluggable format announced by NVIDIA and Cisco); improvements to RoCEv2 in the form of competing standards MRC and MetaRoCE (which really need to be combined somehow); scale-up debates on ESUN and NVLink, with a smattering of UALink/UALoE; inference, CXL and memory wall (and the hot new buzzword, KV Cache).

Adding to that were microVCSEL versus microLED debates (some believe we'll see both for a while), guesses about copper-to-optical transition timelines, and, in general, discussions about the feasibility of slow and wide as an alternative to today's increasingly fast links. Plenty of 1.6T, 3.2T, and beyond discussions on the floor too. Finally, co-design was the new catchphrase across all system builders as the way to achieve ever-higher performance numbers with reasonable energy use and costs.

Anyhow, the experts we interviewed on the ground have many more insights to share, so check out the OCP Global Summit 2026 Showcase on nextgeninfra.io. One last observation: I was struck by how many finance and equities folks were on the show floor. You could tell from how they were dressed (and the questions they asked). A sign of the AI and data center boom for sure. It felt like there were even more of them this time than at OFC earlier in the year.

Funny how I was heading back to the same venue the very next week. Thankfully, there were 12,000 fewer people (yes, I was counting). PyTorchCon had roughly the same number of attendees as last year. For those less familiar with PyTorch, it’s another Meta innovation (like OCP, which Meta/Facebook started) that was donated to the open-source community. It’s the leading machine learning and deep learning framework used for building, training, and deploying AI models. To understand what data center networking demands might look like in the near future, I always find it instructive to look at the workloads running on top. So understanding the innovations in PyTorch, and inference frameworks like vLLM, SGLang, and orchestration like NVIDIA’s Dynamo, or what’s going on with Ray and Slurm, might give hints on the impact of AI workloads on networks. I don’t necessarily understand all the discussions and talks at the conference, but I still find the little I do helpful.

It was straight from San Jose to Chicago that same week for Megaport’s Connect event. I’m grateful to the Megaport team (shout out to Alexis Bertholf for including me) for their super gracious hosting and making the presenters feel like VIPs. It was wonderful to meet the Megaport executive team, including Michael (Chop) Reid, their CEO. Seldom do you see a public company CEO do detailed live technical demos on stage, and this one went very well. And kudos to the Megaport team for executing on their strategy for distributed computing resources (including multiple GPU clusters) post their acquisition of Latitude.sh. I spoke about the evolution of workloads over time (including AI) and their demands on the networks across access, middle mile, backbone, and data centers. You’re welcome to check out the presentation recordings on the Megaport Connect 2026 page or in the YouTube playlist (yes, togas were involved).

I'll share my thoughts on FutureNet North America in next month's newsletter since I have to submit this for scheduling shortly. Suffice it to say that the FutureNet World team did a great job on their inaugural show in North America, pulling together a collection of top telco networking executives over two days. More importantly, many of these execs were willing to share their real-world lessons and learnings candidly. Less of a one-way broadcast about their achievements and more conversations. Well done, Giles, Tom, Collette, Sonal, and the FutureNet team. More on this next month.

Looking forward to heading home from NYC only to turn around on Sunday and head to Salt Lake City, Utah, for KubeCon next week. That's the life of an analyst.

BTW, if you are interested in any of the reports we’re working on (check out our report roadmap), or if you would like to talk and catch up, drop us a line at [email protected].

– Roy for the AvidThink Team

Check out our most recent articles on the AvidThink website:

In addition to the OCP Global Summit 2026 Showcase with many informative videos that our media partner, Jim Carroll of Converge Digest captured, we are glad to announce that our AI for Networking 2026 Showcase and Report is now available.

As part of the 45-page report (27 content pages plus interviews and appendices), we have exclusive interviews with some of the leaders in the AIOps space. Do take a few minutes to download the report and read the interviews — you might find something useful. As always, we do not collect personal information to download the report and it’s complimentary. A huge THANK YOU to the wondreful sponsors who make these reports possible!

6G

Verizon Calls for 6G Uplink Breakthrough to Support AI
Verizon CTO Yago Tenorio said AI applications requiring continuous contextual awareness could push the uplink-to-downlink traffic ratio toward 1:1, demanding architectural advances beyond incremental capacity improvements. Sub-band full duplex could help, while T-Mobile is exploring uplink/downlink decoupling to select spectrum independently for each direction and improve capacity and coverage flexibility.

Ericsson Forecasts Rising AI Uplink Demand, Urges Standalone 6G
Ericsson forecasts a sixfold increase in uplink demand, describing a threefold rise over the next few years followed by a further threefold increase as adoption expands from AI glasses, gaming and cameras to autonomous vehicles, agents and drones. For 6G deployments expected in 2029–2030, the company advocates standalone architecture from launch to avoid 5G’s deployment complexity. Uplink/downlink decoupling could address device power constraints by combining lower-band uplink coverage with higher-frequency downlink capacity.

6G Core Timeline Tightens as Architecture Remains Unsettled
Keysight’s Stephen Douglas expects 6G to launch with a dedicated standalone core, with initial trials around the 2028 Los Angeles Olympics and commercial deployments in 2029–2030. Proposed capabilities include AI-native control, integrated sensing and communications (ISAC), tighter compute integration and native non-terrestrial network support. The core architecture and network functions remain under study and have not been agreed.

EU Allocates €22 Million to Five 6G Projects
The European Smart Networks and Services Joint Undertaking (SNS JU) awarded €22 million (US$24.7 million) across five projects to advance 6G research toward large-scale testing and commercial adoption. 6G-AID will develop a European data foundation for AI-native networks; Hemes-6G will connect eight EU testbeds; and EUnicon-6G will explore connected-device opportunities beyond smartphones. EU-India-6G-Bridge will deepen cooperation with India, while SNS Impact will support cross-project collaboration.

GSMA Backs Global 6G Standards, Warns of AI-Enabled Fraud
The GSMA board called for action on smartphone affordability, network investment and online fraud, highlighting risks from AI-generated content and deepfakes. It projects US$1.2 trillion in operator capital expenditure during 2025–2030 and expects 5G to account for 57% of mobile connections by decade-end. Supporting global 6G standards and harmonized spectrum, GSMA said countries representing more than 80% of the world’s population support mobile use of the upper 6 GHz band, which could provide the 200–400 MHz channel bandwidths anticipated for 6G.

5G

GSA: 98 Operators Have Launched Standalone 5G
More than 390 operators have launched commercial 5G, but only 98 offered standalone (SA) services as of October 2026, according to the Global mobile Suppliers Association (GSA). Slow SA adoption limits network slicing and Reduced Capability (RedCap) services for devices that do not require full 5G functionality; only 17 operators had launched RedCap by September. GSA also reported 41 operators investing in SA-Advanced, including 19 with commercial launches.

Verizon Expands Gigabit FWA With Millimeter-Wave Spectrum
Verizon is deploying 1 Gbps millimeter-wave fixed wireless access (FWA) using 1,700 MHz of previously underused spectrum, complementing metropolitan fiber coverage and freeing mid-band capacity for mobile services. Rolling out this quarter through 2027, the upgrade targets an additional 10–15 million premises beyond Verizon’s existing goal of reaching 40–50 million premises by 2032; the company also reiterated its target of 32 million fiber passings by year-end. Verizon added 159,000 FWA customers in Q3, bringing its total to approximately 6.36–6.37 million.

C Spire Uses AI Capacity Planning to Reach 500,000 More Homes
C Spire expanded 5G FWA availability to 500,000 additional Mississippi homes through improved AI-assisted capacity planning and upgrades to 36 of its 1,300 cell sites. Its in-house predictive machine learning tool combines Nokia RAN and transport-vendor data in an interactive map showing capacity forecasts, subscriber penetration, churn and other metrics. Launched in January 2025 to serve locations beyond its fiber footprint, the service uses these forecasts to identify spare capacity and support additional customers.

Satellite

FCC Proposes 25 MHz Auction and Expanded Satellite D2D Access
The FCC proposed auctioning 25 MHz across the 1675–1695 MHz and 2020–2025 MHz bands for terrestrial, direct-to-device (D2D) or hybrid networks, alongside rule changes opening another 482 MHz to Supplemental Coverage from Space (SCS). Auctioning 1675–1695 MHz depends on an NTIA study, while the SCS changes require public comment. CTIA, AT&T, T-Mobile and Verizon oppose including 2.5 GHz and 3.45 GHz, warning that satellite frequency-division duplex (FDD) operations could interfere with terrestrial time-division duplex (TDD) networks serving 300 million users, particularly their uplinks.

FCC Approves Amazon Leo’s US$11.57 Billion Globalstar Acquisition
The FCC approved Amazon Leo’s US$11.57 billion acquisition of Globalstar, providing approximately 25 MHz of internationally harmonized mobile-satellite spectrum for D2D services and adding Globalstar’s XCOM RAN private 5G business. Amazon and Apple also agreed that Leo will power satellite services for iPhone and Apple Watch, including Emergency SOS. The FCC cited benefits for underserved coverage and emergency communications, while Amazon confirmed in July that Leo service would launch later this year.

Private Wireless

Nokia to Sell Enterprise Campus Edge Business to Aurelius
Nokia agreed to sell its Enterprise Campus Edge (ECE) private 5G business to Aurelius for undisclosed terms, with closing expected in the first half of 2027. ECE serves nearly 900 customers across energy, mining, manufacturing, logistics, airports and ports; the sale follows Nokia’s November 2025 announcement and supports its focus on larger defense, railway and utility contracts while reducing competition with carrier customers. Aurelius is described as employing approximately 1,000 people across Finland, the US, Hungary and India, with annual sales exceeding US$190 million across Europe, the Americas, APAC and the Middle East.

Ericsson Reorganizes Enterprise Wireless, Retains Private 5G
Ericsson is dissolving its enterprise wireless unit and transferring much of the business, including private and neutral-host 5G portfolios, into its networks division. Unit head Asa Tamsons will support the transition before leaving in 2027. The announcement coincided with Nokia’s private-network divestiture and followed essentially flat year-over-year enterprise sales in Ericsson’s Q3 results.

Dell’Oro: Private Wireless RAN Growth Slows to Mid-Single Digits
Global private wireless RAN revenue grew at a mid-single-digit rate in the first half of 2026, down from double-digit growth in 2025, prompting Dell’Oro Group to cut its forecast again as campus deployments lagged expectations. Huawei, Nokia, Ericsson, ZTE and Samsung represented approximately 95% of market revenue. Huawei led globally, while Nokia and Ericsson dominated outside China.

AI-RAN

T-Mobile, Nokia and Nvidia Complete AI-RAN Call at Commercial Site
T-Mobile, Nokia and Nvidia completed what they described as the world’s first AI-RAN call at a commercial cell site in Bellevue, Washington, following months of joint engineering. The call connected a commercial smartphone to T-Mobile’s existing 5G standalone core using Nokia massive MIMO radios and CUDA-accelerated anyRAN software on Nvidia Aerial RAN Computer servers. BT, Deutsche Telekom and Elisa also announced AI-RAN pilots, with BT planning live Birmingham trials in 2027.

Nokia and SoftBank Plan AI-RAN Field Trial Under Commercial Traffic
Nokia and SoftBank will conduct an AI-RAN field trial in Japan in 2027 under commercial traffic conditions, combining Nokia’s AI-native anyRAN cloud RAN software, SoftBank’s AITRAS platform and orchestrator, and Nvidia’s ARC platform. The trial will place radio and AI workloads on shared infrastructure to test improved utilization and potential new services. Both companies position it as a step from laboratory validation toward commercial deployment.

Ericsson Defends Custom Silicon for AI-RAN, Cites 10% Efficiency Gain
Ericsson’s Per Narvinger defended its custom ASIC approach to AI-RAN, citing rapid feature delivery, competitive costs and approximately 10% higher spectral efficiency, while leaving open future GPU adoption. Three operators have deployed its AI-native link adaptation across 60,000 base stations and are seeing spectral gains. The approach contrasts with Nokia’s initial GPU deployments at a handful of operators this quarter.

Elisa Demonstrates Voice, Data and Video Analytics on Shared AI-RAN
Elisa reported what it called Europe’s first live-network demonstration of simultaneous voice, mobile data and video analytics on one AI-RAN platform over its 5G-Advanced network, supporting commercial deployment plans for 2027. Building on implementation plans announced in March, the demonstration used Nokia’s AI-RAN solution and Nvidia Aerial RAN Computer with RTX PRO 4500 Blackwell Server Edition GPUs. Elisa also reported 61% 5G penetration, up three percentage points during the quarter.

AT&T and Verizon Report Operational Gains From AI in the RAN
AT&T and Verizon executives said AI-based RAN deployments are delivering measurable results as operators advance toward larger-scale testing. AT&T uses AI across planning, forecasting, deployment, configuration, optimization and lifecycle management, and is testing AI-native functions embedded in the RAN. Verizon combines centralized, non-real-time rApps with real-time AI inside the gNodeB; work with Ericsson has improved downlink spectral efficiency through AI-based link adaptation, while a separate field trial improved coverage.

AI-RAN Alliance Prioritizes Implementation and Edge Business Models
The AI-RAN Alliance is expanding beyond individual demonstrations to address workload placement, shared infrastructure and the commercial case for distributed AI services on mobile networks. Priorities include edge inference, physical AI, semantic communications, multimodal RF sensing and integrated sensing and communications (ISAC), with security and AI workload placement emerging as strategic concerns.

AI

Deutsche Telekom Extends Agentic AI Beyond the RAN
Deutsche Telekom is extending agentic AI into broader network automation, with RAN Guardian live in Germany and the cross-domain MINDR platform under development. RAN Guardian prepares networks for major events through agents that scan public and social media, categorize events and assess their likely network impact. A remediation agent can then take preventive action before events disrupt normal traffic patterns.

Nokia Q3 AI and Cloud Sales Rise 123%
Nokia reported 123% year-over-year growth in AI and cloud sales, US$1.7 billion in new quarterly orders and an expanding backlog. Total net sales rose 10% to almost US$5.8 billion, net profit increased 95% to US$174.8 million, and Mobile Infrastructure revenue grew 5% to US$3.3 billion. CEO Justin Hotard highlighted AI-RAN development for physical-world AI applications and said Infinera integration remains on track for completion by year-end.

T-Mobile and Ericsson Test AI Beamforming on 5G-Advanced
T-Mobile US and Ericsson are testing AI-Managed Beamforming at scale on live 5G-Advanced commercial traffic, with initial results showing the largest gains during peak demand. The approach optimizes sounding reference signals (SRS)—brief handset transmissions used to measure radio paths—to direct beams more accurately, using models trained on T-Mobile’s own network data. T-Mobile views its nationwide 5G-Advanced network as a proving ground for the sensing and adaptive capabilities needed for distributed AI, robotics and physical AI in 6G.

Ericsson Reports 7% Organic Growth, Warns of AI Component Costs
Ericsson beat Q3 earnings forecasts and reported 7% organic sales growth, but warned that AI-driven component costs will pressure margins through 2027; some customers accelerated purchases ahead of price increases. Facing a flat market, it is reorganizing enterprise wireless and directing some R&D toward satellite networks as providers such as Starlink expand. A new APAC sales organization will combine North East Asia, South East Asia, Oceania and India under current EMEA leader Patrick Johansson.

Cloud

Lambda Sees Opportunity in Hyperscaler Capacity Shortages
Lambda head of cloud services Dave Ward said hyperscalers including AWS are turning away large enterprises because of capacity shortages, creating an opening for neocloud providers. Lambda is accelerating data center additions over the next 12–18 months, although retaining customers after shortages ease remains uncertain. Ward said neoclouds need a competitive baseline software platform plus differentiated capabilities, such as specialized storage or workload-specific stack optimizations.

Vultr: Agentic AI Drives CPU Demand Toward GPU Parity
Vultr CMO Kevin Cochrane said agentic AI is driving CPU demand nearly one-to-one with GPUs, while enterprise buyers increasingly seek complete platforms rather than bare-metal infrastructure. Vultr is expanding its enterprise AI platform with identity and access management, upgraded cluster provisioning, bootable storage and cloud VPN capabilities.

SD-WAN

HPE Extends Marvis AI Across Aruba, Mist and SASE
HPE claims its Marvis AI engine unifies Aruba and Mist into an autonomous network capable of making decisions and executing tasks without human involvement, approximately 16 months after its US$14 billion Juniper acquisition. Network World reports that HPE has also integrated SASE products with Mist AI through the EdgeConnect suite, combining SD-WAN and security service edge. The integration supports wired and wireless network control and troubleshooting through a single AI-based management platform.

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?