Note: This month’s newsletter is a little longer than usual and may be clipped on your Gmail client. We apologize, but given the rich set of updates, we felt it best to leave the full content in place rather than attempt aggressive edits.

The month of May saw us attend another successful FutureNet World in London (congrats to the team for another excellent show), the Red Hat Summit virtually, an almost impromptu MCP Developer Summit, and the ONUG AI Networking Summit in Dallas.

NextGenInfra En Route To Series

As we shared in last month’s newsletter, Jim Carroll (our media partner) beta-tested our new “En Route To” series at OFC 50, and we welcomed our first batch of interviewees during the recent RSAC Conference at the Moscone Center. As a tip of our hat to the increasing prevalence of AI in our lives, we conducted chats with leading executives in infrastructure technology (Infratech) whilst riding self-driving Waymo cars on the streets of San Francisco. Check out our En Route To series on the NextGenInfra Channel on YouTube and watch our inaugural interviews with Larry Lunetta of HPE, Anubhav Arora of Ericsson Enterprise Wireless Solutions, Galeal Zino of NetFoundry, and Stephan Rettenberger of Adtran. If you’re interested in participating in this series, drop us a note at [email protected] (or [email protected]).

Report on AI-RAN

If you haven’t yet downloaded our report on the opportunity for telcos in AI RAN, you can find it on SoftBank’s website here. It’s what we hope to be an optimistic but realistic take on the opportunity. On a related topic, we recently hosted a webinar with the SoftBank team on large telecom models and their use cases. The SoftBank team was kind enough to share their learnings from training and fine-tuning a model on their extensive catalog of in-house data. They also discussed early use cases and what they plan to do next. Check out the on-demand recording of the webinar on Large Telecom Models and their Use Cases on the Fierce Network website.

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

Roy’s Travelogue — FutureNet World London, Red Hat Summit, MCP Developers Summit, ONUG AI Networking Summit
Latest Reports — AI in the Mobile RAN
Recent Articles— The Role of Observability and DPI in Open RAN
NextGenInfra — OIF 448G, En Route To
April News Roundup — AI investments continue – Middle East pivot
Other Content — AvidThink in the News, Webinars, Meet Us Live

BTW, sign up and watch the AI and the Network Fierce Network Research Virtual Summit

running now from June 3 – June 4, 2025 PDT. Roy’s on a panel on June 4 (Wed) at 1040 PDT discussing Making AI Work at the Edge in the Age of Inferencing. The same link will get you on-demand access post event.

In addition, Roy will be joining Ericsson Enterprise Wireless Solutions’ Camille Campbell and Nathan Herman on June 24 to discuss technologies and strategies shaping the next generation of connected enterprise sites as part of their Ericsson Unwired (previously Cradlepoint Unwired) Webinar series on the Designing the Branch of Tomorrow. Sign up today!

Roy’s Travelogue

May was slightly busier than April in terms of travel and conferences. We’ll start with the FutureNet World 2025 conference in London, which saw record attendance (over 700 execs) — congratulations to Giles Cummings, Tom Winter, and the team for putting together another outstanding show. As an indication of their success, the show will be moving to a larger venue, the Intercontinental, near the O2 Dome and Canary Wharf, in 2026.

Key takeaways from FutureNet World (in addition to AI, GenAI, and Agentic AI) include:

Data, data, data — Data is crucial for deriving value from any AI, from traditional to agentic. There is no shortcut (though AI can assist with data preparation and pre-classification) to identifying sources of data, cleansing data, and building a proper data pipeline.
Cloud, cloud, cloud — Building a cloud-type infrastructure (cloud-native, cloud-first, cloud-forward) is necessary, as is cultivating an agile cloud mindset. It doesn’t matter whether it’s a public cloud-first strategy, a private cloud-first approach, or a hybrid approach (the most realistic).
AI is like any transformation — there’s typical change management, cultural transformation (which includes hiring and firing), and upskilling or reskilling for AI savviness.
AI is unlike any transformation — Prior technology waves promised higher efficiencies, which eventually led to in-house labor force reduction (outsourcing, offshoring, SaaSifying, use of public cloud – a form of outsourcing). However, those technologies didn’t include one-to-one replacements for human workers. With AI, there could be a direct replacement for specific tasks or even entire human roles. While the open conversation at the level of governments and technology leaders is improving, we’re still employing platitudes that we will “upskill/reskill” to pretend that there will be no layoffs. I was most appreciative that the telco leaders at FutureNet World were candid about that possibility and more open to a thoughtful conversation than at other conferences.

For more takeaways, watch the analyst panel closing keynote on-demand video below in the Upcoming and On-Demand Webinar section with fellow analysts Amy Cameron (STL Partners), Robert Curran (Appledore Research), and I discuss our thoughts.

The Red Hat Summit for this year was in full swing (according to our friends Amar Kapadia of Aarna, Hanen Garcia of Red Hat, and others who were there in person). For us analysts, the key announcements include Red Hat Enterprise Linux (RHEL) 10 with better AI integration via RHEL AI, Lightspeed command line assistant, and Image Mode for containerized OS updates (increased stability and security). Other improvements included support for post-quantum cryptography (PQC) and other security enhancements. Red Hat enhanced OpenShift AI’s capabilities with optimized model catalogs, improved compression tools for model optimization, multi-node training, and enhanced GPU resource management. OpenShift Virtualization introduced improved parallelism features, enhanced networking, security cluster integration, and compliance operator support.

For me, the key news was the release of llm-d, an open-source distributed inference framework built on Kubernetes. llm-d’s release, combined with Red Hat’s acquisition of Neural Magic — the team behind the vLLM open-source project, a high-performance and memory-efficient library for inference and serving — makes Red Hat a player to watch in the enterprise inference space. CoreWeave, Google Cloud, IBM Research, and NVIDIA are founding contributors of llm-d, along with AMD, Cisco, Hugging Face, Intel, Lambda, and Mistral AI as partners. Notably, llm-d incorporates the NVIDIA Inference Xfer Library (NIXL), which helps accelerate communications in AI inference frameworks for NVIDIA GPUs. NIXL is also part of NVIDIA’s Dynamo, NVIDIA’s inferencing framework, which Jensen Huang highlighted during his GTC keynote a few months ago. I’m watching to see how llm-d and Dynamo play out.

A one-day fun summit for me was the MCP Developers Summit, which came together in under four weeks. Spearheaded by the folks from Acorn Labs (the same ones who gave us Rancher and who spun out from SUSE after Rancher was acquired), the rapidly sold-out conference brought together MCP leaders from Anthropic, OpenAI, Microsoft, Cloudflare, AWS, Postman, and Okta, and end-users deploying MCP at scale from Block and Paypal. A16Z (Andreessen Horowitz) brought an investor’s perspective. Anthropic created MCP (Model Context Protocol) to provide a standardized way for AI models to access resources and tools — e.g., search and retrieve information from filesystems, integrate with your notes library, Git repository, etc. Other AI companies, such as OpenAI, quickly adopted it, and MCP has become a hot new topic on the AI circuit. It’s only a few months old, but I’ve experienced substantial value in using it to attach the Claude Desktop client to my filesystem, Obisidian notes database, and other resources.

There’s plenty of work remaining to flesh out and improve MCP, including authentication, authorization, discovery/directory, security, and remote MCP support, among others. However, based on the discussions at the conference, numerous efforts are underway by highly motivated and innovative technologists. Google recently announced its Agent-to-Agent (A2A) protocol, designed to complement MCP. Meanwhile, IBM’s ACP (distinct from Cisco’s ACP) and other proposals are likely to struggle to gain mindshare in this busy space. Google’s A2A, IBM ACP, Cisco ACP, and other similar proposals downplay MCP as a means for models to access resources and relegate it to agent-to-agent communication. However, MCP proponents don’t see it that way. The MCP camp has visions of expanding MCP from its original capability for local services to remote services (which is already operational) to agent-to-agent services in the longer term. They discuss directories for agent discovery and more advanced permission management. MCP has momentum, and just as Kubernetes overwhelmed Docker Swarm, Apache Mesos, etc, during its time, we could see MCP take the lead. Additionally, seeing some prominent figures from the Linux Foundation at the Summit (yes, Jim was there) leads me to speculate that the MCP protocol may find a home at the Linux Foundation.

Ironically, heading to the ONUG AI Networking Summit in Dallas, MCP and A2A followed me. MCP-related discussions also abounded at the AI Networking Summit, where enterprise networking leaders discussed AI network automation. Nick Lippis, Steve Collins, Paul Agranat, Tony Farinacci, and the rest of the ONUG team put together an excellent set of content and human networking opportunities for enterprise networkers. The highlight was seeing John Chambers join via video conference for a fireside chat with Nick. He’s still the super-positive, classy statesman that he used to be while leading Cisco. Most instructive for me was talking to young networking and IT professionals and hearing their excitement and concerns about the impact that AI is having and will have on their careers. I worry that, as an industry, we’re not doing enough to openly discuss and think through the implications that this acceleration in generative AI and AI agents will have on the workforce and society.

I’m always happy to discuss any of the topics above further. Drop us a line at [email protected].

– Roy for the AvidThink Team

________________________________________________________________________________________________________________________

2025 May News Roundup

AI/GenAI

Dell Emphasizes AI and Disaggregated Data Centers at Annual Conference
At Dell Technologies World, AI was a central focus, with Dell introducing the Dell AI Factory with Intel, adding to existing collaborations with NVIDIA and AMD. Dell also highlighted its vision for disaggregated data centers, aiming for greater flexibility by creating independent pools of storage and computing resources accessible on demand. The company is also working to make its VxRail HCI products compatible with other vendors like Red Hat OpenShift and Nutanix’s AHV.

U.S. Government Proposes 16 Federal Sites for AI Data Center Development
The U.S. government plans to offer 16 federal sites for potential data center development to bolster the nation’s AI capabilities, seeking collaboration with major tech companies like hyperscalers and OpenAI. OpenAI provided input on site requirements for AI supercomputers, emphasizing access to roads, water, power, and a skilled local workforce. The government is considering incentives such as tax breaks and streamlined permitting, with organizations like the Data Center Coalition (DCC) and INCOMPAS stressing the importance of such support.

Swedish Consortium, Including Ericsson and Nvidia, to Establish AI Factory
A Swedish consortium comprising Ericsson, AstraZeneca, SAAB, SEB, and Wallenberg Investments AB is creating an AI factory to drive AI innovation and competence in Sweden. The initial phase will feature two NVIDIA DGX SuperPODs with Grace Blackwell GB200 systems, becoming Sweden’s largest enterprise AI supercomputer for tasks like training domain-specific models and large-scale inference. Nvidia will also establish its first AI Technology Center in Sweden to collaborate on AI research.

OpenAI’s Project Stargate Aims for $500B Investment in AI Data Center Infrastructure
OpenAI’s Project Stargate is a major initiative to build extensive data center infrastructure for advanced AI, targeting an initial $500 billion investment over four years for facilities in the U.S. and internationally, with partners like SoftBank, Oracle, and MGX. The first U.S. facility in Texas, featuring two buildings with 50,000 Nvidia chips each, is expected to be operational in the first half of 2025. OpenAI has also launched “OpenAI for Countries,” with Stargate UAE in Abu Dhabi being the first international collaboration, a 1 GW facility part of a larger 5 GW campus.

McKinsey Report Advises Telecom Operators on AI Integration Strategies
A McKinsey & Company report guides telecom network operators on integrating AI into their businesses, warning that inaction poses the most significant risk, potentially leading to missed growth opportunities. The report highlights various AI implementation examples by telcos, such as Verizon’s AI Connect product and Lumen’s AI-traffic-focused fiber network. Other operators are exploring AI applications in their radio access networks (RAN).

KT Corp and Viettel Group Announce $95M AI Partnership in Vietnam
KT Corp and Viettel Group have signed a $95 million deal to collaborate on AI, forming a joint venture to build a dedicated AI data center and GPU server clusters to enhance Vietnam’s cloud and AI infrastructure. The partnership will also develop AI products for the Vietnamese market, including a national AI model and Vietnamese-language AI agents. KT will introduce AI training courses and establish a global development center in Hanoi.

5G

International Research Team Demonstrates 5G Non-Terrestrial Network Satellite Transmission
An international research team, including Singapore University of Technology and Design (SUTD) and SKY Perfect JSAT (JSAT), successfully transmitted a 5G signal from Singapore to Japan via a geostationary (GEO) satellite using an electronically steered antenna (ESA). This demonstration at World Expo 2025 in Osaka highlighted the potential of 5G Non-Terrestrial Network (NTN) GEO communications for industries like maritime and autonomous vehicles, and for connecting low-Earth orbit satellites. This was reportedly the first instance of a GEO satellite transmitting a 5G signal with an ESA.

T-Mobile Achieves 550 Mbps Uplink Speed in Commercial 5G Advanced Network
T-Mobile, in collaboration with Nokia and MediaTek, announced a record uplink speed of 550 Mbps in a commercial setting using 5G Advanced technology in sub-6 GHz spectrum. The test utilized 100MHz of TDD spectrum (n41) and 35MHz of FDD spectrum (n25). T-Mobile emphasizes the growing importance of uplink speeds for applications like 4K video uploads, real-time gaming, and virtual reality.

European Operators Increasingly Adopting 5G Standalone (SA) Technology
European telecom operators are increasingly deploying 5G Standalone (5G SA) networks, with a focus on enterprise applications such as network slicing and automation. Virgin Media O2 launched its 5G SA network in major UK cities in February 2024, and BT has enabled 5G SA for a significant portion of the UK population. In Italy, TIM is adopting a hybrid approach, deploying both 5G SA and non-SA, believing the SA architecture will enable new applications like automation and network exposure.

Verizon CEO Reports Growing Demand for Private 5G Networks
Verizon CEO Hans Vestberg noted a significant increase in enterprise interest for private 5G networks across retail, manufacturing, and financial services, a notable change from a year prior. He attributed this to 5G Standalone (SA) technology, enabling Verizon to deliver private 5G network slices more efficiently, within days rather than weeks. Verizon’s 5G SA core is currently deployed in 60% to 70% of its network, with a C-band mobility deployment focus aiming for 80% to 90% coverage by year-end.

China Mobile Leads 5G-Advanced Deployment, Investing $1.4B for Upgrades
China Mobile is spearheading 5G-Advanced deployment in China, having invested approximately $416 million to cover 300 cities and planning an additional $1.4 billion to upgrade over 400,000 base stations this year. Reports indicate the 5G-A icon is appearing on user devices in some areas. Competitors China Telecom and China Unicom also have 5G-Advanced rollout plans, with China Telecom targeting 330 cities by October and China Unicom aiming for 300 cities by year-end, focusing on consumer, enterprise, and low-altitude services.

EE Deploys Small Cells in London to Enhance 5G Coverage
EE partnered with Ontix to deploy 80 new 5G small cells in Westminster, London, integrating them into existing street infrastructure like lampposts and BT kiosks to improve 4G and 5G network performance in dense urban areas. This initiative supports EE’s goal of nationwide 5G coverage by 2028, using network analytics for optimal small cell placement. EE’s 5G SA network now offers at least 95% coverage across 15 U.K. cities.

Vodafone Idea Accelerates 5G Rollout, Targeting 17 Service Areas by August
Indian operator Vodafone Idea plans to make 5G available in 17 of its 22 service areas by the end of August, expanding from the greater Delhi area to Bengaluru and Mysura next. The company reports that 70% of eligible users in Mumbai are using 5G, which handles about 20% of total data traffic there. Vodafone Idea intends to invest approximately $6.4 billion in its 5G rollout over the next three years, while competitors Reliance Jio and Bharti Airtel completed their nationwide 5G deployments in 2024.

Open RAN

Zain Kuwait to Deploy Rakuten Symphony’s O-RAN and Cloud Platform for 5G SA
Middle Eastern operator Zain will utilize Rakuten Symphony’s Open RAN software and cloud-native virtualized software platform for its 5G standalone (SA) service in Kuwait. The companies will begin with a pilot project to establish cloud-native Open RAN. The SA deployment will leverage Zain’s existing base stations, data center infrastructure, and SA core network.

6G

Apple Intensifies 6G Preparations with Strategic Hiring
Apple is actively recruiting for 6G-related positions in locations including Israel and San Diego, California, signaling a proactive approach to the next-generation cellular standard, which is not expected commercially until around 2030. This contrasts with Apple’s later entry into 5G modem development. Experts emphasize the importance of acquiring 6G talent early for patent and intellectual property development.

Cloud

Middle East Emerges as a Hub for Sovereign Cloud Initiatives
The Middle East is becoming a key region for sovereign cloud offerings, fueled by significant investments in AI and data centers. Recent collaborations include Google Cloud with Accenture in Saudi Arabia, stc Group with SambaNova for an AI-specific sovereign cloud, and Microsoft with Core42 for Abu Dhabi’s government. AWS and Nvidia are also involved in AI infrastructure projects in Saudi Arabia, while Bain Capital launched hscale, a hyperscale data center provider for the region.

Dell Introduces Private Cloud Solution with Software Flexibility
Dell Technologies unveiled Dell Private Cloud, a new offering available in the second half of the year, designed to prevent vendor lock-in by allowing enterprises to use their preferred software from vendors like Broadcom, Red Hat, or Nutanix on Dell’s disaggregated hardware. The solution features zero-touch onboarding for compute and storage, centralized inventory management, Dell AIOps, and automated provisioning. This setup allows enterprises to decommission and reprovision Dell hardware with different software as needs evolve.

Alibaba Cloud Invests $52.7 Billion in Global Cloud Expansion
Alibaba Cloud is investing $52.7 billion to establish a unified global cloud network, providing AI infrastructure services for Chinese enterprises both domestically and internationally, a move accelerated by U.S.-China tensions over AI hardware. The company plans to focus on strategic markets including Japan, South Korea, Southeast Asia, the Middle East, Europe, and the Americas, expanding from its current 87 availability zones in 29 regions. Analysts project Alibaba Cloud’s revenue could double to $33.3 billion by 2028.

Private 5G

AWS Sunsets its Private 5G Service Offering
Amazon Web Services (AWS) has discontinued its AWS Private 5G service, which was launched in 2022 to simplify private mobile network deployment for businesses. AWS cited challenges such as limited wireless spectrum availability and reliance on third-party hardware as reasons for not meeting the “original vision.” However, AWS will continue offering its private wireless on AWS solution through service provider partnerships, integrating private 5G/4G LTE with AWS services.

Industry Experts Highlight AI’s Critical Role in Scaling Private 5G Networks
While private 5G networks can support basic automation without artificial intelligence, AI is deemed crucial for scaling these networks and realizing their full potential, according to panelists at a recent FutureNet event. Telecom companies view AI as essential for advanced private 5G functionalities like threat detection, load balancing, resource optimization, and self-healing networks. Telcos may also offer domain-specific AI services alongside private 5G, with global deployments expected to rise significantly by 2030.

Celona CEO Highlights Growth in Healthcare and the Impact of RedCap and AI on Private Networks
Rajeev Shah, CEO of private network startup Celona, reported international growth, including collaborations in Saudi Arabia and the U.S., where partnerships with T-Mobile and AT&T have facilitated entry into the healthcare sector, using private networks as an overlay to Wi-Fi for critical communications. Shah anticipates that Reduced Capability (RedCap) 5G for lower-cost IoT devices and AI will be vital for future private network expansion.

Industrial Network Owners Form 5G-OT Alliance to Advance Private 5G
A diverse group of companies, including John Deere, Hamburger Containerboard, BASF, Miami International Airport, and OneLayer, has formed the 5G Operational Technology Alliance (5G-OT Alliance). The alliance aims to accelerate private 5G and LTE network deployment in operational technology environments by sharing best practices, developing practical use cases, and creating standardized implementation guidelines and security frameworks. Membership is open and free to other interested companies.

SASE/SD-WAN/ZTNA

Arista Reportedly Considers Acquiring VeloCloud from Broadcom
Arista is reportedly planning to acquire VeloCloud from Broadcom for nearly $1 billion, according to The Information. Analysts suggest this acquisition would broaden Arista’s portfolio, making it more competitive with Cisco in the enterprise cloud market, and VeloCloud’s managed SD-WAN product would enhance Arista’s existing SD-WAN offerings. Broadcom, which acquired VeloCloud in 2023, is reportedly looking to sell as SD-WAN is not considered a high-growth area and it focuses on its VMware acquisition.

T-Mobile Launches Managed SASE Solution with Palo Alto Networks
T-Mobile is partnering with Palo Alto Networks to offer a managed Secure Access Service Edge (SASE) solution to its T-Mobile for Business customers. The solution integrates T-Mobile’s 5G Advanced network, T-SIMsecure, and T-Mobile Security Slice with Palo Alto Networks’ cloud-native zero-trust Prisma SASE 5G. This collaboration aims to enhance network and device security for connections to private, SaaS, and internet applications without requiring additional equipment.

Thetford Leverages Aryaka’s Unified SASE for Cost Reduction and Network Management
Thetford, a manufacturer for recreational vehicles and marine applications, adopted Aryaka’s unified SASE solution to manage its global network and cloud presence, resulting in lower operational costs and improved networking services. By outsourcing its network to Aryaka, Thetford reduced SD-WAN management overhead and licensing costs. The Aryaka deployment offers a single interface for managing network security, including policy monitoring, web filtering, intrusion detection, and next-generation firewalls.