Artificial Intelligence Monthly Insights

KDDI, in collaboration with Osaka city officials and robotics vendor Avita, will run a public trial starting 31 August of a humanoid robot assisting passengers with information on the Osaka Metro, powered by Google’s Gemini AI model.

Yandex introduced Yandex SIM, a mobile virtual network service embedding an AI assistant into calling to analyze conversations, flag potential fraud, and later personalize plans based on user needs. No host network or technical architecture details were disclosed.

On Aug. 28, SK Telecom stated it was chosen to lead operation of the national 'AI for Everyone' program in South Korea, indicating a formal role in scaling AI access and services.

Mavenir partnered with Sanas to embed Sanas’ real-time voice processing engine into Mavenir’s MAVcore Voice AI portfolio, allowing mobile operators to process, enhance, and secure voice calls within the network rather than on devices. The design targets compliance with sovereign AI and data residency requirements.

LG’s creative agency HSAD applied AI to behavioral data to identify higher-propensity prospects for a specific item. Campaigns using AI-targeted ads achieved a 2.1x conversion rate versus general commercials, with average purchase value up 18%.

LG Uplus connected its 5G network to a Yonsei University AI-RAN testbed powered by Nvidia GPUs and verified that mobility handoffs can be maintained for moving physical AI and robotic devices.

The UK Home Office and National Police Chiefs' Council co-developed an AI service with Vodafone Business acting as the delivery partner.

At MWC Barcelona 2026, TM Forum introduced three core projects under its AI-Native Blueprint to standardize and scale AI operations for CSPs: 1) Model as a Service (MODaaS) defining enterprise-grade model sourcing, governance, lifecycle, and a common control plane across cloud/edge/on-prem; 2) Data Products Lifecycle Management (DPLM) establishing domain taxonomies, data product contracts, and metadata standards to make data discoverable and agent-accessible; 3) Agentic Interactions Security defining a policy language and ontology to secure agentic AI interactions at scale. Developed collaboratively with global members; MODaaS is referenced by the World Economic Forum.

AT&T indicates it is in the later stages of migrating Lumen fiber customers onto AT&T Fiber and highlights growing enterprise interest in AI with attention to open‑source models.

Anthropic updated its support documentation confirming that all AI models launched on or after 2 August will embed watermarks in generated text to meet the EU AI Act’s Article 50(2) Code of Practice on transparency, enabling other digital systems to identify AI-generated or AI-processed content.

NTT DOCOMO and Samsung are collaborating to extend AI-driven RAN functions to the device layer, placing on-device intelligence in smartphones to assist RAN optimization and efficiency within DOCOMO’s network.

LG Uplus trialed an edge-executed AI system to manage and self-heal quality on a Wi‑Fi 8-compatible home broadband setup. The AI runs on the router rather than the cloud, enabled by Broadcom’s Wi‑Fi 8 platform with an integrated neural processing unit. The operator adapted its existing cloud-based AI to operate on-device.

Zain introduced an AI-powered calling translation program that provides real-time Arabic translation during voice calls, marking the first such service by a Kuwaiti operator.

Ooredoo, with Nvidia, Nokia, and IOH, is launching Zankore, a neocloud GPU-as-a-Service platform in Indonesia, targeting 1 GW of AI capacity over three years. Initial 200 MW is slated for H1 2027, sourced from multiple Indonesian data centers, with blue-chip customers already secured. Ooredoo holds 49% and is committing $800M over five years; funding combines equity and debt. The platform uses Nvidia technology to optimize GPU fleet utilization, targeting up to 40% more effective compute.

A joint TM Forum Catalyst, championed by Deutsche Telekom and Vodafone, demonstrated agentic AI to automate translation of dense documentation, suggesting applicability for streamlining telecom processes and knowledge workflows.

Nvidia announced SpaceXAI will adopt its 'Vera' CPUs to accelerate agentic AI workloads and support planned orbital AI data center deployments.

Vodafone and Google Cloud outline cloud-executed agentic AI to automate operations in autonomous mobile networks, referencing 3GPP-aligned approaches.

IOH and Nvidia established an AI research hub in Indonesia to co-develop and accelerate AI capabilities for telecom and the local ecosystem using Nvidia’s accelerated computing platform.

Analyst commentary outlines ten agentic AI trends for telecom, asserting AI agents will assume roles across customer interactions, procurement, workforce tasks, and network operations. No specific deployments, vendors, or geographies are cited.

A group of over 120 organizations, including Nvidia, Cisco, and CrowdStrike, proposed a standardized incident-reporting framework to document and share security and operational incidents involving AI agents.

BizTech Magazine profiles Honeywell’s view on moving from AI copilots to autonomous agents on factory floors. Agents execute approved workflows, interact with OT systems, and verify outcomes within deterministic engineering and safety constraints. Recommended early use cases include sensor fault handling, routine startup/shutdown, weather responses, field comms, shift handovers, logbook updates, and process optimization. Human oversight remains for ambiguous, safety-critical, regulatory, or novel scenarios. Strong cybersecurity controls (segmented networks, tightly managed access, continuous monitoring, system validation) are emphasized.

Alphabet has curtailed or paused aspects of its AI image generation capability after safety and accuracy concerns were raised, adjusting availability and outputs while it conducts further review.

Ciena-commissioned survey (Censuswide, July 13–23, 2026; n=1,200 global, 100 in India) reports 67% of Indian respondents strongly agree AI-driven network services will be a primary revenue driver in 3–5 years (27% somewhat agree). 98% expect managed optical fiber network services (MOFN) to generate revenue from interconnecting distributed AI compute clusters within three years (48% significant). 88% say optical network upgrades are urgently needed to meet low-latency enterprise demands (38% want major upgrades now). 93% view next-gen optical (higher-capacity wavelengths, power‑efficient coherent optics) as critical to profitable AI connectivity; 94% rate advanced network automation as important/critical. 88% expect GPU-as-a-Service, edge compute, localized AI inference, and private wireless to contribute ≥10% of enterprise revenue within three years (35% >20%). 59% expect physical AI (robots, machine vision, remote control) to exceed 15% of AI enterprise revenue within five years.

Survey results indicate most telecom operators anticipate AI workloads will necessitate accelerated network upgrades, with 96% citing the need for advanced network automation—including agentic AI—to manage operations and performance.

AT&T implemented model routing with LiteLLM in its internal 'Ask AT&T' platform, directing routine coding and employee queries to open-source models while reserving complex tasks for proprietary providers. The shift reduced AI coding costs by 56% with an estimated ~2% quality impact, processing ~45B tokens/day. Open-source models (Nvidia Nemotron, Meta Llama, Google Gemma) now handle ~40% of requests (targeting 60–70%), and telecom-tuned models trialed Feb–Jul 2026 delivered up to 90% lower inference costs for industry-specific tasks.

At TM Forum DTW Ignite 2026, Avenga emphasized moving telecom operations to outcome-based AI with clear decision rights, incremental deployment, and strong governance to enable enterprise-scale adoption.

Telecom operators are moving from isolated AI use cases toward AI-native networks that embed intelligence across RAN, core, edge, and service orchestration. Cloud-native virtualization, disaggregated architectures, coordinated AI agents, strong governance, and measurable operational outcomes will be critical to improving efficiency, resilience, energy use, and network performance at scale.

OpenAI and Thailand’s Ministry of Higher Education, Science, Research and Innovation (MHESI) launched the OpenAI x MHESI AI Accelerator in Bangkok, a public–private program supporting 10 Thai start-ups in health, wellness, and education, delivered with the National Innovation Agency.

KT, Samsung SDS, and LG CNS are deploying ultra-high-speed 5G campus networks and AI/robotics testbeds at universities in South Korea to support R&D and application validation.

Tech Mahindra and ServiceNow have entered a multi-year partnership to help enterprises move from fragmented AI pilots to production-scale deployments. The collaboration combines Tech Mahindra’s industry, engineering, and transformation expertise with the ServiceNow AI Platform to develop industry-specific solutions for telecom, manufacturing, BFSI, media, and technology. Tech Mahindra will also establish an AI & Innovation Center of Excellence within its ServiceNow practice, supporting capabilities including AI Control Tower and EmployeeWorks to strengthen AI governance, productivity, and enterprise automation.

Ericsson outlined an AI-native RAN approach that embeds telco-grade AI within the radio access network to enable real-time optimization.

Apple is collaborating with Alibaba Group to train a large language model tailored for the Chinese market, shifting from prior reliance on domestic partners’ prebuilt models. Apple Intelligence is slated to roll out in China in the coming months following an iOS update.

Apple is negotiating multi-year agreements with news publishers to supply licensed, real-time content to an upgraded Siri. Instead of upfront fixed licensing fees, Apple proposes variable compensation, aiming to improve Siri’s real-time accuracy and utility.

SK Telecom will run simultaneous evaluations of network equipment from Samsung, Nokia, and HFR to compare performance and vendor options for an ongoing project.

Millennium Management is collaborating with Anthropic to co-develop an AI-powered digital risk analyst to support human risk managers and extend adoption of Anthropic’s Claude models across the firm.

Integrated Anthropic’s Claude into Rakuten AI for Business to enhance research and data analysis, including collection and organization of external information for market research, competitive assessment, and strategy planning.

Policy experts from AT&T, Lumen and industry groups report increasing network capacity needs, infrastructure permitting bottlenecks, and supply chain constraints driven by growing AI-related traffic.

LG CNS is building robotic dogs for autonomous security patrols, using a dedicated training system to accelerate development. The work is part of a joint development contract with residential facility manager Tower PMC, with initial verification underway and commercial deployment targeted for Q4.

Media snippet references 2026 telecom AI themes, focusing on agentic AI approaches for progressing toward self-driving network operations.

Calix leadership signals a strategic shift emphasizing AI across the company, characterizing the coming pace of AI-driven change as exceptionally rapid. No specific products, deployment models, or timelines were detailed.

EDOTCO Group implemented an AI-based approach to speed up a key operational process in its telecommunications infrastructure operations, using an internally developed solution to reduce cycle time.

Pegasystems outlines that CSPs scaling GenAI will see rising inference spend as tokenized usage grows with agentic loops, retrieval, and context expansion, despite falling unit token prices. The analysis quantifies typical interaction costs (~$0.016 per 2,400 tokens; ~$160k per month for 10M interactions) and notes ~40% of tokens fund retrieval rather than business logic. It highlights latency and auditability pressures and recommends decision-tier routing and pre-invocation governance to choose rules, small models, or frontier LLMs based on complexity, risk, and value.

Sponsored article outlines how private 5G can support edge analytics in manufacturing environments and notes anticipated shifts in mobile network traffic driven by generative AI. No concrete deployments, vendors, spectrum bands, or geographies are specified.

Brief industry note indicating that AI alters traditional telecom revenue models that historically correlated connectivity revenue with traffic volumes.

A sector report references aggressive AI adoption in telecom but provides no specific companies, geographies, architectures, or deployment details.

SK Telecom reported a 67% year-over-year increase in Q2 profits, attributed primarily to expansion of its AI/data center (AIDC) business, indicating growing contribution from AI infrastructure services.

GSMA's Open Telco AI initiative provides open AI models and tooling that multiple operators, including AT&T, SoftBank, and China Telecom, are using to support steps toward autonomous network operations.

OpenAI named Wiz executive Dali Rajic as CRO, with incumbent Denise Dresser exiting after under a year. The move continues recent senior leadership changes as the company scales commercial operations ahead of an anticipated IPO.

Two separate consortia are competing to develop a cybersecurity-focused AI foundation model, with one consortium to be selected in early September 2026.