Telecom Becomes an AI Vendor: AI & Automation Insights, August 2026

At a glance – AI & Automation Insights for August 2026

• Telecom operators kept building AI infrastructure directly: Ooredoo Group and Indosat Ooredoo Hutchison launched Zankore, a GPU-as-a-Service platform targeting 1GW of AI capacity in Southeast Asia, while SK Telecom‘s Q2 profit rose 67% on AIDC business growth.
• AI-RAN validation reached the device and per-user level: NTT Docomo and Samsung pushed AI-RAN to the smartphone layer and validated per-user optimization in Japan, while LG Uplus and Yonsei University confirmed handoff continuity for moving robots and drones.
• Agentic AI moved deeper into live network operations: Vodafone and Google Cloud outlined 3GPP-aligned agentic network automation, and NVIDIA‘s 2026 telco survey found autonomous networks are operators’ best-performing AI ROI use case.
• Customer-facing AI applications multiplied: KDDI began piloting a Gemini-powered humanoid robot on the Osaka Metro, and Zain launched real-time Arabic call translation in Kuwait.
• Governance and cost economics caught up with capability: Anthropic began watermarking AI-generated text under the EU AI Act, AT&T cut AI coding costs 56% via model routing, and new analysis shows falling token prices are masking rising total inference spend.
• Ecosystem partnerships and capital kept consolidating: Tech Mahindra partnered with ServiceNow on enterprise AI deployment, and a new report pegs AI-driven space tech as a $100–200 billion global investment opportunity.

1. Telecom becomes an AI infrastructure and compute provider

Operators kept moving from AI customer to AI infrastructure vendor this month, monetizing compute and platform access directly rather than only procuring AI as a managed service.

• Ooredoo Group & Indosat Ooredoo Hutchison: each launched a branded Zankore GPU-as-a-Service platform in Southeast Asia with Nvidia and Nokia as technical partners, targeting 1GW of AI capacity in Indonesia over three years — an initial 200MW slated for H1 2027, with Ooredoo holding 49% and committing $800M over five years. Indosat and Nvidia separately opened an AI research hub in Indonesia to co-develop capabilities for the local ecosystem. See TeckNexus’s “When Your Connectivity Provider Becomes Your Compute Provider”.
• SK Telecom: reported Q2 profit up 67% year-over-year, attributed primarily to growth in its AI/data centre (AIDC) business, and was separately selected to operate South Korea’s national “AI for Everyone” program.
• Yandex: launched Yandex SIM, a mobile virtual network service embedding an AI assistant into calling to analyze conversations, flag potential fraud, and personalize plans based on usage.

Why it matters for buyers: SK Telecom’s AIDC-driven profit jump and two operators launching branded GPU-as-a-Service platforms in the same month are early evidence that AI infrastructure can be a genuine revenue line for telecom, not just a cost centre — worth tracking if your own vendor relationships could shift from buying AI to buying compute from a carrier. → Compare build, buy, and partner options with the TeckNexus AI Use Case Prioritiser.

2. AI-RAN and network autonomy keep expanding

AI-RAN validation pushed further down the stack this month — to the device and per-user level — while industry bodies kept building the standards layer underneath it.

• LG Uplus & Yonsei University: connected LG Uplus’s commercial 5G network to a Yonsei University AI-RAN testbed powered by Nvidia GPUs and validated that mobility handoffs can be maintained for moving physical AI and robotic devices as they cross between the two networks with automatic access selection.
• NTT Docomo & Samsung: are extending AI-RAN to the device layer, placing on-device intelligence in smartphones to assist RAN optimization, and separately validated a Japan trial that uses AI to analyze individual user movement and behaviour to predict service degradation and adjust RAN configuration per user — architectures aimed at 6G-era predictive network objectives.
• SK Telecom: is running a parallel evaluation of Samsung, Nokia, and HFR equipment to compare performance and vendor options for an ongoing AI-RAN project.
• Ericsson: detailed an AI-native RAN strategy that embeds telco-grade AI directly within the radio access network to enable real-time optimization.
• GSMA: reported that its Open Telco AI models are being adopted by AT&T, SoftBank, and China Telecom to advance steps toward autonomous network operations. See GSMA’s official newsroom announcement.
• NGMN Alliance: published a roadmap treating standards, security, architecture, operations, and governance as prerequisites for genuine Level 4 network autonomy rather than items to backfill after automation is deployed — see TeckNexus’s “Level 4 Autonomous Networks Are Coming: NGMN’s Roadmap Explained.”

Why it matters for buyers: AI-RAN moving to the device and per-user level, alongside a formal Level 4 autonomy roadmap, signals the field is maturing from network-level pilots toward a defined architecture stack — buyers should start mapping their own RAN modernization plans against that stack now rather than treating AI-RAN as a single monolithic upgrade. → Map AI-RAN and autonomy architecture options against your deployment horizon with TeckNexus Network Planning tools.

3. Agentic AI moves deeper into live network operations

Agentic AI kept crossing from demonstration into operational deployment this month, backed by fresh survey evidence on where the ROI is actually landing.

• Deutsche Telekom & Vodafone: championed a joint TM Forum Catalyst demo using agentic AI to automate translation of dense technical documentation, pointing at broader applicability for telecom process and knowledge workflows.
• Vodafone & Google Cloud: outlined cloud-executed agentic AI to automate operations in autonomous mobile networks, built around 3GPP-aligned approaches.
• NVIDIA: its 2026 State of AI in Telecommunications survey (n=1,038) found 90% of operators report AI is already driving revenue and reducing costs, and 89% plan to increase AI spend in 2026, up from 65% a year earlier; a parallel cross-industry survey of 3,200+ enterprises ranked telecom highest for agentic AI adoption at 48%. Autonomous networks were cited by roughly half of respondents as the best-performing AI ROI use case, ahead of customer service (41%) and internal process optimization (33%). See NVIDIA’s full 2026 State of AI in Telecommunications report.
• TM Forum: introduced three core projects under its AI-Native Blueprint at MWC Barcelona 2026: Model as a Service (MODaaS) for enterprise-grade model sourcing and governance, Data Products Lifecycle Management (DPLM) for agent-accessible data standards, and Agentic Interactions Security defining a policy language to secure agentic AI at scale — MODaaS is already referenced by the World Economic Forum. See TM Forum’s official announcement.
• Avenga: argued at TM Forum DTW Ignite 2026 for shifting telecom operations to outcome-based AI with clear decision rights, incremental deployment, and strong governance as prerequisites for enterprise-scale adoption.

Why it matters for buyers: NVIDIA’s survey giving autonomous networks the strongest self-reported AI ROI of any use case — ahead of customer service — is a useful benchmark when your own organization is deciding where to point agentic AI first. → Evaluate agent vendors against evidence, not architecture claims, with the TeckNexus RFP Scorecard Generator.

4. Consumer and customer-facing AI applications multiply

AI kept showing up directly in front of end customers this month, from physical robots to real-time call translation.

• KDDI: in collaboration with Osaka city officials and robotics vendor Avita, began a public trial (starting August 31) of a humanoid robot assisting passengers with information on the Osaka Metro, powered by Google’s Gemini AI model.
• HSAD: LG’s creative agency applied AI to behavioural data to identify higher-propensity prospects; campaigns using AI-targeted ads achieved 2.1x the conversion rate of general commercials, with average purchase value up 18%.
• Zain: introduced an AI Calling Translation Pioneer Program providing real-time Arabic translation during voice calls — the first such service from a Kuwaiti operator.
• Mavenir & Sanas: integrated Sanas’s real-time voice processing engine into Mavenir‘s MAVcore Voice AI portfolio, letting operators process, enhance, and secure voice calls within the network rather than on-device — a design aimed at sovereign AI and data-residency requirements.
• Vodafone Business: acted as delivery partner to the UK Home Office and National Police Chiefs’ Council on a co-developed AI service.

Why it matters for buyers: a humanoid robot, AI-targeted advertising, and real-time call translation landing in the same month shows customer-facing AI applications are no longer confined to chatbots — evaluators comparing vendors in this space need criteria that span physical, voice, and data-residency requirements at once. → Compare vendor categories and capabilities with the TeckNexus AI Use Case Prioritiser.

5. Governance, security, and cost economics catch up to capability

Governance and cost scrutiny kept pace with AI capability this month, from a regulatory transparency mandate to a hard look at what AI inference is actually costing operators.

• Anthropic: confirmed that all AI models launched on or after August 2 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.
• Two consortia: are competing to build a cybersecurity-focused AI foundation model, with a single award expected in early September.
• AT&T: cut AI coding costs 56% by implementing LiteLLM-based model routing in its internal “Ask AT&T” platform, directing routine queries to open-source models (Nvidia Nemotron, Meta Llama, Google Gemma) while reserving complex tasks for proprietary providers; open-source models now handle roughly 40% of requests (targeting 60–70%), processing about 45B tokens per day.
• Falling token prices are masking rising total inference spend: an analysis from Pegasystems, published via TM Forum, found that retrieval overhead (roughly 40% of tokens), agentic loops, and expanding context windows are driving total AI spend up for telecom operators even as per-token prices fall — one estimate puts a 10-million-interaction monthly workload at roughly $160,000. See TeckNexus’s “Why Telecom AI Costs Keep Rising Even as Token Prices Fall,” which recommends decision-tier routing and pre-invocation governance as the practical fix.
• A coalition of 120+ organisations: including Nvidia, Cisco, and CrowdStrike, proposed a standardised framework for documenting and sharing AI agent security and operational incidents, modelled on existing incident-reporting standards in cybersecurity and aviation — see TeckNexus’s “A Coalition of 120 Organisations Wants a Standard Way to Report AI Agent Incidents”.
• Honeywell: outlined a bounded-autonomy framework for industrial AI agents — deterministic, repeatable tasks (sensor fault handling, shift handovers, routine startup/shutdown) get agent autonomy, while ambiguous or safety-critical decisions keep human oversight — see TeckNexus’s “Where Industrial AI Agents Should Be Allowed to Act Alone”.
• Carrier-delivered trust and security services: Glide.id’s MagicalAuth launched in public beta with AT&T, T-Mobile US, and Verizon, replacing SMS one-time passcodes with hardware-rooted SIM/eSIM authentication verified through open network APIs; separately, Lockheed Martin‘s upcoming 5G drone-detection service on Verizon‘s network and AT&T’s cloud-managed Video Intelligence platform both point at physical security AI shifting from owned systems to carrier-delivered managed services. See TeckNexus’s “Carriers Are Quietly Replacing SMS Passcodes With the SIM Itself” and “Carriers Are Becoming the Vendor for Site Security AI, Not Just Connectivity”.

Why it matters for buyers: the same month a frontier AI lab began complying with EU transparency law, an operator proved a 56% cost cut is achievable through routing discipline, and independent analysis showed why most operators’ inference bills are still rising anyway — governance, cost control, and vendor evaluation are converging into a single procurement conversation rather than three separate ones. → Score vendor governance, security, and cost-discipline evidence with the TeckNexus RFP Scorecard Generator and model total inference spend with ROI/TCO tools.

6. Ecosystem partnerships and capital positioning

The broader AI ecosystem kept forming new partnerships and positioning around telecom this month, backed by fresh capital and survey evidence of where operators expect revenue to land.

• Tech Mahindra & ServiceNow: entered a multi-year partnership to scale enterprise AI deployments on the ServiceNow AI Platform, building go-to-market playbooks across manufacturing, telecom, BFSI, media, and technology; Tech Mahindra will establish an AI & Innovation Center of Excellence within its ServiceNow practice.
• KT: is building AI and 5G campus testbeds with Samsung SDS and LG CNS at South Korean universities to support R&D and application validation.
• Apple & Alibaba: are training a large language model tailored for the Chinese market, shifting from Apple‘s prior reliance on domestic partners’ prebuilt models, ahead of an Apple Intelligence rollout in China.
• Rakuten Mobile: integrated Anthropic’s Claude into its corporate generative AI service to support research and data analysis, including market research and competitive assessment.
• LG CNS: is developing autonomous patrol robotic dogs with residential facility manager Tower PMC, targeting Q4 commercial deployment.
• Ciena: a global survey found 90% of operators expect high-capacity AI networking services to drive revenue over 3–5 years, with 56% expecting it to be their main source of net-new revenue; a parallel India-focused survey found 67% of Indian respondents strongly agree AI-driven network services will be a primary revenue driver, 98% expect managed optical fiber network revenue from interconnecting distributed AI compute, and 88% say optical network upgrades are urgently needed to meet low-latency enterprise demand.
• Space tech investment: a new report pegs AI-driven space technology as a $100–200 billion global investment opportunity.
• EDOTCO Group: applied an internally developed AI-based approach to accelerate a key operational workflow in its telecom infrastructure operations, reducing cycle time.

Why it matters for buyers: Ciena‘s survey giving operators strong conviction that AI networking will be their primary net-new revenue source, alongside Tech Mahindra’s and KT‘s fresh AI partnerships, suggests the ecosystem is positioning for AI-driven revenue now rather than waiting for it to materialize — worth factoring into your own vendor roadmap conversations this quarter. → Compare partner and vendor track records with the TeckNexus Technology Selector.

What AI & Automation Insights for August 2026 means if you’re evaluating AI and automation investments

August’s throughline is telecom operators stepping into the AI supply side directly — as compute providers, model builders, and managed-security vendors — while governance and cost discipline finally caught up with a year of rapid agentic-AI deployment. Six moves follow directly from the month:

• Watch whether your telecom or cloud partners are becoming AI infrastructure providers themselves — Zankore, SK Telecom’s AIDC business, and Yandex SIM all show carriers monetizing AI directly, which changes the vendor relationship on offer.
• Benchmark AI-RAN engagement against NVIDIA’s ROI survey, not marketing volume — autonomous networks are the best-performing self-reported use case, ahead of customer service and internal process optimization.
• Ask agentic AI vendors for evidence against TM Forum’s AI-Native Blueprint categories (model sourcing, data lifecycle, interaction security) rather than architecture claims alone.
• Build AI cost models that account for retrieval overhead and agentic loops, not just per-token pricing — this month’s analysis shows falling token prices don’t guarantee falling total spend.
• Treat AI agent incident reporting and bounded-autonomy frameworks as procurement criteria now, not later — the 120-organisation coalition and Honeywell’s framework both give buyers a concrete standard to ask vendors about.
• Factor carrier-delivered trust and security services (SIM-based authentication, managed site-security AI) into your vendor shortlist if you’re evaluating identity or physical security investments this quarter.

For this month’s private network deployment evidence, see our companion Private Network Insights, August 2026 →

Catching up? Last month’s signal is here: The Trust Gap Widens as AI-RAN Goes Government-Scale: AI & Automation Insights, July 2026 →

→ Start with the TeckNexus Intelligence Platform — independent, buyer-neutral tools for AI, ROI, network planning, and RFP decisions, useful to enterprises, vendors, and the broader ecosystem alike.

This analysis is drawn from TeckNexus’s full curated AI & Automation Monthly Insights for August 2026. See every deployment, product, and partnership update →

The TeckNexus Intelligence Platform is buyer-neutral and TeckNexus-authored. Vendor co-branded Intelligence Packs are labelled as such and kept separate from the neutral core tools.

RELATED TOOL

Is your AI agent vendor’s ROI claim backed by evidence — and can you see the real cost?

With NVIDIA’s 2026 survey showing autonomous networks as operators’ best-performing AI ROI use case, and independent analysis showing total inference spend can rise even as token prices fall, the decisive questions are evidence and total cost — not architecture claims. TeckNexus’s AI Use Case Prioritiser ranks candidate AI applications by impact, feasibility, and payback, while the RFP Scorecard Generator structures vendor evaluation around governance, security, and incident-reporting readiness, and ROI/TCO tools model total inference spend against your own workload.

Explore the AI Use Case Prioritiser →