AI is creating enormous new businesses, but telecom operators are barely participating. While the industry concentrates on efficiency, networks, and GPU infrastructure, other companies are moving into markets that telcos should be fighting for.

Machine learning is increasingly transforming optical access networks, which form the critical infrastructure connecting service providers to end users. This article examines how artificial intelligence and machine learning algorithms are being applied to optimize network performance, enhance resource allocation, and improve operational efficiency in optical communications systems. Key applications include predictive maintenance, signal quality optimization, and automated network management—technologies that enable real-time monitoring and adaptive responses to network conditions. These techniques leverage data analytics to identify patterns in network behavior, predict potential failures before they occur, and automatically adjust optical signal parameters to maintain service quality. The significance of this advancement lies in addressing growing bandwidth demands while reducing operational costs and minimizing service disruptions. By implementing machine learning in optical access networks, telecommunications providers can achieve superior network reliability, faster problem resolution, and more efficient spectrum utilization. This integration represents a major shift toward intelligent, self-optimizing network infrastructure that adapts dynamically to changing demands, ultimately delivering improved customer experiences and competitive advantages in increasingly data-intensive telecommunications environments.

The European Space Agency, European Commission, and over 20 European space companies have launched a collaborative initiative to develop Eagle-1, Europe's first space-based quantum key distribution satellite system. This partnership, formalized through a contract between ESA and Luxembourg-based SES, aims to establish an unbreakable, physics-based secure communications network for the European Union. Eagle-1 will demonstrate quantum key distribution technology, which leverages quantum mechanics to distribute encryption keys while immediately detecting any eavesdropping attempts. The satellite will build upon technologies developed under ESA's Scylight programme, validating critical EU-sourced components. The system will operate from low Earth orbit to ground stations, including upgraded optical terminals from Germany's DLR and a new terminal from the Netherlands, with Italy's Sitael providing the platform and Germany's Tesat Spacecom supplying the quantum payload. Scheduled for launch in 2024, Eagle-1 will undergo three years of in-orbit validation, ultimately supporting the European Quantum Communication Infrastructure and enabling the EU to achieve sovereign, autonomous cross-border quantum-secure communications that significantly enhance European cybersecurity autonomy.

5G non-terrestrial networks (5G NTN or NR-NTN) represent a fundamental evolution in mobile connectivity, leveraging satellites across low Earth (LEO), medium Earth (MEO), and geostationary (GEO) orbits, alongside high-altitude platforms (HAPS), to extend coverage to oceans, remote regions, and disaster zones. Standardized beginning with 3GPP Release 17, with subsequent enhancements in Releases 18 and 19, 5G NTN has transitioned from research and prototypes into practical deployments and early service rollouts. The technology operates across FR1 (410 MHz to 7.125 GHz) and FR2 (24.25 GHz to 71 GHz) spectrum bands, with recent expansion into Ka and Ku bands and future targeting of Upper C-band, Q/V, E, and D bands. The architecture comprises four segments—ground, air, space, and network—supporting diverse deployment models. Two satellite approaches exist: simple transparent relays enabling faster deployment but with higher round-trip times, and regenerative payloads offering onboard processing and reduced latency at increased complexity. Managing interference across wide coverage areas requires specific metrics including power flux density (PFD) and equivalent power flux density (EPFD). Physical layer adaptations address Doppler effects, propagation delays, and pathloss through advanced antenna systems and beamforming, with comprehensive protocol stack modifications required across all network layers.

A major industry coalition led by NVIDIA, Nokia, T-Mobile US, Deutsche Telekom, Orange, and SoftBank is accelerating the deployment of AI-RAN (artificial intelligence radio access network) infrastructure, transforming traditional cell sites into converged platforms capable of simultaneously handling real-time radio scheduling and external AI inference workloads. The initiative moves beyond marketing rhetoric into concrete implementation, with nearly 20 global telecom operators publicly adopting Nokia's anyRAN software alongside NVIDIA's ARC architecture. NVIDIA is launching two ARC server configurations—the MGX 2U and MGX 4U—powered by RTX PRO 6000 Blackwell Server Edition GPUs with x86 and NVIDIA Vera CPU options, to be offered by Dell Technologies, Quanta Cloud Technology, and Supermicro. This edge-centric approach aims to monetize distributed carrier infrastructure by hosting AI processing directly within decentralized network footprints. However, the rapid commercial deployment exposes a critical gap: the 3GPP SA2 Working Group remains technically fractured on formalizing underlying 6G core architecture standards, potentially repeating the interoperability fragmentation that undermined 5G Standalone adoption. Bridging vendor innovation velocity with standardization frameworks has become essential to preventing architectural splintering.

Nokia and SoftBank have announced a commercial AI-RAN field trial launching in Japan by 2027, marking a significant milestone in next-generation network technology development. The trial will integrate Nokia's AI-native anyRAN Cloud RAN software with SoftBank's AITRAS platform, supplemented by Nvidia's ARC accelerated computing platform and AITRAS Orchestrator. The infrastructure incorporates Nvidia Grace CPU, RTX PRO 4500 GPU, and ConnectX-8 networking to deliver the computing power and low-latency connectivity required for cloud-native RAN workloads. This collaboration will validate AI-RAN technology under real commercial network conditions and actual traffic loads, advancing both companies' commercialization strategies. The trial demonstrates how AI and RAN workloads can operate on unified, programmable infrastructure, offering deployment flexibility beyond traditional implementations. For Nokia, this represents a crucial transition from laboratory development to real-world deployment, while SoftBank progresses toward commercializing AI-RAN services. The partnership highlights the industry's movement toward AI-native 6G networks, potentially unlocking improved spectral efficiency and distributed intelligence capabilities.

Sur près de 94 pages d’avertissements, Anthropic reconnaît que ses IA ont déjà manifesté, lors de tests simulés, des comportements proches du chantage pour éviter leur désactivation, qu’elles peuvent modifier leur comportement lorsqu’elles détectent qu’elles sont surveillées ou évaluées, qu’une erreur de configuration a permis à quatre modèles autonomes, au cours de tests de cybersécurité, d’accéder sans autorisation à de véritables systèmes informatiques extérieurs pendant des sessions de 10 à 34 heures — l’un d’eux parvenant même à obtenir des droits d’administrateur sur un serveur, sans qu’Anthropic ait trouvé de preuve d’une volonté d’échapper à la surveillance —, et qu’en cas de catastrophe majeure liée à l’IA, les dommages pourraient dépasser les capacités de couverture d’un assureur, voire de l’ensemble du secteur de l’assurance. Anthropic joue-t-elle avec nos peurs pour rester au centre du jeu ? Plongée en 10 points dans un des documents les plus fascinants et déstabilisants de l’histoire de la Silicon Valley.

It is crazy what you hear about Elon Musk in telecom. Some people think he can defy physics, eliminate the need for cell towers, and build magical networks that will destroy every mobile operator on Earth. Others dismiss everything he does simply because they cannot stand him. Some commentators have financial interests in SpaceX succeeding. Others seem personally invested in seeing Musk fail. Personally, I couldn’t care less about him. I love engineering, I understand the telecom business, and I hate bullshit. So here are my 20 truths and lies about Musk in telecom.

The pitch from vendor keynotes is sexy. Mobile operators already own millions of cellular towers covering the planet. By upgrading those towers to 6G, every base station will supposedly double as a high-resolution radar. Operators will stop just selling cheap data plans; they will sell a live “digital twin” of the physical world, charging cities to monitor street intersections, aviation authorities to track low-altitude drones, and defense ministries to guard the skies. Let’s call this narrative what it is: commercial bullcrap.

Anthropic wants to replace millions of white-collar workers and justify a $2 trillion valuation. But getting enterprises to actually use its AI is proving much harder than building it.

Derrière cette affaire se cache une réalité peu connue : le secteur des téléviseurs intelligents fonctionne désormais en partie comme celui des réseaux sociaux, avec un modèle économique fondé sur la collecte et la revente de données personnelles à des fins publicitaires. Le hacker affirme avoir contourné le solide chiffrement AES du micrologiciel de son téléviseur en exploitant des vulnérabilités de type zero-day, c’est-à-dire non encore connues et corrigées, afin d’obtenir un accès administrateur complet. Grâce à cela, il a pu analyser le trafic et les processus internes, ce qui a rapidement révélé un tableau qu’il a qualifié d’« horreur pour la vie privée ». Concrètement, l’appareil intègre des mécanismes d’analyse des paquets réseau, c’est-à-dire du trafic du réseau wi-fi, ainsi qu’un identifiant publicitaire interappareils permettant de recouper les usages sur plusieurs écrans d’un même foyer.

We are probably in the Windows 1.0 era of humanoids. The hardware is improving, but the software remains limited, and robots still fail at tasks a five-year-old could do without thinking. But things are moving remarkably fast. Unitree already sells its R1 humanoid starting at $4,900, while 1X is offering its NEO home robot for $20,000 or a $499 monthly subscription. These are very different products, and the cheaper machines are far from being useful household workers, but prices are moving in an interesting direction. According to Counterpoint Research, more than 22,000 humanoids shipped in the first half of 2026, nearly 300% growth from the previous year. The firm expects shipments to exceed 50,000 this year. Most of these machines still go into research, demonstrations, and data collection, but industrial deployments are starting to happen. Figure has tested robots at BMW, and Agility Robotics is already moving goods in GXO warehouses.

Analysts warns of a massive neocloud shakeout. Banks are getting nervous, GPU financing is becoming expensive, and trillions are being committed to AI infrastructure. Is an AI bubble burst near?

Walk into any telecom boardroom right now, mention Elon Musk, and watch twenty executives reach for their blood pressure medication. Between SpaceX’s Q3 numbers showing Starlink at 13.6 million subscribers, direct-to-device constellations maturing, and Musk’s move on 800 MHz spectrum to build “Starlink Mobile,” operators are staring at the sky in a cold sweat. They are looking in the wrong direction. Again. Sure, Musk is coming for your rural coverage and a slice of your mobility margin. But Mark Zuckerberg is coming for the only slide in your investor deck that still shows growth. For the past three years, every telco CEO on Earth has told analysts the same story: connectivity is flat, ARPU is dead, so future growth will come from B2B digital services. Operators will sell cloud, SaaS, automation, and AI agents to small- and medium-sized businesses and the mid-market. Meta Muse is about to wipe that revenue off the map. Completely.

If I were running a TowerCo, I would not approach SpaceX with another presentation explaining how many towers I own. I would offer to build and operate its entire terrestrial physical network. My proposal would be simple: “Elon, you build the radios, satellites, software, and rockets. Give us the technical specifications and RF plan. We will deliver 80,000 operational sites in 18 months, with a contractual fallback to 24 months. Then, if you want, we hand you the keys.” That is a Build-Operate-Transfer (BOT) business, not another tower lease. And the commercial opportunity could be considerably larger.

Voice over Narrowband Internet of Things Non-Terrestrial Networks (NB-IoT NTN) represents a significant evolution in 3GPP Release 20 standardization efforts. This work extends NB-IoT technology, originally designed in Release 13 for simplified IoT applications like smart metering and asset tracking with data rates of only 10-100 kbps, to now support voice communications over satellite infrastructure. Building upon the satellite adaptation introduced in Release 17, Release 20 expands voice capabilities beyond smartphones to encompass wearables and automotive applications, including satellite-based emergency voice services for vehicles. The enhancement addresses emerging requirements for voice communications in non-terrestrial environments while maintaining NB-IoT's core advantages: extended coverage through transmission repetition, efficient spectrum utilization via in-band deployment with 4G, and simplified device architecture. The development leverages existing optimization approaches including Cellular IoT Control Plane Optimization (CIoT CP) and Cellular IoT User Plane Optimization (CIoT UP). This standardization work matters because it enables critical voice connectivity in remote and underserved areas via satellite networks, extending essential communications capabilities to diverse device categories and emergency services.

The emerging AI-RAN (Artificial Intelligence Radio Access Network) technology represents a strategic convergence of telecommunications, artificial intelligence, and computing infrastructure with profound geopolitical implications through 2031. On October 21, 2026, NVIDIA, Nokia, and T-Mobile completed a landmark AI-RAN call demonstration on T-Mobile's operational 5G network in Bellevue, Washington, utilizing Nokia Massive MIMO radios, CUDA-accelerated anyRAN software, and NVIDIA Aerial RAN Computer infrastructure. This achievement marks a significant commercial-network integration milestone, though production-scale economic viability and autonomous 6G functionality remain unproven. Beyond accelerating mobile networks, AI-RAN could transform telecommunications infrastructure into geographically distributed computing systems capable of supporting connectivity, real-time sensing, and automated decision-making. The industrial consequences are substantial, as telecom operators' distributed sites, fiber connections, and infrastructure could evolve into edge-computing platforms. Geopolitically, competitive boundaries between semiconductor manufacturers, telecom equipment vendors, cloud providers, and network operators are converging, concentrating strategic dependence in GPU ecosystems, AI toolchains, and accelerator supply chains. The decisive phase is expected during 2027–2029, when field trials, performance metrics, and early 6G standardization will determine viable architectures and inform government policy on autonomous infrastructure deployment.

# Summary This article discusses integrated positioning and communication systems utilizing low Earth orbit (LEO) satellites, representing a convergence of two traditionally separate technological domains. LEO satellite constellations, operating at altitudes between 160 and 2,000 kilometers, offer unique advantages for simultaneously delivering precise positioning and reliable communication services globally. The integration of these capabilities addresses the growing demand for seamless connectivity in remote areas and emerging Internet of Things applications while reducing infrastructure costs through consolidated satellite networks. By combining positioning, navigation, and timing functions with communication services, integrated LEO systems enable more efficient spectrum utilization and simplified user equipment requirements. This dual-functionality approach has significant implications for telecommunications, emergency response, autonomous vehicle operations, and maritime applications. The work appears in npj Wireless Technology, which offers article processing charge waivers available on an ad-hoc basis following acceptance. This development reflects the industry's strategic shift toward multi-purpose satellite platforms that can serve increasingly diverse global connectivity needs.

1oT has launched an eSIM connectivity service designed to address a critical pain point in the Internet of Things industry: vendor lock-in. The platform enables IoT companies to switch between telecommunications providers seamlessly, prioritizing operational speed and flexibility. The service offers 12 distinct telecoms profiles, allowing organizations to select connectivity solutions tailored to their specific use cases, geographic regions, and technological requirements. Currently, 1oT's infrastructure supports three million IoT devices across 173 countries, ranging from wildlife tracking systems like bird trackers to micro-mobility solutions such as e-scooters. This innovation matters significantly for the IoT sector because it reduces dependency on single providers, enhances cost competitiveness, and enables organizations to optimize their connectivity strategies dynamically. By eliminating contractual constraints, 1oT empowers enterprises to make agile connectivity decisions that align with evolving business needs and market conditions.

This paper addresses the challenge of dynamically orchestrating resources in 5G edge network slicing by proposing a novel two-timescale architecture that combines Large Language Models with multi-agent auction mechanisms. The system uses an LLM-based Strategist component to translate natural-language network operator intents into economic policy parameters such as budget and bidding aggressiveness at the macro scale. Concurrently, SPADE-based slice agents execute a deterministic Vickrey-inspired second-price auction mechanism to resolve real-time resource contention at the operational scale, with allocations enforced directly on Kubernetes CPU limits and Calico CNI bandwidth annotations. The researchers evaluated their approach on a real Kubernetes/Open5GS edge cluster across two scenarios repeated thirty times: a Smart-City scenario demonstrating rapid propagation of public-safety intents into measurable bid and bandwidth allocation changes, and a Hierarchical-slicing scenario showing priority-dependent differentiation, adaptation to demand fluctuations, and preservation of minimum guarantees. This approach matters because it enables mission-critical 5G edge applications requiring strict latency bounds and reliable bandwidth access to maintain quality of service through intelligent, responsive resource orchestration without the computational overhead and non-determinism of continuous LLM inference.