Telco CEOs have changed as much as the networks they run. Two decades ago, most were engineers obsessed with coverage maps; today, they’re diplomats managing politics, investors, and cloud partners.

In 2008, Europe’s telcos were the most valuable companies in technology. Today, their worth has fallen by more than 80%. We spoke to insiders to understand how the empire collapsed.

In classical AI, an agent is nothing mystical. It is a system that perceives, makes decisions, and acts within an environment. A thermostat is an agent. A Roomba is an agent. The term never implied consciousness or autonomy. It was a clean abstraction for building systems that respond to inputs and produce outputs in pursuit of a goal.In the era of generative AI, the word “agent” has been hijacked. Vendors present agents as if they were digital employees who can sit in your enterprise, learn the ropes, and work unsupervised.

Instead of every device depending on a single WiFi router, ESP32 boards link together, passing data from node to node. The result? A self-healing, decentralized network that keeps devices connected, even if one path fails. 

Derrière cette affaire, la découverte est impressionnante : plus de 300 serveurs SIM, abritant pas moins de 100 000 cartes SIM, étaient dissimulés sur différents sites stratégiques. Ces équipements n’étaient pas cantonnés à des appels anonymes : ils pouvaient tout autant permettre des attaques complexes sur l’infrastructure mobile, comme le sabotage d’antennes-relais ou l’exécution d’attaques par déni de service. À travers une communication chiffrée et opaque, ces dispositifs offraient aussi un canal privilégié entre des individus déjà surveillés par la police fédérale et des acteurs étatiques étrangers.

Les terminaux d’aujourd’hui ne ressemblent plus guère à ceux d’hier. Désormais, la transformation numérique s’impose comme le socle de l’aviation moderne, révolutionnant bien au-delà des simples systèmes de réservation. Grâce à une alliance de la 5G, du Wi-Fi 6, des réseaux privés et de l’edge computing, les aéroports se métamorphosent en véritables plateformes connectées. Cette mutation n’a rien d’anodin : elle vise autant l’efficacité opérationnelle que l’expérience vécue par chaque voyageur.

Telecommunications is entering a new investment cycle under very different financial conditions than those of the last cycle. The era of near-zero rates that supported the rollouts of 4G and early 5G has given way to higher capital costs, stricter credit markets, and slower revenue growth. These shifts expose structural weaknesses that are often overlooked in public reporting.This week’s three confessions examine them directly: the composition of telco boards and how it limits innovation, the role of debt as a permanent feature of the industry’s balance sheets, and the widening gap between returns and the cost of capital that will determine how 6G is financed.

By 2025, GPUs will be the most contested resource in the technology sector. Demand from AI training and inference has pushed supply chains to their limit, and a single rack of high-end accelerators can generate more revenue per square meter than most telco switches ever did.Telecom operators have begun to take an interest in this activity. With mobile data revenues flat and infrastructure costs rising, the appeal of a service category that analysts expect to expand from approximately $5 billion in 2023 to more than $50 billion by the early 2030s is obvious. When combined with the parallel rise of edge computing, projected to exceed 300 billion by 2030, the potential looks transformative.

Seven outsiders step into the scanner to expose what telcos are getting wrong, and how to heal the fractures before it’s too late

La région de l’Afrique subsaharienne connaît une transformation numérique sans précédent, où le déploiement progressif de la 5G agit comme un véritable catalyseur. Si l’on observe d’ores et déjà des avancées notables en Afrique du Sud, au Kenya ou au Nigeria, d’autres pays comme le Ghana, la Tanzanie ou l’Ethiopie commencent également à tirer parti de ce nouvel élan technologique. L’accès plus rapide à Internet et la diminution de la latence modifient profondément les usages, ouvrant le champ à des expériences inédites, du streaming haute définition aux jeux mobiles en temps réel.

L’opérateur télécoms Orange se dit désormais prêt à participer activement à une consolidation du secteur en France, marquant un tournant potentiel dans le paysage concurrentiel des télécommunications et ouvrant la voie à d’éventuels rapprochements entre acteurs majeurs.

Telecom networks run on models. Erlang distributions, busy-hour factors, peak-to-average ratios, and oversubscription levels are the tools that let operators design networks without provisioning for absolute maximum demand. The assumption is that human behavior follows statistically smooth curves, allowing for the forecast of aggregate demand and the absorption of rare spikes with a margin of overcapacity.AI breaks this foundation. Agent-driven traffic is non-ergodic: past averages do not predict future bursts. Each inference request is unique, decorrelated from the next, and capable of spawning chains of sub-requests across geographies.

The numbers don’t lie. The global AI value chain is expected to expand from $280 billion in 2024 to over $1.6 trillion by 2030. Chips, power, data centers, platforms, models, applications, it’s the most significant technology shift in three decades. And telcos? Even in the best-case scenario, we capture only 1–2% of that pool, approximately $30 to $45 billion per year. At the same time, we’ll spend more than that, roughly $50 billion annually, buying GPUs, cloud services, applications, and AI software. We are not invited to the party. We are paying for it.

#DOTDOT, LA STARTUP FRANÇAISE QUI TRANSFORME LES GIGAS INUTILISÉS EN ACCÈS INTERNET LOCAL ET À LA DEMANDE, EST FINALISTE DU GLOBAL STARTUP AWARDS – SOUTH EUROPE 2025 DANS LA CATÉGORIE STARTUP OF THE YEAR.

Yet a new source of demand is emerging — not from humans but from machines. AI agents are beginning to transact thousands of times per second, executing trades, scheduling deliveries, authenticating identities, and negotiating contracts with no human in the loop. Connected vehicles already generate terabytes of sensor data daily, and while most of it will be processed locally, even a fraction flowing through networks represents a seismic shift. Industrial robots on factory floors, drones in the airspace, and smart energy grids will all rely on constant streams of verified, low-latency interactions.This machine-driven internet is not about how many gigabytes are consumed. A drone monitoring an oil pipeline doesn’t care if it sends twenty megabytes or two hundred; what matters is that the signal arrives in five milliseconds, not fifty, and that it can prove the device on the other end is genuine. A trading algorithm won’t choose a network based on marketing, but on whether its transaction is cleared deterministically with zero packet loss. An AI-powered medical device doesn’t care about “unlimited data,” but whether the network can guarantee integrity, trust, and compliance in real time.

We spend less than 1% of our revenue on R&D. In 2022, AT&T, with more than $120 billion in revenue, spent just over $1.2 billion on research and development. Vodafone and Telefónica are in the same range. Compare that with Alphabet, which allocates approximately 15% of its $280 billion in revenue to R&D, or Meta, at nearly 20%. Even Apple, often described as conservative, spends more than 7%.Our spending goes elsewhere. CAPEX intensity for the sector runs at 15% to 18%. For Vodafone, €7.8 billion on €45 billion of revenue; for AT&T, roughly $24 billion a year. Spectrum auctions, towers, and fibre consume everything. Then come dividends. European operators yield between 5% and 8%. AT&T yields around 6. Debt service takes another 10% of revenue. The sector carries more than $1.7 trillion in debt, with leverage at three times EBITDA in many cases. After those obligations, there is nothing left for invention.

À la sortie des caisses, Élodie, mariée et mère de famille, explique que, désormais, avec les craintes liées aux taxes, aux impôts et à l’emploi, elle fera plus attention à ses dépenses : "C'est la rentrée en sixième, un gros cap avec le passage au collège, donc forcément des dépenses plus importantes et une attention vraiment portée sur les prix."

The environmental impact of streaming extends far beyond servers and networks: it reaches our mobile devices and living rooms. At Greening of Streaming, we’ve been investigating where and how energy is consumed across the streaming ecosystem. These four recommendations emerged from our research and testing. We don’t claim they are the most important actions, but they are practical, user-facing steps that operators and service providers could promote or support today.

The combination of ARPU obsession and gross adds worship locks telcos in a bundle-and-churn model. It encourages volume chasing, discounting, and endless promotions, instead of building long-term value. It makes operators look busy, more gross adds, and higher ARPU, while masking structural weakness.This is why telcos, even after pouring over $3 trillion in CAPEX over the last decade, haven’t seen revenue growth keep pace with data traffic. The KPIs they still cling to are built for a world of selling minutes and megabytes, not for today’s digital ecosystems.

The problems were obvious the moment the decision was taken. Non-standalone meant carrying two cores in parallel, the EPC and the new 5G core. That doubled everything: testing, integration, operations. Every new feature had to be checked twice, and even then, it would not behave the same across both paths.Worse, most of the real innovations we had been working toward could not even run on NSA. Network slicing, ultra-low latency, voice over new radio, all of that depended on the standalone core. With NSA, the user got a faster radio link but was still anchored in a 4G brain. It was like upgrading the wheels of a car and leaving the old engine under the hood.