In his DTW Ignite opening keynote this week, TM Forum CEO Nik Willetts told operators sovereign AI is “not about owning GPUs or building data centers,” but about controlling AI behavior in real time, with operators positioned as the audit and governance layer. He’s right on the first part. But the proposed solution doesn’t survive the market test. Enterprises—JPMorgan, Klarna, Walmart, Goldman—are already shipping production AI and they won’t wait for their telco to invent a safe version for them. Plus, AWS, Azure, and Google Cloud Platform are already shipping AI guardrails and releasing updates monthly. By the time operators stand up sovereign AI governance products, hyperscalers will have shipped 50 versions of theirs. Yes, telcos are trusted. But trust without velocity isn’t a product. Telcos should focus on things we can deliver with excellence. This is not it.
The new “AI-native ODA” rolled out at DTW Ignite last week, and the architecture is worth a close look. Agents sit at the execution layer, deciding and acting across BSS while the Canvas handles policy and audit. Translation: semantics get decoded by the model at inference time. When an agent resolves SLA compensation, it’s reconciling three definitions of “eligible customer” inside the model, every time, at the moment of action. That’s unpredictable. That’s unauditable. And it’s the architecture the whole industry is standardizing on. Totogi’s answer: solve semantics before the agent runs, not while it’s running, and model your allowed decisions. Then use that to take action. That’s how telco builds enterprise AI that works.
Sid Nag at Tekonyx says operators “messed up the whole cloud opportunity.” He’s right. Fierce Network gets the diagnosis right but the prescription wrong. Its argument that AI is the second chance (because inference must run close to the user) misses where the workloads are already landing. Most AI workloads will run on hyperscaler clouds where the software lives and where enterprises have already moved their data. The actual operator opportunity is the operational layer above the pipes: AI that recovers $200–600 million a year in revenue leakage for a Tier-1, plus subscriber AI services that only work with network context. Hyperscalers can’t replicate either. That’s the AI opportunity for telco.
SK Telecom’s AX Innovation 2.0 redesigns every job around AI agents. Agents get employee IDs, departments, job functions, and a lifecycle from “hiring to termination.” In their AI Sandbox trial, what used to be three jobs—planner, developer, designer—is now one human working alongside multiple agents. Compare that to an Amdocs managed services team: dozens of people supervising agents forever. Same word, “agent,” opposite architecture.new Operators that rebuild org charts AI-first will run leaner companies in 18 months than the ones buying augmentation as a service line. Redesign the role. Retire the headcount.
Finnish MEP Aura Salla nailed it: Europe cannot keep building its tech stack on access that can be “switched off overnight by a foreign government.” She’s right. The kill switch is real. But the current response, of telcos racing to build gigawatt-scale AI factories, confuses sovereignty with infrastructure ownership. There are two problems with the bet. The disruption may be brief (Kalshi traders expect Fable 5 and Mythos 5 back by July 1, weeks after the order), and gigawatt factories take years to build and decades to depreciate against a regulatory event that may last a month. Plus, the hyperscalers are already adding EU residency zones. Once those land, the sovereign compute premium evaporates and operators are left holding depreciating GPUs in a commodity market. The real sovereignty that operators can defend is the operational layer they own: subscriber data, network state, billing logic. Build there. Plug in whichever frontier model survives Washington’s next review.
What do I mean about hyperscaler sovereignty plans? Last week in Delhi, AWS committed $48 billion to India through 2030, with $13 billion specifically for AWS AI infrastructure in Mumbai and Hyderabad. Spain got €33.7 billion for the same buildout. France, €15 billion. That’s the AI infrastructure envelope from one hyperscaler, extended country by country, partnered with national governments under equivalent sovereignty frameworks. Operators thinking they can match this with gigawatt AI factories are bringing telco CapEx (seven-year cycle, debt-financed) to a hyperscaler-scale fight ($13 billion per country, equity-financed at 35x earnings). The math doesn’t work. The operators’ best bet is the operational layer above the pipes, or the connectivity layer beneath the data centers. Pick one. Don’t pick the middle.
Mavenir CEO Pardeep Kohli has a new revenue idea for telcos: sell AI tokens. Tokens. Buy NVIDIA GPUs, run open-source models Kohli himself admits are “six to eight months behind” OpenAI and Anthropic, and charge enterprises by the token. NVIDIA sells the chips; Red Hat sells the platform; Mavenir sells the orchestration. Three vendors win. Yet the operator carries depreciation on hardware that obsoletes in 18 months against a seven-year CapEx cycle. In AI today, six months is the gap between a product enterprises pay for and one they ignore. DR’s thought: skip this idea unless you can sell tokens of models people actually want to use.
Talk about knockout. The World Cup Round of 32 is underway, with the first knockout round in the new 48-team format running June 28 through July 3. The operator stats for this tournament are wild. Per Verizon (the official telecom sponsor): 50 terabytes of data per match inside a single stadium, equivalent to streaming HD video for over three years, for an expected 6 billion viewers worldwide. Verizon’s Broadcast Contribution Network uses 64x100G wave circuits to connect every stadium to the International Broadcast Center in Dallas, with a 3–5x bandwidth increase across all 11 US host stadiums. AT&T deployed over 2,000 network upgrades across 11 US host cities and engineered for 2–3x normal traffic. Per Ookla, T-Mobile led download speeds in every US host market. This is the work telcos deliver with excellence: high-density, multi-jurisdiction, real-time connectivity at scales nothing else on Earth tests. With the USA making the knockout round, I expect usage to only go up. Go USA! ⚽🇺🇸⚽
