O2 Telefónica is now running production 5G core network functions on AWS Outposts inside its own data center, with Nokia providing the cloud-native core software. The operator says the new 5G core is supporting roughly one million subscribers alongside the existing network. This is the first time an operator has run a production 5G core on hyperscaler infrastructure inside its own data center. I’ve been making this case since CLOUD CITY at #MWC21: put hyperscaler infrastructure in the operator’s facility and you get AWS tooling, APIs, and deployment velocity while data never leaves the building. The Outposts model doesn’t just answer the sovereignty objection—it makes it irrelevant, and with it the entire argument for bespoke telco cloud stacks that cost more and innovate slower. Every operator still running #fakecloud just lost the last excuse standing between them and hyperscaler economics in the core network.
To prepare for an “AI-driven digital economy,” AT&T committed $250 billion over five years to expand fiber, 5G, and satellite connectivity across the U.S.—roughly $50B a year including its existing $23–24B annual capex. The operator is getting ready to support autonomous vehicles, cloud computing, and other data-heavy services. But notice what’s not in the announcement: no GPU cloud, no sovereign AI play, no “we’ll compete with hyperscalers on compute.” AT&T is making the right bet—that the most valuable role in the AI economy is being the connectivity layer where everything runs, not trying to out-invest companies spending billions a year on data centers. Operators chasing AI inference workloads should study this press release carefully—then double check the AI bets they are making themselves.
SoftBank unveiled “Telco AI Cloud” at MWC26—a strategy to evolve from carrying data packets to hosting and processing AI workloads across GPU cloud, AI-RAN edge compute, and a custom software stack called Infrinia AI Cloud OS. The ambition is to transform the telco network into a distributed AI platform that comprehends traffic, not just transports it. SoftBank has Masayoshi Son’s conviction and deep NVIDIA ties behind the play, but here’s the math: each hyperscaler spends $50–70B per year on AI infrastructure, and SoftBank’s entire telco operation doesn’t generate that kind of free cash flow. Son’s Vision Fund burned through $100B learning that conviction doesn’t override unit economics—and GPU cloud infrastructure is an even more capital-intensive lesson. Softbank may want to stop trying to compete against $200B hyperscaler data center buildouts on their telco margins...
Telenor partnered with NVIDIA, Supermicro, and others to demo a “Telco Cloud Continuum” at MWC26, but the headline play is Telenor AI Factory—Norway’s first sovereign AI cloud, running NVIDIA GPU-accelerated infrastructure entirely within Norwegian borders. But I still don’t get it. Sovereign AI requires operators to build and maintain GPU data centers, compete for NVIDIA allocations, hire scarce ML talent, and keep pace with a hardware refresh cycle that moves faster than any telco procurement process. AWS, Azure, and Google Cloud already operate regions inside European borders with local data residency—sovereign cloud without the operator having to own the iron. Telenor is building a capital-intensive AI compute business while the hyperscalers will happily sell them the same sovereignty guarantees as a managed service. The smarter move is to be the connectivity and edge layer the hyperscalers need—not to become a GPU landlord in a market where your competitors refresh hardware every 18 months and you refresh every 7 years.
Amdocs’ GenAI division president says that GenAI success in telco won’t come from “magic” and that the future of AI is “people supervising an environment that has both people and agents.” That’s not an AI vision—that’s Amdocs’ business model talking. They have 27,000 employees whose jobs depend on humans staying in the loop permanently. The reason their AI needs constant supervision is aOS, their new agentic operating system, keeps the semantic resolution proprietary and locked inside their stack. But good news for Amdocs customers: Totogi does have magic that works, in the form of the Totogi Ontology and BSS Magic—give it a try!
Telecoms.com covered my MWC 2026 message on why scaling AI across a telco isn’t a weekend project—and StarHub in Singapore is the case in point. Their enterprise sales reps were using horizontal conversation intelligence tools that could transcribe calls but couldn’t tell you whether the deal was sellable, at what price, or through which fulfillment path. The Totogi Ontology encodes that logic—product constructs, pricing rules, eligibility constraints—so when a rep discusses a custom enterprise bundle, the system knows instantly what StarHub can actually deliver. Horizontal platforms start blank and stay blank. Vertical AI ships with the domain knowledge that makes agents useful on day one.
Most telcos are barely past square one on network autonomy, according to Accenture research showing 79% of operators are still at Level 0 or Level 1. Just 22% expect to reach Level 4—highly autonomous, AI-driven, mostly self-managing—by 2030. Accenture blames the usual suspects: legacy systems, technical debt, AI talent shortage. But those are symptoms, not the disease. The reason telcos can’t automate is that their systems don’t agree on what anything means—you can’t hand autonomy to AI agents that have to reconcile 15 definitions of “customer” before they can act. Fix the semantic layer first and autonomy follows. Leave it broken and you’ll still be at Level 1 in 2030, paying consultants to do what your AI should be doing. Want to get ahead of the pack? Totogi can help. Call me! 📞
Saying that the next wave of telco innovation is about people—developers, team structure, organizational change—is popular, and for once I agree with the conventional wisdom. AI is bringing the biggest shift in how we work in a generation, and every single job in your organization is going to change. You can’t just throw AI tools at your teams and see what sticks. You have to redesign each process, each role, to be AI-first. Not “how do we add AI to what we already do” but “what does this job look like if AI is the starting point.” If your CHRO isn’t leading this conversation alongside your CxOs, you’re already behind. This is the kind of culture shaping jobs strategy HR leaders DREAM of. It’s time to LEAN IN! 🙋🏻♀️
