Global telecom CapEx is projected to fall 2% in 2026 and Dell’Oro projects just 1% annual growth through 2030. Meanwhile, five US hyperscalers are spending more than $700 billion this year, with 75% aimed at AI infrastructure. That’s a 2:1 ratio—and it’s widening. Operators are tightening belts post-5G while everyone else is doubling down on AI. This kind of slowness is how you end up watching someone else build the flywheel. The companies spending aggressively on AI right now aren’t being reckless. They understand that AI compounds and the cost of waiting is permanent. Will telco get off the sidelines? Time to get moving.
Margins are down and costs are up for Ericsson, caused by AI demand. And it’s going to get worse, the CFO warned in the vendor’s latest earnings call. The industry has spent two years promising that AI-RAN will transform networks. What Ericsson’s numbers reveal is that AI demand is eating the equipment supply chain before operators have written a single check for the AI-native future. Huang’s Law is showing up in Ericsson’s cost base, but not yet in its revenue. How long will it be able to cover the gap?
Verizon CEO Dan Schulman cut 13,000 jobs in his first month on the job, then told the Wall Street Journal that AI will drive unemployment to Great Depression levels within five years. I love the honesty, but I don’t buy the doom. More new jobs will be created because of AI than will be eliminated. What I can get behind: Schulman’s $20 million reskilling fund and his challenge to other Fortune 100 CEOs to match it. That’s leadership. Redesign every role AI-first, fund the transition, and get your Chief HR Officer in the room. The companies that move fastest will attract the best talent.
At FutureNet World in London, Vodafone’s CTO Scott Petty said it plain: Vodafone doesn’t buy GPUs anymore. It only rents them, because AI workloads lifecycle every 18 months and nobody runs a telco CapEx program on that clock. Orange’s Laurent Leboucher piled on, saying telcos should compete on “connectivity-plus,” providing trusted, sovereign network infrastructure with AI capabilities on top. Yay! This is music to my ears. Stop trying to own the AI stack. Rent the compute, own the network, build on top of hyperscaler infrastructure the way every smart software company does. That’s the modern telco playbook.
Is it the end of public cloud egress fees? Amazon Web Services’ (AWS’) Interconnect – Multicloud is now generally available, meaning customers can securely move data between AWS and Google Cloud today, with Azure and Oracle coming later this year. AWS has replaced per-GB egress fees with flat bandwidth billing. The telco angle: Lumen is the first partner for AWS Interconnect – Last Mile, with AT&T and Megaport joining next. Hyperscalers are building the intelligence layer; operators are becoming the physical on-ramp. That’s not a bad role, but only if operators price it like strategic infrastructure, not commodity bandwidth.
Deutsche Telekom CEO Tim Höttges wants Germany to be a global leader in physical AI, offering sovereignty for European data centers. The industrial AI thesis is real, as Germany’s manufacturing base is a genuine asset. But the sovereignty framing conflates two different problems: data residency, which hyperscalers already solve with in-country regional zones, and infrastructure ownership, which is an expensive GPU ownership trap. Instead of trying to compete in a race you can’t win, invest that capital in your own AI intelligence layer—the software, the data, the ontology that makes your network and your subscribers uniquely valuable.
Is 2026 the year of the AI ontology? After chasing agentic AI last year, the industry is now realizing that the prerequisite was always semantic consistency. IEEE ComSoc’s tech blog quotes leaders from Telstra, BT, and Vodafone, all saying the same thing: without shared semantic grounding, multi-agent systems produce “agent drift,” or different agents interpreting the same concepts differently and generating errors at machine speed. This is exactly what we’ve been building at Totogi. The Totogi Ontology is live and in production at Tier-1 operators, and it’s why our AI agents work. If your agentic AI strategy doesn’t start with semantic consistency, your agents will be confident, fast, and wrong.
Anthropic is playing both sides of the Atlantic, briefing European Commission officials on AI models including cybersecurity-focused ones not yet available in the EU, while CEO Dario Amodei meets Trump administration officials as its Pentagon blacklisting lawsuit is still live in court. The foundation models powering an increasing share of your AI vendor stack are caught in geopolitical crossfire, which is exactly why you need your own intelligence layer underneath them. Models will come and go. The ontology, the data, the semantic consistency you build on top—that’s yours. That’s what compounds. So go own it!
