Anthropic just released Claude Opus 5.5, making frontier AI a commodity. The new model produced slightly better code than OpenAI’s GPT-6 Astra on Cognition’s FrontierCode 1.1, at about one-fifth the cost per task. This is the second Opus release since July, following a ten-day stretch in which Anthropic, Google, Meta, OpenAI, and DeepSeek each launched a new AI model. Your competitors got every one of those upgrades the same day you did. Any AI strategy built on having the best model has an expiration date. Rent the model, and put your money into the part that compounds: the context only you own.
Not to be outdone, Google announced Gemini 4 Argon, its “most advanced model yet.” The catch: you can’t use it. It’s only available to “trusted cyber defenders” in Google’s Fairwind Program, because Argon is trained to hunt down, verify, and fix software vulnerabilities on its own. Developers and enterprises get access later. Naturally, Google’s charts show it beating OpenAI’s GPT-6 Astra, Anthropic’s Fable 5.1, and the new Opus 5.5. Telcos running 30-year-old code should want an AI bug-hunter ASAP, but there will be another “best model ever” in a month or so. Don’t marry one LLM. Build so you can swap in the leader of the week.
Deutsche Telekom is replacing phone menus with AI agents, and finding that it only takes a couple months to handle 40% of calls without a human. But each 20% beyond that is exponentially harder, and getting past 80% is “really, really hard,” because all that’s left are edge cases that combine several problems. DT is making smart calls: owning its evals, exposing a model context protocol (MCP) layer so partners stay swappable, and running every agent inside guardrails. DT operates across 10 countries, each with its own tech stack. MCP can give its agents a consistent way to access those systems, but not a common vocabulary. That last 20% is an ontology problem. Solve it once, and getting to 100% stops being a fantasy. 🎯
Google is building data centers in space. Project Suncatcher just launched Trillium TPUs on a SpaceX rocket to test AI compute running on up to 8x the solar power available on Earth, with a laser-linked satellite pair due in 2027. Telcos, meanwhile, are breaking ground on sovereign AI data centers. You can’t win a compute war against companies that treat low Earth orbit as expansion space. The GPUs you buy today are obsolete in 18 months. Your customer relationship is the one asset Google can’t launch. Let them fight over orbit. Own the subscriber.
McKinsey’s Ivan Ostojic argues that agentic AI could erode the developer lock-in protecting API resellers and make telcos’ harder-to-replicate network signals more valuable. Twilio shows why the identity layer is worth watching: in Q2 2026, it reported 17% organic revenue growth, while its Verify product grew more than 30% selling identity. Twilio aggregates that identity data across carriers, wraps it in a clean API, and keeps the margin while operators collect wholesale fees. When AI agents start buying, booking, and paying on your subscribers’ behalf, someone has to prove they’re authorized. Make sure it’s you.
85 Group runs 21 MVNOs on one charging platform, and each one runs like it has its own. One of South Africa’s largest mobile virtual network aggregators (MVNAs), 85 Group now routes all traffic through a single master tenant on Totogi’s multi-tenant charging, while every MVNO controls its own plans, balances, pricing, and promotions. That includes offers the old system couldn’t support, like a 250MB bundle shared across WhatsApp, TikTok, Instagram, and Facebook. Launching the next brand is as easy as adding a tenant; two more go live in November. We moved all 21 over in a single cutover of under four hours. MNOs, why aren’t you managing your MVNOs this way? 🤔
Verizon CTO Yago Tenorio says 6G should be built from use cases up, and he’s signed NVIDIA, AWS, Cisco, Intel, and 11 other partners to his 6G Innovation Forum. 5G had use cases too, and the industry spent hundreds of billions building for customers who never paid. This time around, start with a simpler question: which of your cells actually make money?
That’s exactly what VIVA Bolivia is doing with the Totogi Ontology. Normally, revenue, cost to serve, and network performance live in a myriad of systems, so most operators settle for a once-a-year consultant study of sampled sites. Our ontology connects VIVA’s commercial and network systems (on-prem and in Azure) into one model of the business, without replacing any of them, and ranks every cell site by the revenue it generates. Spoiler: the busiest sites aren’t always the most profitable. The industry bet big on 5G using traffic charts. Don’t repeat that mistake with 6G. 🚀
