On Friday, SpaceX launched Starship Flight 15 from Starbase. The ship reached orbit, dropped off 26 Starlink V3 satellites, circled the Earth 14 times, and came back to Starbase on Saturday afternoon, landing in Mechazilla’s chopstick arms, the first time anyone has caught an orbital spacecraft at its own launch pad. Forget the money for a second. That maneuver is a 170-foot vehicle coming out of orbit, flipping upright, and parking itself inside a tower. A team of engineers made it work on the first try. The same team ships rockets, satellites, chips, and a cellular network, and it iterates every few months because that’s the only way to reach Mars. Unfortunately for telco, Mars needs a communications system, so one of the best engineering teams on the planet has wandered into your lane to build one. They never wanted your business. They need to solve your problem on the way to theirs, and they’ve found customers willing to pay them while they figure it out. 🚀
The Washington Post reports SpaceX is in talks to buy or partner with Boost Mobile, EchoStar’s 7.4-million-subscriber MVNO, after AT&T, Verizon, and T-Mobile all refused to host Starlink Mobile on their networks. On Deutsche Telekom’s Q3 call, Tim Höttges did the math: a greenfield US network needs at least $200 billion, T-Mobile holds 400 MHz of spectrum to SpaceX’s 76 MHz, and renting towers alone runs $20 billion to $30 billion. “Welcome to the party,” he said. Two days later, Musk answered that SpaceX could buy a legacy carrier outright and use its spectrum “ten times more effectively.” The industry keeps warning Musk about the price of admission as if he were a startup with a pitch deck. SpaceX closed Friday worth $1.89 trillion, more than three times AT&T, Verizon, and T-Mobile combined. Elon is the richest man alive, who hasn’t ever stopped because an incumbent told him the economics didn’t work. Plan for the competitor who arrives, and stop hoping he gives up. 🚀
Synergy Research says enterprises spent $168 billion on cloud infrastructure in Q3, up 55% year over year, the twelfth straight quarter of accelerating growth. The market has doubled in eight quarters, AI services are now a third of it, and SpaceX, which launched a cloud business this year, is already a top-ten provider by revenue. Even with that much money coming in, Synergy says Amazon, Microsoft, and Google are losing share to neoclouds that did not exist three years ago. This is a market where a trillion-dollar company with unlimited capital falls behind in a single year. Telcos run on seven-year depreciation schedules, 18-month procurement cycles, and sovereign AI business cases that take a quarter to approve. By the time your data center is live, the GPUs inside it are two generations old and renting for less than you paid. If the fastest companies on Earth cannot hold their position here, are you sure this is your arena? 🤔
The distance between people working at the frontier of AI and everyone else is “almost incomprehensible,” and the lines are moving farther apart, per Aaron Levie, CEO of Box. His reasoning: had AI progress leveled off, everyone would have caught up and standardized on the same tools, and the advantage would have disappeared. Instead the curve went vertical, so anyone who codified their AI workflows six months ago is already behind. Telcos run $200 million to $500 million transformation programs launched with RFPs, planned as two- to three-year timelines, and abandon 30% to 50% of them before they finish. Those timelines are relics of a world where the target held still. Reevaluate every AI decision your team made in 2025 because today AI can do many things it couldn’t a year ago.
In October, AI wrote 90% of the code at Proton, the Swiss company behind Proton Mail, according to internal meeting audio obtained by 404 Media. Its CTO Bart Butler told engineers “your old job is gone.” Its CEO said 2026 is the last year Proton hires engineers. Starting next year, it hires product managers who build with AI. Amazon just rewrote its engineering role guideline with the same message, telling engineers to get close to customers and the business because that’s where AI pays off. The person who replaces the engineer in both stories is someone using AI who understands the customer AND the business. In telco, the person who knows how your rating rules work has become more valuable than the person who knows Java. Every role in your IT and network organization is changing this way, and the redesign belongs to your CHRO as much as your CTO. Proton and Amazon have started. Have you?
Fierce reports that decision APIs quickly went from “obscure to everywhere.” TypeSafe AI shipped the first one in September. Microsoft, AWS, OpenAI, Cloudflare, and Perplexity all followed within six weeks. Decision APIs are small, fast models with a few hundred million to a few billion parameters that return structured answers, sometimes in milliseconds. The use case that should get your attention is one Fierce lists almost in passing: gating AI agents by checking whether an action is permitted. A decision model answers that question by inference, which means it’s still guessing. In your business, whether a past-due customer can be upgraded is a rule, written down, and the agent should read it from your rulebook rather than ask another model to estimate it. That rulebook is what the Totogi Ontology is: your rules, encoded once, so any agent can only attempt the actions you allow. Use decision APIs for fraud triage and model routing, where fast and mostly right is fine. Leave the question of what your agents may do to the ontology.
In a survey of 600 senior AI leaders, 57% still can’t generate returns that beat their AI spend, unchanged from a year ago. Forbes columnist Gary Drenik’s diagnosis is that AI dies in the last mile, between a working model and a business that actually runs on it. His prescription is forward-deployed engineers (FDEs) embedded in the workflows. He’s right, and Palantir built a $400 billion company proving it. Totogi runs the same model for telcos. Our FDEs sit inside your operation, build your ontology from your systems and your rules, and ship working software against your real backlog. In Düsseldorf last year, one of our FDEs built six BSS modules, live, in nine hours; the same build today takes about two hours. Work with Totogi, and at the end of the engagement you get an ontology you own, which every future agent and every future model plugs into. Your AI roadmap has been a PowerPoint for two years. Send us the deck, and we’ll hand it back as running code.
Amdocs shares fell by half on November 4, to their lowest level since 2014, after new CEO Shimie Hortig guided fiscal-year revenue down 14% to 18% and earnings per share down 26% to 30%. The cause is a legal dispute with T-Mobile US (which Omdia estimates supplies about a fifth of Amdocs’ revenue), over a recently signed agreement that went to arbitration last quarter. Hortig called it an isolated case, yet even excluding planned business exits, revenue is guided down 7% to 11%, which is more than T-Mobile cutting back. Bank of America’s Tal Liani asked the million-dollar question: now that T-Mobile “decided to do things in a different way,” what stops the next big customer from leaving? Hortig’s reassurance was a new agreement with AT&T to implement aOS, the agentic AI platform, which puts the same services engagement under a new name. Light Reading quoted my February blog on why that matters: more than 80% of Amdocs’ revenue depends on services-driven delivery. When software does the integration work people used to do, pricing power goes with it. It seems like T-Mobile is trying something different. Maybe it's time you make a move like T-Mobile and take back your telco, too.