Benedict Evans’ latest presentation, AI eats the world, argues that AI is becoming a general-purpose infrastructure layer, like cloud or electricity, but its lasting value will likely be above the models—in apps, workflows, and product design. The Big Three hyperscalers (plus Meta) are spending $700 billion on AI infrastructure this year alone, and Evans’ point is clear: building the commodity layer is a losing game unless you’re already at planetary scale. The winners will be whoever turns raw model capability into useful, sticky, scalable software people actually adopt. For telcos, that means deploying to production—not admiring the platform shift from the sidelines.
China Telecom released new AI token subscription plans starting at $1.45/month for 10 million tokens, bundled with access to DeepSeek and its own Telechat model. China Unicom, China Mobile, and Jio are all racing to offer the lowest-cost token. Telcos monetized voice-to-data by selling megabytes. Now they’re betting they can do it again with AI tokens. But the model providers are already selling directly to consumers. Telcos inserting themselves as token resellers risks the same disintermediation that killed carrier app stores and content portals. The real play isn’t reselling someone else’s AI; it’s using AI to make your own services indispensable.
KDDI just unveiled its three-year strategy through March 2029 built entirely around what it calls the “AI-native society,” which is AI embedded into everyday life, not just in the network. The plan includes $7.5 billion in infrastructure investment, 100,000 base stations repurposed as real-time sensing hubs, Starlink direct-to-device (D2D) already live, and a convenience store chain (Lawson) turned into AI-enabled customer touchpoints. This is what it looks like when a telco stops thinking about AI as a network optimization project and starts redesigning the entire business around it. Using AI, KDDI is building a new kind of company. This is one to keep your eyes on… 👀
SK Telecom is planning to build its own AI stack, partnering with Arm and chip startup Rebellions to co-develop custom inference servers and software. It plans to run its 519-billion-parameter Korean-language foundation model on the hardware, and then sell the whole package globally as sovereign AI infrastructure. So SKT wants to be an AI chip, model, and cloud provider all at once, entering into a game with Martingale stakes, where the leaders are already betting $700 billion this year and doubling every round. SKT is one of the most innovative telcos on the planet. I’d love to see that energy focused on the software and services layer where telcos actually have structural advantages.
Michael Dell says telcos have “some of the richest data sets of any company in the world” and should build sovereign AI on Dell’s disaggregated server architecture. But servers are just the hardware layer. To actually run AI, you still need models, orchestration, MLOps, and the full-cloud services stack that hyperscalers bundle natively. Dell’s answer is bolt-on partnerships with Google, OpenAI, and Palantir, which means telcos get to assemble it themselves. For operators already managing 200+ systems, “buy our servers and integrate the rest” is a little self-dealing coming from Dell. I’d stick to the public cloud … ⛅️
TM Forum surveyed 200+ IT execs at 117 operators and found one in three now cite Open Digital Architecture (ODA) as their core transformation framework, with more than half already embedding ODA elements. That’s real momentum! ODA gives operators a shared modular architecture and standardized APIs, which is necessary infrastructure. But ODA defines how systems connect, not what the data means when it crosses those connections. “Customer” in your billing API and “customer” in your provisioning API are still two different things. You need an ontology alongside your ODA—like the Totogi Ontology—that resolves what your systems mean, not just how they talk.
TM Forum’s Willie Stegmann has a refreshingly honest take on telco transformation: it never stops. It was digital, then cloud, and now AI is forcing another reinvention. His key insight: AI use cases cut across business, IT, and network teams, so if those groups operate in silos, AI adoption stalls. The orgs using AI at scale are connecting tasks and insights across functional groups—that’s where the 10x value lives. At Totogi, we’re seeing this firsthand: one customer uses our ontology to connect cell tower data with energy optimization, another to link network insights to advertising spend. The ontology is what stitches all the AI together.
TelecomTV just launched ANTA, the AI-Native Telco Accelerator community, with Deutsche Telekom, Orange, NTT Docomo, Rakuten Mobile, and Axiata on the steering board—plus Totogi and Wind River as founding technology partners. The mission: help operators stop running AI pilots that go nowhere and start deploying at scale. This is the group that’s going to figure out what AI-native actually means for telcos in production, not in PowerPoint. Totogi is proud to be a part of this innovative group! 🙋🏻♀️
The Independent ran a piece on easy ways for consumers to cut monthly expenses, and TelcoDR company Goji Mobile made the list. CEO Thad Hwang told readers that switching from a brand-name carrier to an MVNO can save $70 to $100 per month, and most people never notice a difference in coverage. On Goji’s marketplace, 74% of users pick an MVNO once they see plans side by side. Operators, this is your wake-up call: when the only differentiator is price and coverage is identical, you lose. What are you offering subscribers that makes them stay: personalized plans, real-time offers, loyalty that actually means something? You can’t ride postpaid coattails forever. If you want to talk about how to personalize at the individual subscriber level, call me.
Here’s some really exciting personal news: I have been drafted to my hometown pro pickleball team, the Austin Ignite! I’ll be playing in the Major League Pickleball (MLP) Champions Series, which will have three co-appearances with MLP events this year: Chicago July 23-26, and Dallas August 6-9 and October 30-November 1. Swing by and check it out!
