SpaceX is acquiring EchoStar’s unpaired AWS-3 spectrum licenses for $2.6 billion—73% below book value because EchoStar couldn’t meet FCC buildout deadlines. Combined with SpaceX’s $17 billion AWS-4 purchase and 650+ direct-to-cell satellites already serving six million people, the strategy crystallizes: Musk is systematically acquiring the rural footprint telcos never properly monetized. Over 50% of Earth’s land mass lacks terrestrial coverage, and telcos spent decades avoiding those unprofitable tower builds. SpaceX is turning your coverage gaps into its market entry point. Dense urban areas still need ground infrastructure—satellite can’t handle stadium crowds or subway tunnels—but for the millions living in rural America paying $120/month for spotty service? SpaceX offers “cell towers in space” working with existing phones, no hardware changes required. Next-generation satellites launching in 2026 will deliver 100x capacity increases and full 5G speeds comparable to terrestrial LTE. And since SpaceX owns Starship, deploying thousands of upgraded satellites costs them the launches they’re already doing, not the decade-long vendor contracts telcos need for rural buildouts. When regulators punish slow buildouts with forced asset sales, first movers with capital and execution velocity don’t just win spectrum auctions—they cherry-pick which parts of your addressable market to competitively attack.
South Africa’s MTN completed a migration of its Enterprise Value Analytics (EVA) platform to Microsoft Azure, creating the largest telco cloud implementation in Middle East and Africa—processing 22 billion records daily through 800+ analytics workflows pulling from 1,700 data feeds. Built on Azure Databricks with Microsoft Defender security, EVA 3.0 delivers real-time insights into service performance and customer behavior, letting MTN detect issues and design more relevant offers before customers complain. The kicker? MTN’s Cloud Centre of Excellence achieved over 1,350 Azure certifications in 2025—more than any other African organization—proving its building genuine cloud expertise, not just buying vendor services. This isn’t a proof-of-concept or a pilot—it’s production infrastructure at hyperscale, and MTN plans to template it across its other African markets. While developed market telcos debate cloud strategies and worry about hyperscaler lock-in, MTN just built the blueprint for modern telco analytics and trained a continent-wide engineering team to run it. Bravo!
Finally Verizon makes a move I can get behind. Verizon Business and Amazon Web Services (AWS) have partnered up to build high-capacity, low-latency fiber routes linking AWS’s data centers using Verizon’s AI Connect portfolio. AI workloads require 10–100x more bandwidth than traditional cloud services, and Scott Lawrence from Verizon says network traffic from AI will grow dramatically through 2030. Instead of competing with hyperscalers on AI infrastructure or trying to build GPUaaS offerings that’ll get crushed, Verizon recognized it owns something AWS needs—long-haul fiber routes between data centers. The operator is building dedicated “fiber superhighways” for AI traffic, becoming essential plumbing rather than fighting battles it can’t win. This is the telecom transformation playbook—leverage your actual assets (nationwide fiber) to serve the players building the AI economy, not waste capital pretending you’re one of them.
Deutsche Telekom and NVIDIA are building a €1B AI factory in Munich—one of Europe’s largest AI infrastructures. Launching Q1 2026, the “Industrial AI Cloud” will house over 1,000 NVIDIA DGX B200 systems with up to 10,000 Blackwell GPUs, delivering 0.5 exaFLOPS of computing power. It seems as if DT is buying the chips, providing the data center, handling operations and security—while NVIDIA gets another telco building out its GPU infrastructure without capital risk. What’s DT’s business model here? Competing with hyperscalers on AI infrastructure is expensive, and we’ve seen NVIDIA use telcos to build distributed compute before. Unless DT can prove it’ll generate returns that justify competing with AWS, Azure, and Google on AI workloads, this looks like another case of telcos spending billions to become NVIDIA’s infrastructure arm. Remind me who’s going to be the winner here?
Microsoft shipped two proprietary AI models in August 2025: MAI-Voice-1, a highly expressive speech generation model that produces one minute of audio in under one second on a single GPU, and MAI-1-preview, Microsoft’s first in-house foundation model trained end-to-end without OpenAI involvement. The models signal Microsoft’s deliberate shift from total OpenAI dependence to strategic independence. For years, OpenAI’s GPT models powered everything Microsoft did with AI—Azure OpenAI services, GitHub Copilot, Bing chat. But tensions emerged as OpenAI grew more independent and Microsoft reportedly criticized GPT-4 as “too expensive and slow” for consumer needs. Microsoft hired Mustafa Suleyman (DeepMind co-founder) in 2024 to lead internal AI development and quietly began testing alternative models. Meanwhile, OpenAI has been reducing its own Microsoft dependence, raising billions in outside capital and exploring paths beyond Azure infrastructure. AI’s original power couple appears to be consciously uncoupling. It goes to show that both of their offerings—LLMs in OpenAI’s case, public cloud for Azure—are commodities, and both can diversify. The takeaway for telco execs: make sure you’re keeping your options open when it comes to model dependency.
Telefónica Germany and Tech Mahindra are also partnering with NVIDIA to create a Large Telco Model (LTM)—an industry-specific generative AI model built on Meta’s Llama 3.1 using NVIDIA AI Enterprise software. The model processes structured data (alarms, counters, events) and unstructured data (logs, images, procedures) to enable automated root-cause analysis, dispatch optimization, and predictive network management. Why are telcos burning resources training custom models to be 10% better at understanding telecom jargon when off-the-shelf LLMs already handle complex technical domains? The smarter play is to build an ontology—a structured knowledge layer that any LLM can use. Ride the wave of innovation that hyperscalers and AI labs deliver every six months, and invest your resources into the ontology that captures your network relationships, business rules, and domain expertise. That’s reusable across model generations and gives you the competitive edge.
Are MVNOs the new celebrity tequila? Celebrity MVNOs are having a moment following T-Mobile’s $1.35 billion Mint Mobile acquisition. In the last six months alone: SmartLess Mobile launched (podcasters Jason Bateman, Will Arnett, Sean Hayes), Trump Mobile debuted with Donald Trump Jr. pitching $47.45/month unlimited plans, and Noble Mobile went live with Andrew Yang as CEO promising cash back for using less data. The new entrants hope to replicate Mint’s success: leverage celebrity reach for constant visibility while cross-promoting across other ventures. But let’s not forget that Mint already had product-market fit and operational excellence before Reynolds joined—all he had to do was amplify an already-working business, not build it from the ground up. Running a telecom service requires ongoing customer support, network quality management, and retention focus, not just marketing charisma. Can’t wait to look back on these launches in a year when these celebrities realize mobile service isn’t a passive brand licensing deal—and they’re stuck nursing their own MVNO hangover. 😵
