Anthropic’s Fable 5 is back after the US feds lifted export controls, and it’s a monster: the most capable generally available model on the planet, shipped with the strongest safeguards Anthropic has ever built. Now we know governments can switch models off overnight. That risk isn’t going away. The lesson: keep your stack flexible so you can pivot to the next-best model and maintain high-quality inference. Don’t let the man bring you down!
Shout out to Susana Schwartz and RCRTech for featuring my “SKT’s stake” blog as an op-ed in their AI Infrastructure Daily newsletter. The setup: SK Telecom wrote one $100 million check for equity in Anthropic in 2023, and that stake could be worth $1–2.6 billion ahead of its IPO. Morningstar reclassified SKT as an AI stock and its P/E ripped from 11x to 60x (AT&T sits at 7x, Deutsche Telekom at 15x: still priced as utilities.) Twenty years of telco-to-techco consulting programs, and SKT did it with a single check. The platform layer is now worth more than the entire infrastructure layer beneath it. Read the piece, then think about how to use your unique data to build AI products, instead of being a landlord for someone else’s.
Context must live separately from your AI agents. It’s the insight of the year and it’s buried in the framing of the new Gartner and Verizon paper, How to Build an AI Platform at Scale. (Paywalled; also distilled on LinkedIn here.) Agents come and go—you’ll swap frameworks, models, and vendors constantly—but the layer that knows what “customer,” “order,” and “eligible” actually mean has to persist underneath all of them. Gartner predicts 60% of agentic projects relying on model context protocol (MCP) alone will fail by 2028 for exactly this reason: lack of enterprise context. Verizon wrote the paper and it is embarking on building it out. Want the killer shortcut? Use our telco-specific ontology at Totogi (where I’m CEO) and get a head start on this foundational problem. Give it a try!
SK Group just committed $642 billion to AI datacenters, with SKT in charge of building 15GW of capacity—5GW by 2029—anchored by 1,000+ Nvidia B200s and a “sovereign AI cloud” mandate from the Korean government. That would make SKT one of the largest AI datacenter operators on the planet, and I get the appeal: the government is the anchor tenant, and somebody has to build Korea’s AI backbone. But GPUs refresh in 18-month cycles while datacenters pay back over decades, and that math doesn’t care how patriotic the workload is. Nearly two-thirds of a trillion dollars buys a lot of depreciating silicon. Before you copy this playbook, make sure you have SKT’s government mandate, because without one, you’re just a GPU landlord with better branding.
SpaceX is now courting Charter for a consumer mobile deal after AT&T, Verizon, and T-Mobile all said no to a mobile virtual network operator (MVNO). Analysts note that Charter’s Wi-Fi hotspots and small cells can’t close the coverage gaps in the direct-to-cell satellite network without Verizon’s network underneath, so this version wouldn’t give SpaceX what it wants. But Charter stock jumped 9% on a rumor, which tells you exactly what the market thinks proximity to Starlink is worth. Musk will find a willing partner eventually; he always does. When your phone rings, don’t sell yourself as spectrum and towers. Structure it like SKT structured Anthropic: equity and upside.
AWS telco chief Jan Hofmeyr says operators face an AI “cost reckoning” if they keep pointing generic large language models (LLMs) at their networks. LLMs were built to consume text and video, but networks speak in protocols and telemetry. His fix: run agents on top of a network-aware context layer to translate. That’s now a hyperscaler, Gartner, and Verizon all landing on the same architecture in the same month. The context layer has moved beyond a Totogi talking point to become the consensus. Context matters. Totogi said it first.
Palantir CEO Alex Karp went on Squawk Box and delivered the most important argument of the quarter: enterprises are paying for tokens while the frontier labs capture the value, and the real money sits in the platform layer that turns context into action. Every telco exec should borrow Palantir’s nine-point framework about AI sovereignty: who keeps the data, who controls the weights, who owns the alpha? Ask your AI vendors those three questions this week. If they squirm, you have your answer.
