SKT’s stake

The core problem: SK Telecom’s $100 million equity stake in Anthropic prompted Morningstar to reclassify it as an AI stock, expanding its P/E from 11x to 60x while AT&T (7x) and Deutsche Telekom (15x) remain priced as utilities. The rest of the telecom industry is spending its AI budget on GPU clouds and API credits: infrastructure and consumption, the same position in the value chain telcos have always held. The market rewards ownership of the AI platform layer, and SKT is the only telco that bought in—even though our industry has the raw materials for an AI platform that no hyperscaler can replicate.

I don’t know if you noticed, but SK Telecom is now considered an AI stock after a recategorization by Morningstar.

Not a telecom stock with AI exposure. An AI stock. Shares are up more than 40% this year.

How did that happen? With one check.

In August 2023, SKT wrote $100 million for equity in Anthropic at a $5 billion valuation. In early June 2026, Anthropic filed its S-1 to prepare for an initial public offering (IPO), and its most recent private round valued the company at nearly $1 trillion. SKT’s stake is now worth somewhere between $1 billion and $2.6 billion—a 10x to 26x return in under three years.

In October 2024, a little over a year after SKT’s initial investment, T-Mobile reportedly agreed to pay OpenAI about $100 million over three years to use its AI technology for the IntentCX customer-experience platform. Same price tag, but T-Mo paid to consume the technology rather than invest in it.

For one telco, the $100 million is now worth billions. For the other, it’s a diminishing credit on the balance sheet.

Before the Anthropic investment, SKT traded at a P/E of roughly 11x, indistinguishable from any other telco on earth. AT&T trades at 7x and Deutsche Telekom at 15x. Today, SKT trades at 60x. A 60x multiple is what you pay for NVIDIA or Microsoft, not a phone company. The telecom industry has spent the last decade talking about the “telco-to-techco” transformation, pouring billions into consulting-led programs to get there. SKT did it with a single check.

And if you understand why the market responded so positively, the implications for every other telco are enormous.

History repeats itself

Every major technology platform shift follows the same structural arc: value migrates from the infrastructure layer to the platform layer, and the infrastructure providers who funded the transition end up on the outside looking in. Just ask Benedict Evans.

In the late 1990s and early 2000s, telcos poured hundreds of billions into fiber and broadband buildouts—the infrastructure that made the consumer internet possible. Google, Amazon, and Facebook built the platforms that rode on top of it. The platform layer became worth trillions while the infrastructure layer became a commodity priced at single-digit P/E multiples.

Then starting around 2007, telcos subsidized smartphones and rolled out 3G and 4G networks, expanding the mobile broadband that made the app economy possible. Apple and Google captured the platform layer while telcos carried the traffic.

It’s happening again. SpaceX just priced its IPO at $1.77 trillion. Anthropic is approaching $1 trillion, and OpenAI is north of $850 billion. Three companies are going public in the coming months with a combined market value of roughly $3.6 trillion. The entire global telecom industry is worth about $3.1 trillion. The platform layer is already worth more than the infrastructure layer beneath it.

And here’s what makes this so frustrating for me. Telcos have raw materials for an AI platform that no hyperscaler can replicate: real-time network data across hundreds of millions of connections, billing relationships that no tech company has, location and behavioral data at population scale, and deep operational knowledge of how networks, services, and customers actually interact. That’s the foundation of an AI platform only a telco could build.

Looking at this opportunity, what is telco doing about it?

Building more infrastructure

SoftBank just launched a GPU cloud and announced its “Telco AI Cloud” vision, integrating large-scale GPU data centers with AI-RAN edge computing. Deutsche Telekom and NVIDIA built an Industrial AI Cloud in Munich, powered by up to 10,000 Blackwell GPUs. Across the industry, the prevailing AI strategy is some combination of GPU-as-a-service (GPUaaS), sovereign AI clouds, and AI-powered radio access networks.

That’s right: building more infrastructure.

These are real investments, and I don’t want to diminish them. But they are also, every single one of them, infrastructure plays. We’re doing the thing we’ve always done, except now we’re building infrastructure for AI workloads instead of voice and data. We’re building GPU racks instead of cell towers and AI data centers instead of central offices. Different hardware, same business model, same position in the value chain. This is our industry’s definition of an AI strategy: become the landlord for someone else’s AI.

Here’s the thing: SKT is building AI infrastructure too. It launched GPUaaS; it’s building a hyperscale AI data center in Ulsan, South Korea; it’s doing AI-RAN with NVIDIA. But the company also did something none of the others did: bought a stake in an AI platform company. It moved up the stack. And the market didn’t reward it for the data centers. It rewarded the Anthropic stake.

How do you build a platform?

Here’s what I keep coming back to: every telco executive I talk to says AI is a strategic priority. The budgets are growing, the pilots are multiplying, and there’s genuine energy around what AI can do for our industry. And yet not one of them has bought the thing the market actually pays for. The enthusiasm is real. The capital is real. It’s all going into the two categories telcos have always operated in—and none of it into the layer that gets a telco reclassified.

Now look at the full picture of what SKT is actually doing. Yes, it wrote the Anthropic check. But it also built A.X K1, a 519-billion-parameter sovereign foundation model that it’s currently expanding in a phase-2 buildout. It launched A-dot, an AI personal assistant with over 10 million subscribers, and Aster, agentic AI built for global markets. It created an AI Company-in-Company to consolidate all AI businesses under one roof. It’s building AI solutions for manufacturing with SK hynix. The Anthropic investment got the headlines, but the strategy behind it is a full-stack platform play: foundation models, AI products, consumer services, enterprise solutions… and yes, infrastructure too.

That’s the difference. SKT is building AI products, AI services, and AI businesses, and investing in the platform layer at the same time.

Telcos need to realize that AI is more than another workload to host or another tool to run through procurement. It’s redefining work, redefining industries, and redefining what a telco can be. And telcos could have an amazing AI platform, if only they would build it.

SKT understands that. It didn’t stop at investing in someone else’s platform. It built products, shipped them to customers, and used its unique data to compound its advantage. That’s what AI companies do. Infrastructure companies build capacity and rent it out.

For twenty years, the question in every telco boardroom has been the same: how do we build the capability that tech companies have? It’s the wrong question, and it’s why we keep defaulting to the one thing we know how to build, which is infrastructure. SKT didn’t set out to become an AI company from scratch. It bought a position in one, then built around it. The platform layer wasn’t something it had to invent. It was something it could own.

The right question for your board is, what do you actually want to own? SKT already knows the answer.

Instead of doing what we did in the late 1990s, and again in 2007, let’s try something different this time around.

Recent Posts

Get my FREE insider newsletter, delivered every two weeks, with curated content to help telco execs across the globe move to the public cloud.