Telecom Value Migration

AT&T spent $133.5 billion in acquisitions trying to escape the “dumb pipe” fate. It bought DirecTV for $48.5 billion in 2015 and Time Warner for $85 billion in 2018, betting that owning content would protect it from the value migration that had already begun. It failed. By 2022, AT&T had divested all of its media assets and retreated to being exactly what it had spent a decade and $133.5 billion trying not to become: a pure connectivity provider.

Meanwhile, Netflix — a company that started as a DVD-by-mail service with a $350 million IPO in 2002 — grew to $400 billion by 2025, surpassing AT&T in market cap in 2020 and expanding the gap to 2.27x. Netflix never laid a foot of fibre, never built a cell tower, never acquired a spectrum licence. It rented bandwidth at commodity rates and used it to build the world’s most valuable entertainment platform. The application beat the pipe. The content beat the conduit.

The telecom revolution is the closest historical analogue to the AI revolution. Both involve a transport layer — bandwidth in telecom, inference in AI — that becomes commodity while application companies built on top capture multiples of the value. The telecom story is complete enough to measure the endgame: OTT applications (Netflix, Spotify, YouTube, Meta) are now worth 5.7x the telecom infrastructure they run on. The crossover happened in 2020, just 22 years after the broadband build-out began. AI’s crossover is projected at 5–10 years.

The Empire That Could Not Escape

AT&T’s trajectory is the definitive case study in the infrastructure trap. At its 1999 peak, AT&T (then SBC) was valued at $283.5 billion and controlled the dominant voice and data network in America. Twenty-six years later, the company is worth $176.3 billion — lower than its peak in nominal terms, and roughly half in real terms. A quarter century of network build-out, 5G investment, and $133.5 billion in media acquisitions yielded negative returns. The company that once defined American innovation — Bell Labs invented the transistor, the laser, Unix, and the C programming language — became a commodity pipe provider.

The strategic lesson is devastating in its clarity. AT&T had every conceivable advantage: the network, the customers, the capital, the brand, the regulatory relationships. It tried every possible escape route — acquiring content (DirecTV, Time Warner), launching its own streaming service (HBO Max), bundling wireless with entertainment. None of it worked, because the fundamental economics of infrastructure commoditisation are more powerful than any corporate strategy. When the output is undifferentiated — and bandwidth is the definition of undifferentiated — competition drives prices toward marginal cost. No amount of vertical integration changes this fact.

The Rise of the Application Layer

Netflix’s trajectory is the mirror image of AT&T’s decline. From a $350 million IPO in 2002 to $400 billion by 2025, Netflix delivered a 1,144x return by building entirely on infrastructure it never owned. The company’s 2007 pivot from DVD-by-mail to streaming was the strategic masterstroke — it recognised that bandwidth would commoditise and that value would accrue to the content and experience layer above. Every phase of Netflix’s growth depended on telecom bandwidth: streaming requires reliable broadband, and each new resolution tier (720p, 1080p, 4K, HDR) demanded more of it. Yet none of the value went to the bandwidth providers. Netflix generated $33.7 billion in annual revenue with 70%+ gross margins; the telcos that carried every byte of Netflix traffic captured only commodity transport fees.

The crossover chart above is the single most important visual in this chapter. In 2010, Netflix was 5% of AT&T’s size. By 2015, 23%. By 2018, 56%. In 2020, during the COVID-19 pandemic, Netflix crossed AT&T at 1.17x. The brief reversal in 2022 — when Netflix lost subscribers for the first time in a decade and its stock crashed 70% — appeared to undo the crossover. But application-layer setbacks are temporary; value layer shifts are structural. Within two years, Netflix recovered and expanded the ratio to 2.27x. The crossover proved irreversible, just as the Microsoft-Intel crossover proved irreversible after 2003–2005. Once capital markets conclude that the application layer is worth more than the infrastructure layer, the gap only widens.

Netflix’s subscriber trajectory underscores the mechanism. From 7.5 million subscribers in 2007 (when streaming launched) to 325 million in 2025 — a 43x increase in 18 years. Every single subscriber requires telecom infrastructure to receive the service. Netflix consumes enormous bandwidth — at its peak, it accounts for roughly 15% of all downstream internet traffic in the United States. Yet the value accrued entirely to the application layer. The network effects are asymmetric: more subscribers fund better content, which attracts more subscribers. The telcos have no equivalent flywheel. More bandwidth users do not improve the bandwidth. This is the differentiation asymmetry that drives every value layer shift.

The 5.7x Ratio

The bar chart above tells the full story. Five OTT application companies — Alphabet ($2.2 trillion), Meta ($1.64 trillion), Netflix ($400 billion), ByteDance ($330 billion), and Spotify ($94 billion) — are collectively worth $4.66 trillion. The four largest telcos — T-Mobile ($241 billion), Deutsche Telekom ($201 billion), AT&T ($198 billion), and Verizon ($170 billion) — are collectively worth $810 billion. The application layer is worth 5.7x the infrastructure layer. Alphabet alone exceeds all four telcos by 2.7x. Meta alone exceeds them by 2x. The value is in the algorithm, the content library, the social graph, the advertising platform — not in the fibre and cell towers that deliver the bits.

YouTube’s revenue trajectory is perhaps the most striking illustration of the value migration at work. Google first disclosed YouTube ad revenue in 2017 at $8.15 billion. By 2025, it had grown to $40.37 billion in ad revenue alone, with total revenue (including subscriptions) exceeding $60 billion. A single video platform — one that produces almost none of its own content — generates more revenue than most traditional media companies combined. Every video watched on YouTube flows through telecom pipes. The bandwidth providers see none of this $60 billion. They provide the commodity transport and capture commodity returns.

The Physical Evidence

Cord cutting is the physical manifestation of the value migration. US pay-TV subscribers peaked at 105 million in 2010 — 88% household penetration. By 2025, they had fallen to 49.6 million — a 53% decline. In the same period, US streaming subscriptions grew from 110 million to 339 million. The lines crossed around 2017 and have diverged ever since. Pay-TV revenue declined from $100.1 billion (2017) to $84.9 billion (2023). Users abandoned $150-per-month cable bundles for $15-per-month streaming services — a 10x price reduction for a superior experience. This is not a technology story; it is a value story. The same dynamic is forming in enterprise software: organisations will abandon $50-per-seat SaaS for $0.99-per-outcome AI agents.

The Commoditisation Signal

ARPU decline is the commoditisation signal in its purest form. Global mobile ARPU collapsed 82% from $35 per month in 2000 to approximately $6 per month by 2024 (per PwC/GlobalData projections), and continues to decline. Even with the rollout of 5G — a massive infrastructure investment cycle — telcos have been unable to raise prices. Bandwidth is fungible. Consumers do not care which network carries their data, just as they do not care which power plant generates their electricity. This is the commodity trap: same users, declining revenue per user, growing demand. It is exactly the Jevons Paradox dynamic already documented for AI inference — as the price of intelligence falls, consumption explodes, but the per-unit revenue to infrastructure providers declines. The volume growth cannot compensate for the margin compression.

Meta’s trajectory is the most dramatic single-company illustration of the value migration. From a $63 billion IPO in 2012 to $1.64 trillion in February 2026 — a 26x return built entirely on telecom pipes. One social media company is now worth more than the four largest US and European telcos combined. Even after Meta’s 2022 crash to $319 billion (from metaverse spending concerns), the company recovered within two years and expanded the ratio from 0.49x to 2.0x versus the combined telcos. The value layer shift is stronger than any individual company’s missteps. Meta’s advertising platform, social graph, and data flywheel create value at a scale that no amount of network infrastructure can match.

Spotify completes the picture. A music streaming platform that owns zero network infrastructure, with a market cap of $94 billion — larger than most European telcos. Spotify went from a $26.5 billion direct listing in 2018 through a brutal crash to $16 billion in 2022, then recovered to nearly $120 billion by the end of 2025. The volatility was real, but the structural trend is unmistakable: the application layer captures more value than the infrastructure layer, and the gap only widens over time.

The Strategic Lesson for AI

The telecom revolution offers the most direct and uncomfortable lesson for today’s AI infrastructure providers. AT&T had every advantage — the network, the customers, the capital, the brand, the regulatory relationships — and still lost the value migration. It spent $133.5 billion in acquisitions trying to escape the infrastructure trap and ultimately divested every media asset it had acquired. The value migration was not driven by AT&T’s failure; it was driven by the structural economics of infrastructure commoditisation. When the output is undifferentiated, competition drives prices toward marginal cost. When prices converge, value migrates to the layer above — the layer that controls the customer relationship, the brand experience, and the network effects.

For AI, the parallels are precise. NVIDIA is AT&T circa 1999 — the dominant infrastructure provider at peak value during the build-out phase. Commodity model APIs are becoming the “dumb pipe” of AI — essential but undifferentiated. The AI Netflix — the application company that will capture multiples of the infrastructure value while owning none of the infrastructure — is being built right now. The question is not whether the telecom pattern will repeat in AI. The data from 140 years of infrastructure revolutions says it will. The question is which application companies will emerge as the Netflixes and Metas of the intelligence revolution — and how much faster it will happen.