Electricity Value Migration

In 1882, Thomas Edison controlled the entire electrical stack — generation, distribution, metering, and the application itself (the light bulb). It was the first vertically integrated infrastructure company in history. One hundred and twenty years later, electric utilities capture just 1.8% of US GDP while the economy they enable generates $28.3 trillion. The original infrastructure-to-application value migration set the template that every subsequent revolution would follow.

Edison’s model is the prototype for what NVIDIA and OpenAI are attempting today. Edison did not merely sell electricity — he built the power stations, designed the generators, laid the underground wiring, invented the meters, and manufactured the light bulbs. He sold both the infrastructure and the application, controlling everything from electrons to the end-user experience. This degree of vertical integration was rational during the build-out phase, when the entire system was new and components had to be co-developed. But as the system matured and each layer developed independent competitive dynamics, the forces of specialisation overwhelmed the benefits of integration.

The electricity revolution offers AI strategists the longest data set available — 140 years of value migration from infrastructure to application. The pattern is complete, the endgame is visible, and the implications are unambiguous. This chapter traces the full arc: from Edison’s Pearl Street Station, through GE’s rise and fall, the Rural Electrification Act, the appliance revolution, utility deregulation, and finally the modern reality in which Tesla ($1.5 trillion) is worth 3.9x the top three US utilities combined ($391 billion).

The Great Unbundling

Edison’s integrated model survived in various forms for over a century. But in the late 1990s, states like New York unbundled the three major components of the electric utility business: power generation, transmission, and distribution. Con Edison, which had been vertically integrated for over a century, was forced to separate. Independent power producers competed on generation. Regulated monopolies operated the transmission grid. Local utilities handled distribution. This completed a 120-year migration from Edison’s integrated model to a fully separated infrastructure stack.

The parallel to AI is striking. Today, NVIDIA makes the chips, hyperscalers run the data centres, labs train the models, and application companies build on top. The unbundling that took electricity a century is happening in AI within a decade. The question is not whether this separation will intensify — it is whether any company can resist it. Edison is NVIDIA, the company that controls the fundamental hardware. GE is OpenAI, the company that tried to straddle both infrastructure and application. The appliance manufacturers — Whirlpool, Frigidaire, RCA — are the emerging AI application companies. The same roles, the same economic logic, the same eventual outcome.

The Commoditisation Curve

The dual-axis chart above reveals a subtle but critical distinction. The cost per kilowatt-hour fell 68% in real terms — a meaningful decline, but not dramatic. The household budget share, however, collapsed 93% — from 29% to 2%. This asymmetry is the commoditisation signature. The resource itself does not need to become free; it simply needs to become cheap enough that consumers stop thinking about it. At 29% of household budgets, electricity was a considered purchase, a luxury rationed carefully. At 2%, it is a background cost, a line item no one examines. The same trajectory is playing out for AI inference. When GPT-3 cost $60 per million tokens, organisations deliberated carefully about each API call. At $0.28 per million tokens, inference becomes a utility input — cheap enough that the focus shifts entirely to what you build on top of it.

The GE Warning

GE’s trajectory is the most powerful cautionary tale in the electricity value migration. For over a century, GE straddled both infrastructure and application layers — the last Edison-model integrated company. At its peak in 2000, GE was the most valuable company in the world at approximately $594 billion, spanning power generation, appliances, industrial equipment, media, and financial services. By 2018, it had been removed from the Dow Jones Industrial Average after a 111-year streak. By 2020, it had crashed to $59 billion — a 90% decline from peak. In 2021, the company announced a three-way breakup into GE Aerospace, GE HealthCare, and GE Vernova.

The breakup proved the point: focus beat integration. GE Aerospace alone, as a pure-play jet engine company, trades at $362 billion — more valuable than the combined conglomerate was in 2020. The company that once embodied Edison’s integrated model could not sustain vertical integration as each layer developed independent competitive dynamics. This is the warning for AI companies attempting to control the full stack: vertical integration works during the build-out phase, but specialisation pressure eventually wins.

The Application Explosion

The appliance adoption S-curves above illustrate Phase 3 — the Application Explosion — in its clearest historical form. Once electricity became cheap and reliable, waves of application-layer innovation transformed daily life. Refrigerators went from 5% household penetration in 1925 to 96% by 1960. Television went from 0.5% in 1946 to 87% by 1960 — a 14-year adoption arc that would be unthinkable without commodity electricity. Air conditioning, washing machines, and microwaves each followed their own S-curves, each one building on the invisible infrastructure below.

Nobody cared about the electricity that powered their refrigerator. They cared about the refrigerator. The value accrued to the appliance makers — GE Appliances, Whirlpool, Samsung — not to the utilities that supplied the electrons. The global home appliance market today stands at $710–718 billion and is projected to reach $1.17 trillion by 2032. The AI parallel is forming: nobody will care about the model that powers their AI assistant. They will care about the assistant. The value will accrue to the application, not the resource.

The Rural Electrification Act of 1936 adds an important dimension to the story. When only 10.9% of American farms had electricity, the market alone was not going to close the gap — the economics of running power lines to remote rural areas did not justify private investment. The federal government stepped in, creating cooperatives that electrified rural America in under 20 years, taking farm electrification from 10% to 93% by 1953. This government intervention did not slow the value migration; it accelerated the application explosion by expanding the addressable market. As AI becomes essential infrastructure, expect similar dynamics: government programmes to ensure universal access will expand the application layer, not protect the infrastructure layer.

The Endgame: Essential but Invisible

The 1:55 ratio is the endgame of every infrastructure revolution. US utilities generate $491 billion in revenue with 5–10% margins. They power a $30.6 trillion economy. Virtually 100% of GDP depends on electricity, yet the utility sector captures only ~1.6% of it. The three largest utilities — NextEra Energy ($195 billion), Southern Company ($100 billion), Duke Energy ($96 billion) — have a combined market cap of $391.7 billion, less than half of Microsoft alone ($2.9 trillion). The infrastructure that powers the entire US economy is valued at a fraction of a single software company. There is no clearer illustration of where value migrates when infrastructure commoditises.

Tesla is the modern proof point. An electricity application company — it builds vehicles and energy storage systems powered by commodity electrons — is worth $1.5 trillion, or 3.9x the top three utilities combined. Tesla went from $2.8 billion at its 2010 IPO to $1.5 trillion in sixteen years, all built on infrastructure it never owned or operated. It does not generate electricity, transmit it, or distribute it. It consumes it — and captures orders of magnitude more value than the companies that produce it. The same logic applies to AI: the companies that consume commodity inference to deliver premium outcomes will capture multiples of the value generated by the companies that produce the inference.

The AI Connection: Electricity Demand Comes Full Circle

In an ironic twist, AI is now the largest new source of electricity demand. Data centres consumed 58 TWh in 2014 (1.8% of US electricity). By 2030, they are projected to consume 700 TWh (10%) — a 12x increase driven almost entirely by AI training and inference workloads. AI’s share of data centre electricity is growing from minimal in 2014 to an estimated 70% by 2030. The newest and most power-hungry application of electricity is the very technology whose value migration this chapter analyses. The infrastructure revolution that began in 1882 is now powering the infrastructure revolution that began in 2020 — a concrete manifestation of the “stacking effect” that compresses each successive revolution’s timeline.

What Comes Next

The electricity revolution provides the longest and most complete data set for understanding value migration. From Edison’s six integrated components to the modern reality where utilities capture 1.8% of the economy they enable, the pattern is unmistakable: infrastructure commoditises, applications capture the value, and the timeline compresses with each subsequent revolution. The electricity revolution took 80 years to complete its value inversion. The AI revolution is projected to complete the same transition in 5–10 years — a 12–24x compression factor. The next chapter examines the telecom revolution, where AT&T spent $133.5 billion trying to escape the “dumb pipe” fate and failed — a story with even more direct implications for today’s AI infrastructure providers.