Edison’s Lesson for the AI Age: Why the Capex Boom Isn’t a Bubble

There’s a scene in It’s a Wonderful Life where George Bailey sees his town as if he’d never existed, Bedford Falls transformed into the darker Pottersville. Now play that same thought experiment with Thomas Edison. Remove him from history. Not just the lightbulb, but everything that followed: no Pearl Street Station in 1882 powering Lower Manhattan. No integrated system of generation and distribution. No catalyst for the 70-year, $2 trillion buildout that wired America. Without Edison’s initial spark, imagine the cascade of absences: refrigerators vanish (no cold chain for vaccines), factories grind to a halt (no assembly lines), cities shrink (no elevators, no skyscrapers), modern medicine evaporates (no MRI machines). Keep going: no computers, no internet, no smartphones, no data centers. No AI. This isn’t the 1880s, it’s an alternate timeline where civilization took a profoundly different path because one inventor didn’t build the foundation. And here’s the kicker: in 1920, forty years after Edison lit up Pearl Street, almost none of those revolutionary products existed yet. The capex came first. The innovation came after. Steve Jobs was right, you can only connect the dots looking backward. Edison couldn’t see ChatGPT coming. He just knew energy was fundamental.

Today we’re witnessing a compressed, seven-year sprint to spend $7+ trillion on AI infrastructure, at least $5 trillion for data centers and $1.5 trillion to rebuild the power grid to feed them. Microsoft, Google, Amazon, and Meta spent $251 billion in 2024 alone, a 62% spike from 2023. Now we have OpenAI and its small revenues all over the news spending money they don’t have using future revenues. This has led to memes with the Three Stooges going viral. The question everyone’s asking: isn’t this a bubble? The answer requires understanding what happened when Edison and his successors built foundational infrastructure at this scale. The U.S. electrical buildout took seventy years (1882-1960) and cost $2 trillion in today’s dollars. Edison’s genius wasn’t just perfecting the bulb, it was conceiving an entire integrated system modeled on gas lighting networks, using low-voltage DC to power 5,000 lights within a one-mile radius of Pearl Street, illuminating Wall Street offices and the New York Times. But it was a grinding marathon after that initial spark: 45 competing utilities tearing up Chicago streets, the bitter “War of the Currents” between Edison’s DC and Westinghouse’s AC systems, and a stark reality where by 1930, nine in ten urban homes had power but only one in ten farms did. The market failed rural America, running lines cost $45,000 per mile in today’s dollars, requiring FDR’s Rural Electrification Administration to socialize the final buildout. But here’s what matters: throughout most of this seventy-year period, the revolutionary products didn’t exist yet. The first home refrigerator launched in 1913 but didn’t become common until after WWII. Television wasn’t viable until the late 1940s, forty years of infrastructure before the product that reshaped American culture. Computers emerged in the 1940s and didn’t reach consumers until the 1970s. The internet required 120+ years of electrical infrastructure. Economic historians have proven this lag: factories adopted electric motors rapidly in the early 1900s, but productivity didn’t accelerate until decades later when manufacturers completely re-engineered their production processes from centralized steam engines to distributed electric motors on each tool.

The AI boom follows an identical pattern, we’re building infrastructure before we know what it’s for, just as Edison built his grid before anyone imagined television. As Jensen Hunag said recently “This is the biggest AI infrastructure project in history. We’re literally going to connect intelligence to every application, to every use case, to every device and we’re just at the beginning.” Today’s applications (ChatGPT writing emails, AI analyzing spreadsheets, Midjourney generating images) are like Edison’s first lightbulbs: useful but not revolutionary. The revolution comes when AI disappears into the background and simply makes everything else work better, like electricity. What doesn’t exist yet? Personalized medicine where AI designs treatment protocols tailored to your exact genetic makeup after digesting every clinical trial ever conducted. Real-time supply chain optimization where millions of routing, inventory, and logistics decisions are calculated with near-perfect information from satellites, IoT sensors, and weather data. Automated scientific discovery where AI generates hypotheses, designs experiments, and identifies patterns faster than human researchers, compressing years to weeks. Hyper-personalized education where every student has an AI tutor adapting in real-time to their learning style. Predictive infrastructure maintenance where cities anticipate bridge failures and power outages before they happen. But here’s what I know for certain: I’m wrong about most of these specifics. The actual applications will surprise us, just as the internet surprised people who spent $1 trillion laying fiber in the 1990s expecting phone calls, not YouTube and Netflix. Edison thought he was lighting streets and homes. He couldn’t see he was building the engine for computation itself. The infrastructure gets built for one reason and used for another.

The current buildout has three components absorbing hundreds of billions: silicon (~60% of capex / $3.1 trillion) dominated by Nvidia’s GPUs; power and cooling (~25% / $1.3 trillion) because AI data centers consume electricity at densities far exceeding traditional facilities; and physical construction (~15% / $0.8 trillion) for the warehouse-scale buildings themselves. The most profound aspect? The AI buildout is forcing a second electrical revolution, Edison’s grid must be rebuilt to power the AI age. U.S. electricity demand was flat for a decade until AI shattered the equilibrium, data centers jumped from 4.4% of U.S. electricity in 2023 to a projected 12% by 2028. This forced utilities to revise their entire capital plans, with Deloitte projecting $1.4 trillion in new power sector investment by 2030 just to meet AI demand. The new grid requires rebuilding the old grid. This symbiotic feedback loop, data center capex necessitating power grid capex, pushes the total toward $7 trillion, one of the most concentrated capital deployments in human history happening in just seven years versus Edison’s seventy.

So is it a bubble? Let’s be precise: bubbles are characterized by speculation, leverage, momentum-driven prices detached from fundamentals, and assets producing no cash flow (tulips, flipped houses, revenue-free dot-coms). The AI buildout fits none of these patterns. The capital comes from the most profitable companies in history funding it from their own cash flows, Microsoft, Google, Amazon, and Meta generated $350+ billion in revenue in Q1 2025 alone. These assets produce measurable value today: Azure AI services generate billions, ChatGPT has millions of paying subscribers, enterprise customers deploy AI across operations. The infrastructure has inherent value, data centers can be repurposed, power plants will generate electricity for decades. There’s a crucial difference between infrastructure bubbles and application bubbles: in the 1990s, too much fiber was laid and many telecom companies collapsed, but the fiber didn’t disappear, it was acquired at pennies on the dollar and became the foundation for YouTube, Netflix, and cloud computing. The infrastructure was real; the business models were wrong. Crucially, the companies that laid the fiber weren’t the ones who reaped the rewards, that value went to entirely different players who built on top of it. Today’s AI buildout is different: the same companies spending the capex OpenAI, Microsoft, Google, Amazon, Meta are positioned to capture the value from the applications built on their infrastructure. We see the models they have released. They already see the models of the future that cannot be released without the infrastructure to support the world of users. Edison faced similar skepticism, critics called electric lighting a fad, claimed gas was superior, predicted financial ruin. Could there be a shakeout in AI? Of course! There will be many companies that end up worthless like within all VC fund investment portfolios. Will some companies overspend? Absolutely! But that’s not a bubble, it’s the natural tension in all foundational infrastructure investments where you must build before you know exactly what it’s for.

The legitimate concern is the lag between capex and productivity gains, the same lag Edison faced. He lit up Pearl Street in 1882, but it took decades for the economic transformation to materialize. We’re in the infrastructure phase now, spending $7 trillion to build the computational grid, but economy-wide productivity gains come later when businesses fundamentally re-engineer workflows, when new companies are built from the ground up around AI, when a generation grows up taking this technology for granted and uses it in ways we can’t anticipate. This creates real risk: if the lag is too long, if revenue growth doesn’t materialize fast enough, there could be a “capex hangover” where spending moderates and stock prices correct. This isn’t irrational worry, it’s the inherent vulnerability of infrastructure investment. But it’s also not a bubble. It’s patient capital betting that intelligence, like energy, is fundamental, and making intelligence abundant and cheap will unlock capabilities we can articulate and many more we cannot. History shows that every general-purpose technology that became infrastructure, electricity, telephony, the internet followed this pattern: capital first, applications later, dots connected backward. Edison couldn’t connect the dots forward to computers and AI. He just knew that abundant, cheap energy would change everything.

The final lesson from George Bailey’s Bedford Falls applies perfectly to Thomas Edison’s legacy: the most important infrastructure is invisible until it’s gone. Edison died in 1931, before television, before computers, before the internet, before any of the applications that would justify the decades of capital investment in his grid. He never saw the full consequences of what he built. But imagine removing him from history, no Pearl Street, no integrated electrical system, no catalyst for the 70-year buildout. You don’t just lose lightbulbs. You lose everything that electricity enabled, including the computers that train AI models, the data centers that run them, the cooling systems that keep them operational, the global telecommunications networks that distribute their output. Even AI would not be here without Edison. The AI infrastructure being built today will be equally invisible once complete, your children won’t think about data centers in Iowa or subsea cables connecting continents, they’ll just live in a world where certain things are possible that weren’t before. Medical diagnoses in minutes. Scientific breakthroughs cascading rapidly. Supply chains adapting in real-time. Education meeting each student where they are. Autonomous vehicles driving on the streets and humanoids walking on the sidewalks. Or, equally possible, applications we can’t imagine yet, built by people who take this infrastructure for granted. In 1882, Edison couldn’t see the refrigerator, television, computer, or AI coming. He could only know that energy was fundamental and making it ubiquitous would create unenumerable possibilities. In 2025, I can’t give you a spreadsheet proving AI’s ROI or tell you exactly which applications will matter most. But I can tell you that intelligence is fundamental—it’s not just another tool, it’s the foundation everything else gets built on. And every time we’ve built a new foundation like this, electricity, the internet, it unlocked possibilities we couldn’t have imagined beforehand. We’re spending $7 trillion to build something we don’t yet fully understand. That’s not a bubble. That’s a bet on the future that history suggests is worth making because Edison made the same bet 143 years ago, and without his infrastructure, none of this would be possible. The dots only connect backward, and they all trace back to a power station in Lower Manhattan that changed everything.