Six hundred billion dollars of faith

This week, Alphabet, Microsoft, Meta, and Amazon report quarterly earnings on the same day. Investors want one answer: is the biggest capital spending spree in corporate history actually paying off? The four companies are on track to pour roughly $600 billion into AI infrastructure this year, a figure that would have seemed absurd even twelve months ago. Bridgewater Associates pegged the number at $650 billion in February. Some analysts now say $680 billion when you include Apple, Nvidia, and Oracle. Whatever the precise tally, it dwarfs anything the technology industry has attempted before, a 67% jump from 2025's already staggering $381 billion. And yet, by almost every measure of actual return, the evidence remains thin. MIT found that 95% of AI pilots deliver zero measurable impact on the bottom line. S&P Global reported that 42% of companies abandoned most of their AI projects in 2025. Morgan Stanley found that only 21% of S&P 500 companies could cite a single measurable AI benefit. The gap between spending and proof is the central tension of this cycle. So why does the spending keep accelerating?

Nobody can afford to blink

The simplest explanation is game theory. Davidson Kempner's CIO called it a "prisoner's dilemma" that few can escape. Bradford Cornell, writing in February, argued that when every competitor is forced to make the same massive investment, the primary beneficiaries are consumers and smaller companies, not the investors footing the bill. The logic is brutal in its clarity. If Google cuts its AI budget and Meta doesn't, Google risks falling behind in cloud, search, and advertising, the businesses that generate virtually all of its revenue. If Meta slows down while Microsoft accelerates, Meta loses ground in the enterprise AI market it desperately wants to enter. Nobody can afford to be the one who stopped spending. The rational move for each individual company is to keep building, even if the collective outcome is that nobody earns an adequate return. This is how you get Meta raising its 2026 capex guidance to $145 billion, nearly double what it spent in 2025, and watching its stock drop 6% after hours. The market punishes the spending. The market would punish stopping even more. Reuters described these companies as being "stuck in a wheel" of ever-rising AI spending. That framing is more accurate than most. This isn't a strategy. It's an arms race masquerading as one.

We have seen this movie before

The closest historical parallels are the railroad boom of the 1840s through 1880s and the fiber-optic buildout of the late 1990s. Both followed the same arc: transformative technology, massive capital deployed on faith, a painful bust, and then, decades later, vindication of the underlying bet. During the dot-com era, telecom companies laid over 80 million miles of fiber-optic cable across the United States. The overcapacity was staggering. Four years after the bubble burst, 85% of those fiber lines were still dark. Bandwidth costs collapsed by 90%. Many of the companies that built the infrastructure went bankrupt. But that dark fiber became the foundation on which YouTube, Netflix, and Facebook were eventually built. The builders didn't profit. The next generation did. The railroad story follows a similar pattern. By 1855, Britain had the highest density of railroad tracks in the world, with 90% of those tracks built during the speculative bubble years. The railways transformed commerce, slashed transportation costs, and employed half a million people. They also destroyed most of the investors who funded them. AI data center spending is now tracking at 1.2% of U.S. GDP, exceeding the telecom buildout of the early 2000s (1.0%) and trailing only the railroad boom of the 1880s (6.0%). The pattern is consistent: the infrastructure matters, but the companies that build it rarely capture the value they create.

A bet backed by faith, not revenue

What makes the current moment particularly striking is the acceleration. The numbers haven't just grown, they've nearly tripled in two years. And the growth is shifting from digital to physical in ways that introduce entirely new constraints. Bridgewater's Greg Jensen warned in February that the AI boom has entered a "more dangerous phase," marked by exponentially rising investments in physical infrastructure and growing reliance on outside capital. The early phase of the buildout was funded by Big Tech's own free cash flow. That era is ending. Analysts project that free cash flow across these companies could drop by up to 90% in 2026 as capex outpaces revenue growth. The Magnificent Seven are now raising debt to fund AI build-outs, a significant shift from the self-funded model of previous years. CreditSights estimates roughly 75% of the total spend, about $450 billion, goes directly to AI infrastructure: GPUs, servers, networking equipment, and data centers. These are physical assets with long depreciation schedules. Unlike software investments that can be scaled back quickly, a half-built data center is a sunk cost. Once you've broken ground, the financial logic pushes you to finish. The energy dimension adds another layer of pressure. S&P Global warned that Big Tech's $635 billion spending faces an energy shock test as data center power demands surge. U.S. data center power consumption may exceed 100 gigawatts by 2035, equivalent to adding a mid-sized European country's worth of electricity demand. Each megawatt of AI compute requires tens of tons of copper, straining supply chains that were already tight. None of this is backed by proportional current revenue. It's a bet on future demand that doesn't exist yet at the scale needed to justify the investment. That's not analysis. It's faith.

The earnings call paradox

There's a peculiar dissonance at the heart of this cycle. Every quarter, CEOs take the stage and call AI "the most important technology of our lifetime." Every quarter, investors push back on the spending. And every quarter, the spending goes up anyway. Alphabet's cloud revenue growth hit 48% in Q4 2025, with AI as a meaningful contributor. Microsoft reported Azure growth of 33%, with AI accounting for 16 percentage points. These are real numbers, but they're not remotely proportional to the capex being deployed. Google Cloud generated $17.7 billion in Q4 2025 revenue. Google is spending multiples of that on infrastructure in a single quarter. The companies point to these growth rates as validation. Investors point to the widening gap between revenue and spending as a warning. Both are right, which is exactly what makes this moment so unstable. Earlier this year, these four companies hemorrhaged over $1 trillion in combined market value after revealing their 2026 spending plans. The market recovered. The spending didn't slow down. That pattern, shock followed by acceptance followed by even bigger numbers, is the defining rhythm of this era.

Cheap inference is coming regardless

Here's the part that should keep startup founders optimistic and Big Tech CFOs awake at night: the Jevons paradox. In 1865, William Stanley Jevons observed that more efficient coal engines didn't reduce coal consumption. They increased it. Efficiency made coal cheaper per unit of work, which made coal-powered applications economical in places they hadn't been before. Demand exploded. The same dynamic is playing out with AI inference. Per-token costs have dropped roughly 1,000x in three years. Enterprise AI spending surged 320% in 2025. Both facts are true simultaneously. Cheaper inference doesn't reduce total spending on AI. It expands the universe of viable applications, which drives more total compute demand, which justifies more infrastructure, which makes inference cheaper still. Satya Nadella invoked Jevons paradox explicitly after DeepSeek demonstrated dramatically cheaper inference in early 2025. His message to investors was clear: don't worry about cheaper models eating our margins, because cheaper models will expand the total market. He may be right about the market expanding. The question is whether the companies spending $600 billion are the ones who will capture that expanded market, or whether they're building the equivalent of dark fiber for a generation of startups that hasn't been founded yet. History suggests the latter. The railroad companies built the tracks. The retailers, manufacturers, and logistics companies that used those tracks captured the real value. The fiber companies laid the cables. Google, Netflix, and Meta built businesses on top of them. The infrastructure always matters. The builders rarely win.

The faith economy

There is no precedent for this level of coordinated capital expenditure in the history of the technology industry. There is no precedent for this degree of spending with this little proof of proportional return. And there is no precedent for the particular trap these companies find themselves in: unable to stop, unable to prove the investment is working, and unable to even slow down without signaling weakness to competitors, investors, and customers simultaneously. The $600 billion is not an investment thesis. It's an act of collective faith, a bet that the demand curve will eventually catch up to the supply curve, that the applications will materialize, that the revenue will follow the infrastructure. It might be right. The technology is real. The potential applications are vast. But the same was true of railroads in 1845 and fiber in 1999. The spending won't stop this year. It probably won't stop next year. Because in a prisoner's dilemma, the only move more terrifying than continuing to spend is being the first one to stop.