Let me start with a number that should make every CIO, CFO, and technology strategist stop what they’re doing: 42. In Douglas Adams’ cult classic The Hitchhiker’s Guide to the Galaxy, a supercomputer named Deep Thought spends 7.5 million years calculating the ultimate answer. The result? 42. The catch is that no one actually knows what the question was, leading to the creation of Earth as a giant organic computer to find it. But no, I am not referring to the Deep Thought answer. I am also not referring to the number of months until your cloud contract renewal. It’s the number of AI-related stocks that have generated 65-75% of ALL S&P 500 returns, earnings, and capital spending since ChatGPT launched in November 2022.
JPMorgan’s “Smothering Heights” Report
JPMorgan’s latest “Smothering Heights” report (https://privatebank.jpmorgan.com/nam/en/insights/latest-and-featured/eotm/outlook) —and yes, I’m quoted in the ecosystem analysis (page 17) —drops a truth bomb that Wall Street is pricing in but enterprise technology leaders are dangerously ignoring: we’re witnessing the most extreme market concentration in modern financial history, built on infrastructure spending that dwarfs the Manhattan Project, Apollo Program, and Interstate Highway System *combined*.
And here’s the part that should terrify any traditional industry player: if your enterprise AI strategy looks anything like what these 42 companies are doing, you’ve already lost.
The $18 Trillion Moat No Regular Enterprise Can Cross
Eight companies—just eight—have grown from $3 trillion to $18 trillion in market capitalization in seven years. Microsoft, Google, Amazon, Meta (the hyperscalers) plus NVIDIA, TSMC, ASML, and AMD (the semiconductor ecosystem). Together, they represent:
– 20% of developed world equity markets
– 16% of global markets including emerging markets
– 40-45% of US GDP growth in recent quarters (up from <5% in early 2023)
Let me put that GDP contribution in context. Tech sector capital spending alone is now contributing more to US economic growth than the entire manufacturing sector, all of agriculture, and most of construction *combined*. We’ve created an economy where AI infrastructure spending IS the economy.
JPMorgan’s analysis is brutal in its clarity: “without these 42 AI stocks, the S&P 500 would have underperformed Europe, Japan, and China since ChatGPT’s launch. The market isn’t being driven by broad economic growth or productivity gains—it’s being driven by a capital expenditure arms race that most companies can’t even comprehend, let alone participate in.”
AI Spend – The Capex Juggernaut That Dwarfs History
Here’s where it gets genuinely wild. Since Q4 2022, the four hyperscalers have spent $1.3 trillion on capital expenditures and R&D. Not over decades. Not over a generation. In roughly three years.
To understand how unprecedented this is, JPMorgan compared current tech capex to every major US infrastructure project in history:
“`
Peak Annual Spending as % of GDP:
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Manhattan Project (1944): 0.4%
Interstate Highway (1966): 0.6%
Apollo Project (1964): 0.8%
Tech Capex (2025): 2.1%
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
“`
Current tech spending is 2.6x the Apollo Program at its peak. It’s 3.5x the Interstate Highway System. It’s five times the Manhattan Project that built the atomic bomb and won World War II.
But here’s the critical difference that every enterprise architect needs to internalize: the Manhattan Project was designed to build three bombs to end a war. The Interstate Highway System was designed to move goods and people for 50+ years. Tech capex in 2025 is designed to… we’re not actually sure yet.
That uncertainty is the entire game.
Meta’s 70% Revenue Gamble (And What It Means For You)
Meta’s capex and R&D spending hit 70% of revenue in recent quarters. Read that again. For every dollar Meta makes, it’s spending 70 cents on AI infrastructure and research, up from typical tech company ratios of 10-15%.
The median S&P 500 company spends 10% of revenue on capex and R&D combined. Meta is spending 7x that ratio with no clear line of sight to when it generates returns.
Mark Zuckerberg, Q3 earnings call: “We will aggressively ramp up spending to stay competitive in the AI arms race.”
Translation: *We have no idea what the winning move is, but we know that not spending means certain death.
This is the game theory nightmare that makes enterprise AI so treacherous. The hyperscalers are in a prisoner’s dilemma where:
– Cooperate (moderate spending) = everyone loses to whoever defects
– Defect (massive spending) = potential winner-take-all outcome
– Not playing = certain irrelevance
For enterprise technology leaders, here’s the uncomfortable question: If Meta with $134 billion in revenue can’t afford to play this game sustainably, what makes you think your organization can?
The Infrastructure Layer vs. The Application Layer: A Wealth Transfer Story
JPMorgan breaks down AI stock performance into three categories since ChatGPT’s launch:
Direct AI stocks (28 companies including hyperscalers and semiconductors):
– Price return: 195%
– Earnings growth: 159%
– Capex/R&D growth: 72%
AI Utilities (8 power companies like Vistra, Constellation, NextEra):
– Price return: 66%
– Earnings growth: 64%
AI Capital Equipment (6 companies like Eaton, Trane, GE Vernova):
– Price return: 174%
– Earnings growth: 155%
Everyone else in the S&P 500:
– Price return: 26%
– Earnings growth: 19%
Notice what’s missing from the winner’s circle? AI application companies. AI-enabled productivity companies. AI software companies selling to enterprises.
Goldman Sachs tried tracking these. They built baskets of companies that should benefit from:
1. AI-enabled revenues (software companies selling AI products)
1. AI productivity beneficiaries (companies with high labor costs that AI should reduce)
Both baskets have been flat to the equal-weighted S&P 500 since late 2023. The infrastructure layer is capturing all the value. The application layer—where most enterprises are trying to play—is capturing approximately none of it.
This is the brutal economics of platform shifts: the first trillion dollars of value goes to infrastructure builders, not infrastructure users.
The Tariff Asymmetry That Reveals Everything
Here’s a detail from the JPMorgan report that’s easy to miss but reveals the entire geopolitical game: the Trump administration’s tariff structure.
Semiconductor ecosystem tariff exemptions:
– Semiconductors: 80% of imports exempt
– Computers: 75% exempt
– Computer parts: 70% exempt
Power generation ecosystem tariff exemptions:
– Solar panels: 20% exempt
– Batteries: 25% exempt
– Transformers: 15% exempt
– Electrical switches: 10% exempt
The US government is explicitly protecting semiconductor supply chains while exposing power infrastructure to tariffs. This isn’t accidental. It’s a clear signal about where strategic value lies.
The moat isn’t just technical—it’s geopolitical. The US has essentially declared that chip design and manufacturing are national security assets while power generation is a commodity.
Conclusion
For enterprises trying to build AI strategies, this asymmetry matters enormously. You can’t compete with companies whose supply chains are treated as strategic infrastructure. You can only figure out how to leverage what they build. The next blog discusses what enterprises can do in this era of exploding AI investments
Featured image designed by Freepik