How AI Conquered the US Economy: A Visual FAQ
Artificial intelligence is the biggest tech construction project in a century. Here's what that looks like.
The American economy has split in two. There’s a rip-roaring AI economy. And there’s a lackluster consumer economy.
You see it in the economic statistics. Last quarter, spending on artificial intelligence outpaced the growth in consumer spending. Without AI, US economic growth would be meager.
You see it in stocks. In the last two years, about 60 percent of the stock market’s growth has come from AI-related companies, such as Microsoft, Nvidia, and Meta. Without the AI boom, stock market returns would be putrid.
You see it in the business data. According to Stripe, firms that self-describe as “AI companies” are dominating revenue growth on the platform, and they’re far surpassing the growth rate of any other group.
Nobody can say for sure whether the AI boom is evidence of the next Industrial Revolution or the next big bubble. All we know is that it’s happening. We can all stop talking about “what will happen if AI dominates the economy at such-and-such future date?” No, the AI economy is here and now. We’re living in it, for better or worse.
So, what exactly is the artificial intelligence boom? How did it happen, where did all this money to build AI come from, who is using the technology, and is it making people more productive? Today, in a bit of a throwback to my early blogging years, I’m going to try to walk through an FAQ with graphs to create a visual guide to the question:
How big is the AI boom?
Artificial intelligence has a few simple ingredients: computer chips, racks of servers in data centers, huge amounts of electricity, and networking and cooling systems that keep everything running without overheating.
This hardware is immensely expensive. In the last six months, the four companies investing the most in artificial intelligence—Meta, Google, Microsoft, and Amazon—spent between $100 billion and $200 billion on chips, data centers, and the like. “The most valuable tech companies are buying and building stuff at a record pace,” wrote the Wall Street Journal’s Christopher Mims.
Can you put those numbers in historical context?
This is either the biggest tech-infrastructure project since the 1960s (since the beginning of the computer age) or the 1880s (the heyday of the railroad age).
In January, JP Morgan’s Michael Cembalest calculated that the leading AI chip manufacturer Nvidia is on pace to capture the highest share of market-wide capital spending since IBM’s peak revenues in 1969. Not to be outdone, the economic writer Paul Kedrosky has calculated that AI capital expenditures as a share of GDP have already exceeded the dot-com boom and are now approaching levels not seen since the railroad build-out of the Gilded Age.
Where is all this money coming from?
Today’s AI infrastructure boom is made possible by the extraordinary and unprecedented profits of today’s leading tech companies. As Cembalest explained on my podcast, today's leading tech companies have become so profitable in the last few years that their share of total “free cash flow” (meaning, revenue minus operating expenses and infrastructure) dwarfs anything we’ve seen since the end of World War II. These firms’ existing business models—whether it’s ads for Meta or search ads for Google—are strong enough to generate stupid amounts of money to throw at the next generation of technology. “They’re generating unprecedented amounts of free cash flow,” Cembalest told me. “They make oodles and oodles of money, which is why they can afford to be pouring hundreds of billions of dollars of capital spending each year into AI-related R&D and infrastructure.”
So, does this explain why the stock market is behaving so weirdly?
I think so. There’s an interesting debate in finance circles now about why the stock market seems to be shrugging off the Trump tariffs and slowing growth. I think the clearest answer to this question is some combination of (a) some investors still think Trump will chicken out on the tariffs; (b) they don’t think the final effect of the tariffs will be very big; and (c) the tariffs don’t matter much to the digital economy, and AI-related stocks are dominating returns while the rest of the market collectively putters along.
As this chart from Societe Generale shows, the ten largest companies in the S&P 500 have so dominated net income growth in the last six years that it’s becoming more useful to think about an S&P 10 vs an S&P 490. If you’re a portfolio manager invested in the other 490 stocks, the last six years of equity returns aren’t very impressive because these companies have collectively not managed to increase their profits.
Okay, so companies are spending unprecedented amounts of money. Are they earning it back?
Not yet. As the Wall Street Journal's Greg Ip wrote, the "unsettling” side of the AI boom is that all this spending on chips and data centers is "draining American corporations of cash." OpenAI and Anthropic are losing gobs of money, and the biggest tech companies are still relying on their older business models to generate their largest profit margins. If these firms are spending much more than they’ll ever be able to earn back, it would suggest that we’re in the midst of a historic infrastructure bubble.
As for the bull case: The payments company Stripe is already seeing evidence that AI startup revenue is exceeding the growth rate of any previous generation of technology. “AI companies are reaching revenue milestones faster than previous generations of startups,” the company announced in a recent report. “The top 100 AI companies on Stripe achieved annualized revenues of $1 million in a median period of just 11.5 months—four months ahead of the fastest-growing SaaS companies.”
Who is using this technology?
By one account, generative AI tools like ChatGPT and Gemini are being adopted faster than practically any technology for which we have good data. The St. Louis Federal Reserve has estimated that the rate of adoption for generative AI is roughly twice as fast as the Internet.
In one the largest recent surveys of generative AI—the 2025 paper "The Labor Market Effects of Generative Artificial Intelligence”—economists estimated that more than 50 percent of workers in information services (meaning, software firms) and management are already using the technology at work. That compares with very few people in old-economy firms, such as mining or fishing. AI is also much more popular among college graduates than people who never attended college.
Do workers say AI is making them more productive?
Yes. The iconic study proving that new AI models improve productivity comes from firms with rather repetitive work, such as call centers. But we’re getting more self-reports from workers saying that AI is helping them save lots of time. One surprising example: teaching. According to a recent Gallup survey, roughly 60 percent of elementary school teachers say they’ve used AI to prepare lessons, review instruction material, make worksheets, or do administrative work. Most teachers who use AI say it improves their work, and those who use it regularly say it saves them 6 hours a week—or six weeks per school year. In one very optimistic interpretation, that’s like saying AI gives elementary school teachers a month and a half of paid leave every year.
Perhaps the most bullish indicator that AI is going to help people become more productive at work comes from the AI research nonprofit METR, which found that the length of tasks that AI agents can complete is doubling every 7 months. In 2021, AI could automate a simple Google search: 10 seconds. Two years later, ChatGPT was looking up facts on the Internet that would take the typical person about 4 minutes. Now some models are performing coding tasks that take a typical developer 50 minutes. “Extrapolating this trend predicts that, in under a decade, we will see AI agents that can independently complete a large fraction of software tasks that currently take humans days or weeks,” the researchers said.
So, that’s it then: AI will create an economy-wide productivity boom?
Not so fast. In fact, many workers might dramatically overestimate how much more productive AI is making them.
METR also conducted an in-depth study that asked experienced developers to code with a popular AI assistant. After they finished their tasks, the developers claimed that using the AI had made them 20 percent more productive. But independent evaluators in the study actually concluded that using AI did the opposite: it increased task completion time by about 20 percent. I don’t want to speculate too much about what this study means in the long run. But for now, I think it’s a necessary caveat to boosterish claims that ChatGPT is on the cusp of replacing tens of millions of entry-level white-collar jobs.
Tell me one more funny thing about how AI is changing the world
A new paper in Science found that since the rise of large language models, there's been a huge shift in academic writing. In 2024, the word "delves" has appeared 2,700% more than its historical average, by one account. The analysis suggests that about 1/7th of 2024 abstracts were processed by AI.
The delves into a significant and crucial topic, nice post Derek.
Derek, your analyses of important issues in business, politics, health, and other sectors are better than AI - because of the independent research, critical thinking, and deft summarizing behind them. In just a few minutes of reading, I have a much better understanding of what is going on with AI now, and how we should assess it. I look forward to future articles on this topic, and others in your Substack.
Personally, I have found that AI saves much time in my searches and is good at generating lists of ideas, but comes up short in writing. For example, I helped someone review their wedding speech, which was largely written by AI. The weakest parts were the ones written by AI; they ultimately got edited out. They were easy to spot: empty phrases that sounded good but signified nothing.