Question to ChatGPT: Is there an AI bubble? * (answer appears at the end of this column)
There has been increasing talk of an “AI bubble” in the media and among some investors. Enthusiasm and even irrational exuberance are hallmarks of early investment in disruptive technologies, and history teaches that there will be big winners and inglorious flameouts.
Like most technological revolutions, artificial intelligence requires abundant finding to bring the promise to fruition. What is different is the sheer scope of that investment, which is diverting investment in the rest of the economy. Bubble? Yes, but whether it pops, leaks or inflates is the question.
Spending on AI is clearly dominating all other factors driving of the economy. Harvard economist Jason Furman states that investment in information processing equipment and software for the first half of 2025 accounted for 92% of the growth in GDP, and that this spending is crowding out other critical investment in areas like residential construction and manufacturing.
JP Morgan reports that during 2024 and 2025, more than 100% of the growth in private fixed investment came from communications equipment, software and information processing, offsetting a slowdown in other areas like homebuilding. Somewhere between 50% and 100% of GDP growth in the first half of the year can be attributed to AI.
The massive firms at the forefront of the revolution were already minting money before ChatGPT burst upon the scene, thanks in part to their highly efficient capital structure. Most of the so-called Magnificent 7 benefited from a low capital intensity; that is, their businesses depend much less on physical assets like plant and equipment and more on intangibles like intellectual property and brand loyalty. Contrast General Motors, with 50 cavernous factories and a massive inventory, with the asset-light structure of Meta or Microsoft.
This asset-light capital structure has allowed tech behemoths to generate huge profits and free cash flows, and investors have been rewarded. The market value of the Magnificent 7 now accounts for nearly a third of the entire S&P 500, the highest concentration on record. But change is afoot.
The AI buildout requires massive investment in infrastructure that is swelling the balance sheets of these giants with tangible assets. This is especially true for the “hyperscalers” bankrolling the epic computing and storage capacity that are the mother’s milk of artificial intelligence. The big hyperscalers (Alphabet, Amazon, Meta, Microsoft and Oracle) spent a combined $67 billion on capital investment in 2019. JP Morgan expects that to balloon to $467 billion in 2027. McKinsey & Co. reckons a total commitment of $5 trillion for data centers alone will be required by 2030.
Increased capital intensity amplifies risk. History is replete with examples of overinvestment in new technologies that ultimately lead to excess capacity, price collapse and bankruptcies. The building of the railroads and the dotcom bubble are examples of transformational technological revolutions that left the field littered with early casualties.
Bain capital believes that to justify this monumental investment, data centers must generate $2 trillion in revenues by 2030. That is a 100 fold increase from the expected 2025 revenues of around $20 billion. Practically vertical.
Consider a representative project under construction in rural Indiana. Dubbed Project Ranier, the colossal assemblage of computing power being built by Amazon will consume 2.2 gigawatts of electricity. To paraphrase Everett Dirksen: a gigawatt here, a gigawatt there, and pretty soon you’re talking real juice — 2.2 gigawatts is enough to power all the homes in Nashville. Ten times over. McKinsey believes that total power consumption dedicated to AI will reach 80 gigawatts by 2030. Goldman Sachs says 122 gigawatts.
New transformational innovations often attract huge sums of capital during the early phase. Much or even most of that early capital fails to produce a return, and many early players go bust. That was true of the 19th century deployment of the iron horse, as hundreds of startup railroads went bankrupt, but ultimately the economy prospered tremendously and even today, rail is one of the most efficient modes of shipment. A century later, 80 million miles of fiber optic line were laid for the internet. By 2005, 85% of that fiber remained unused and dozens of telecom enterprises failed, but the cost of data transmission plummeted by 90%. Global Crossing is long forgotten but Netflix lives on.
Is it a bubble? Like an autopsy, the answer can only be known with certainty postmortem. Adjusted for inflation and shorter useful life, AI investment already exceeds the capital that built the railroads. Since the launch of ChatGPT in 2022, artificial intelligence plays have sucked up 90% of all capital investment, and AI-related stocks are responsible for 75% of the gains in the S&P 500.
The U.S. economy is now along for the ride. Consumer spending typically comprises 2/3 of GDP but for the first time ever, capital spending has eclipsed the consumer as the prime mover. Another first: Investment in data centers now exceeds construction of traditional office buildings.
Until recently, the hyperscalers have self-funded out of cash flows. However, financing is becoming increasingly byzantine and circular. Microsoft owns 27% of OpenAI, which has a $100 billion chip deal with Nvidia. Meanwhile OpenAI inked a deal to buy chips from AMD in exchange for 10% of AMD’s stock, and a $300 billion partnership with Oracle, which is borrowing so heavily that its debt ratio is five times the S&P 500 average. If one of the big boys stubs a toe, the contagion could trigger a major selloff.
It is probable that a significant portion of this capital tsunami will be lost. That is the nature of innovation cycles. It is also evident that artificial intelligence is changing the world. What is unclear and unknowable is who the winners will be.
* ChatGPT answers: “That’s one of the most debated economic questions right now — and the short answer is: yes, parts of the AI sector are showing bubble-like behavior, but the whole AI industry is not necessarily in a full bubble.”
Christopher A. Hopkins, CFA, is a co-founder of Apogee Wealth Partners in Chattanooga.