Update 903 — Debate Over AI Bubble:
Could Boom in Valuations End in Tears?
AI has been all over the news recently, especially in the business section. All of the Magnificent 7 companies, the largest publicly traded companies in America, have some level of investment in AI. The sector has pushed tech and, in fact, all major stock indices to record levels. The term “AI washing” has been coined for companies that boost their stock values by exaggerating how much AI plays a role in their products and services.
Yet with all of this buzz, many industry analysts express concern about the AI boom, as some now feel that predictions of AI being “revolutionary” are overblown. Some have gone as far as to raise concerns that the AI industry is a bubble waiting to burst. This week, 20/20 Vision explores the arguments for and against the AI boom being an economic bubble, in the hopes that the readers can use this to assess AI’s place in the broader economy.
Best,
Dana
Introduction to the AI Boom
Since the launch of ChatGPT in November 2022, AI has quickly become not just a booming sector in the economy, but essentially the central sector underpinning continued economic growth. As Harvard economics professor Jason Furman calculated, while investment in information processing equipment and software accounted for 4 percent of GDP in the first half of 2025, it was also responsible for 92 percent of GDP growth.
Source: Jason Furman
The reason AI has sparked such a flurry of investment is simple: with the advent of generative AI (genAI), which can produce new content like text and images, the world has reached another turning point in technological innovation. While much of this innovation remains speculative, artificial intelligence has advanced significantly to accomplish simple tasks faster than humans can, such as writing code, analyzing data, editing images and videos, and more.
Many of the tasks once considered to be grunt work for entry-level employees can now be performed by genAI products. This development has resulted in greater AI adoption in the workplace, with 27 percent of white-collar employees reporting that they use AI frequently this year, up 12 percentage points since 2024. One can imagine that AI will accomplish far more as the technology develops, and investors want to buy stocks on the ground floor, before AI’s potential has been realized. If AI lives up to the expectations of its promoters, it could be the biggest technological breakthrough since the development of the steam engine.
America’s large tech firms are predicted to spend nearly $400 billion on AI infrastructure this year. By the end of 2028, the amount of money spent on data centers worldwide is expected to exceed $3 trillion. CEOs now need to consider how their firms will handle AI adoption in the near future, across every field, and those firms at the center of the AI boom reign supreme over the stock market. There is no better example of this than the computer chip company Nvidia. A decade ago, it was a niche manufacturer of graphics cards primarily for PC gamers. Since 2015, its stock value has increased 35,000 percent, and its market cap is now the world’s largest at $4.5 trillion.
The course that AI development takes will have massive implications for the US economy and beyond. With all of that said, it begs an important question that can no longer be brushed aside: are investors overenthusiastic about AI? Are we, in fact, seeing the formation of an AI bubble? And, perhaps most importantly, when will the answer to these questions become clear?
A Brief Overview of Economic Bubbles
Financial bubbles occur when the assets of a particular company, sector, market, etc., are priced by investors significantly higher than their actual value. The most famous recent bubbles in the US economy were the Dot Com Bubble of the late 90s/early 2000s – where overspeculation in the internet led to the overvaluation of internet stocks – and the real estate bubble, which led to the 2007 financial crisis and the Great Recession. In both cases, the bubbles burst, leading to financial crises, especially in the latter case.
One problem with identifying bubbles is that, generally, they can only be seen clearly as bubbles with the benefit of hindsight. Per many economists, bubbles follow a general lifecycle:
- Displacement, where investors become interested in a “new paradigm” that they think will make a profit, such as a new technology.
- Boom, where the new paradigm gains more attention and thus attracts more investors who do not want to miss out.
- Euphoria, the point where investors begin to completely ignore the risks and red flags, sending the valuation of the assets in question into the stratosphere.
- Profit-taking, where the smart money realizes what is going on and begins to pull their investments from these assets.
- Panic, where the investors that were once euphoric now try to flee the bubble en masse, sending asset values into a tailspin, with anyone left holding the bag losing a great deal of money.
Since the Dot Com Bubble has become the go-to comparison for when a bubble occurs around a new, promising technology, we will use it as an illustration of the lifecycle of a bubble:
- First, a new technology, the internet, served as a new paradigm for investors.
- Second, a boom occurred when people became more aware of the internet as the original internet services, such as AOL, grew, and the World Wide Web took off during the 90s.
- Third, euphoria kicked in as investors increasingly turned to start-ups that had not even turned a profit yet. Some of these companies were focused solely on establishing their brand, and some spent almost all of their budgets on advertising.
- Fourth, many of the biggest tech companies, such as Dell and Cisco, realized that the market had likely peaked in March 2000, after the merger of AOL with TimeWarner, and they began selling huge amounts of their stocks to take their profit while they still could.
- Fifth, panic set in, and mass-selling ensued. By October 4, 2002, the Nasdaq had fallen 77 percent from its peak.
As the Dot Com Bubble illustrated, even when investor enthusiasm is based on a real and profitable innovation, such as the internet, people’s eagerness for easy money can cloud their assessment of the risks. This can lead to financial panic when it becomes clear that their expectations will not be met.
Strongest Arguments that the AI Boom Is a Bubble
There are many troubling signs of an AI sector bubble. The following are some of what we feel are the strongest arguments that the AI boom is a bubble.
Investor Enthusiasm Exceeds Asset Worth
If the most telltale sign of a bubble is that investors’ expectations have grown beyond the likely payoff, there are plenty of signs that this is occurring with AI. Last month, OpenAI unveiled its plans for $1 trillion in AI infrastructure. This push is backed by several other large AI companies, most notably the software giant Oracle, which has pledged $300 billion to this venture in a five-year contract — a massive investment in infrastructure for one company.
But investors do not seem worried about whether or not it will be profitable. According to one company executive interviewed by the Wall Street Journal, they think that this is actually not enough. The executive expects the full demand for AI infrastructure, measured by total gigawatts used, to eventually reach 100 gigawatts, which would require $5 trillion in infrastructure investment. All of this from a company that has not even submitted its IPO.
Investment in AI is growing rapidly, but at best, it will be a few years before many AI companies see a profit. A major study from MIT published in June found that 95 percent of firms polled that had invested in GenAI had achieved a net zero return on investment so far. To be clear, AI revenue is growing, and it should come as no surprise that profits from AI will not be immediate. The infrastructure needed to support it takes time to develop, even with massive investments. Investors may be getting ahead of themselves in ways that practically beg for comparisons to the Dot Com Bubble. For example, former OpenAI executive Mra Murati recently raised $2 billion at a $12 billion valuation for his AI start-up, Thinking Machines Lab. The company is still in the early stages and does not yet have a product to show for all of its investments.
None of this will be a problem, assuming that artificial intelligence accomplishes everything that AI companies promise and becomes truly revolutionary. GenAI is still in the early stages of its development, and nobody knows for sure where it will end up. Apple recently rattled the AI industry by publishing a study that found that the large language models (LLMs) used by many companies to underpin their GenAIs often fell short of their benchmarks for accomplishing medium- to high-complexity tasks, a finding that suggests that AI is still very limited in its ability to compete with human reason and skill. Research suggests that AI is an avenue for boosting workplace productivity, but achieving that productivity is not yet as straightforward as simply replacing low-skilled workers with an AI model.
The AI industry is not simply promising to replace grunt work. AI CEOs, such as Anthropic’s Dario Amodei, are making bold statements about how AI could wipe out half of all entry-level white-collar jobs within the next five years. Firms in sectors ranging from traditional finance to healthcare are banking heavily on the assumption that AI will be truly transformative. Thus, if AI fails to live up to all of this hype, the damage could reach far beyond just the tech sector.
Debt Financing Supporting the AI Boom
Much of the investment has been made by leveraging debt. While it is difficult to get adequate data on private credit, much of the billions of dollars spent on AI infrastructure so far was borrowed from Big Tech companies. For example, regarding the deal between OpenAI and Oracle, Oracle is estimated to need to borrow $25 billion annually for the next four years to meet its obligations. Oracle is already a highly leveraged company, with its debt-to-equity ratio at about 450 percent.
The debt-to-equity ratio is less concerning for other Big Tech companies, such as Microsoft (~33 percent) and Alphabet (~11.5 percent), but the amount of private credit deployed to finance the AI boom is expected to grow. According to Bloomberg, tech companies have raised about $157 billion in 2025 alone, up 70 percent from last year, largely fueled by private credit. OpenAI, in particular, continues to operate at a loss of about $5 billion a year on $3.7 billion in revenue as of last year. Until that gap is closed, it will need to solicit more money from investors or leverage more debt from banks to continue operating.
This debt financing is not expected to slow down: according to private equity firm Carlyle, $1.8 trillion of capital will be deployed by 2030 to meet AI demand. Much of it will need to be financed by leveraging debt, with companies expected to leverage a staggering $1.5 trillion in the construction of data centers by 2028. If too many firms default on their debt, it could seriously destabilize the financial sector and set off a chain reaction throughout the economy. This wouldn’t be such a big concern if the financial payoff from AI were a certainty, but as discussed earlier, it isn’t.
Circular Investment, Corporate Entanglement
The next issue raised by those arguing that this is a bubble is that many large companies are investing trillions of dollars in an intricate web centered primarily on just two companies: OpenAI and Nvidia.
Source: Bloomberg
As illustrated by the chart, many of the deals between these companies are “circular” in nature. An example of a circular deal is when company A invests substantial sums in company B, and then company B uses that exact same money to buy goods and services from company A. For example, back in September, Nvidia signed a deal to invest as much as $100 billion in OpenAI to fuel its infrastructure buildup, and OpenAI, in turn, agreed to spend billions buying Nvidia chips for its planned data centers.
Corporate entanglement can lead to firms artificially inflating their share prices. The layers of interconnection themselves are also a problem: they make the success of the firms involved so interdependent that trouble with one company will spell trouble for the whole web of companies. During the Dot Com Bubble, for example, circular deals were made to finance leverage for funding the buildout of fiber optic cables, and vendors marked the sales of equipment as revenue even though this infrastructure was not yet in use. The current state of affairs is not entirely comparable to the Dot Com Bubble, as the transactions in AI’s case are taking place above board, but it does show why many market experts are worried about history repeating itself.
AI Companies and Questionable Accounting
Finally, there are already signs that AI companies are engaging in significant financial smoke-and-mirrors to make their balance sheets look better than they actually are. In an interview with Derek Thompson, investor and author Paul Kedrosky laid out how, in their earnings statements, many tech firms are spreading out the cost of AI infrastructure, like chips and data centers, over five-year periods.
The problem with this is that, with the current rapid rate of technological development, most analysts think that this equipment will need to be replaced every two or three years. In other words, they are obscuring how quickly much of their own infrastructure investments are depreciating in value, and thus making their balance sheets look healthier than they actually are. This is only the most prominent example laid out by Kedrosky on how AI companies seem intent on covering up the actual costs of their AI infrastructure, costs that are already quite high.
Arguments that the AI Boom Is NOT a Bubble
Even with all of this said, there are many reasons to believe that the talks of an AI bubble are overblown, even if they are not entirely baseless.
Investors Are Not Investing in Nothing
The biggest argument that the AI boom is not a bubble is that investors are already seeing real returns on their investments. As investor and writer Azeem Azhar covers in his own blog covering this exact topic, while AI revenue has not yet reached a level that meets investors’ loftiest expectations, it is real and growing at a rapid pace. According to Azhar’s numbers, GenAI has skyrocketed from $7.1 billion in annual revenue in 2023 to approximately $61.6 billion this year, a massive growth rate that shows no signs of deceleration. This specific measurement does not include revenue from investors, but from the consumption of GenAI itself. In other words, AI is definitely generating revenue, not just selling empty promises to investors. Keep in mind, also, that this is just GenAI, which does not cover the full range of AI’s current uses in today’s economy. It is difficult to know the number for the full amount of annual revenue produced by AI, but it is certainly even higher than $61.6 billion.
Also, the biggest movers and shakers in the AI industry are some of the most well-established and profitable blue-chip companies in the world. To look at the net income of many of these companies as per their most recent quarterly earnings reports:
- Microsoft had a net income of $101.8 billion.
- Nvidia had a net income of $16.6 billion.
- Alphabet had a net income of $28 billion.
Compare and contrast with the Dot Com Bubble Burst, when huge investments were made into companies that wound up producing nothing but advertising for themselves. While investors are likely to invest in a few startups that go under without ever producing anything of value, they are mostly investing in massive companies that provide quite a bit to consumers beyond just AI models. The big question mark here is OpenAI, which, per its own reporting, will not be profitable for at least a few years. Investors are willing to wait for a return on their investment, knowing it takes a few years for a venture to complete its buildup period. Investors tend to be more tolerant when they see the company they’ve invested in making progress towards profitability. With that said, OpenAI already produces a product that millions of people use daily, which automatically makes it a better investment than hundreds of failed startups during the Dot Com Bubble.
Even if these companies are profitable, there can still be valid concerns that their shares are overpriced, as investors let their expectations of future returns grow beyond what is realistic. The best way to address these concerns is by looking at the forward price-to-earnings (P/E) ratio of these companies, which measures the current share price (P) to the future expectations of earnings (E), and comparing it to past bubbles.
Goldman Sachs already did just that in a report released back in April. It found that the aggregate forward P/E for a 24-month period for the Magnificent 7— all of which have significant exposure to the AI boom — stood at 23x. This is to say that the forward P/E for these companies is pretty high, especially since this is not the most precise look at AI companies specifically. That said, 23x is far lower than the forward P/E reached by the leaders of the Dot Com Bubble, which peaked at 52x before it burst. In other words, it is possible that a bubble is forming, but it has not reached the level seen back in 2000, right before the Dot Com Bubble burst.
AI Could End Up Being Revolutionary
On the central question of whether AI will become as much of a disruptor as investors think, AI will have to achieve a high level of disruption to meet expectations. While a lot of things will have to go right for this to happen, experts such as Goldman Sachs economist Joseph Briggs feel that that is, in fact, what we are seeing. AI is seeing greater adoption in the workplace, as we covered previously, and the productivity gains seen from AI will grow as the technology itself is developed.
As Derek Thomson discussed in his piece on the matter, one must also consider the holy grail in AI technology: artificial general intelligence. Basically, artificial general intelligence is the point at which AI achieves genius-level intelligence in a way that exceeds even the smartest human. If that comes to pass, almost everything AI advocates have promised, even the loftiest and most far-reaching, becomes possible. Then, all arguments that the AI boom is a bubble will become moot, as there would be no doubt left that AI was worth the investment. This is not guaranteed — or possibly even likely — to happen by any stretch of the imagination, but it is within the realm of possibility and thus worth considering.
If the Bubble Bursts, Will it Be Catastrophic?
In short, even if AI is a bubble, it is not guaranteed that it will result in a catastrophic burst in the same vein as the Dot Com Bubble or the real estate bubble that led to the Great Recession. Soft landings are possible and easier to achieve when there is more substance to the underlying investment, as is the case with AI. If a bubble is identified, policymakers can establish guardrails for accessing greater credit, thus reducing the amount of debt leveraged to fuel further investment. If a bubble is identified early enough, it can also give investors the advanced warning they need to reduce their exposure to the assets that have inflated values without leading to the kind of mass sellings and death spirals associated with the worst bubbles. The Federal Reserve can also raise interest rates, which would slow down economic growth but, at the same time, reduce access to capital at a key point in a bubble’s development. As it stands, the Fed is not blind to the possibility that AI is becoming a bubble, and earlier this month, San Francisco Fed President Mary Daly stated outright that she didn’t think such a bubble would threaten broader financial stability.
In sum, we have two takeaways. One, AI is probably going to have a major impact on the workplace, even if that impact is not immediate and its full extent is still up for debate. Two, investors are very excited about AI’s potential, and it will take a lot of things going right for their expectations to be met. It may not become apparent for a few years as to whether or not the AI boom is a bubble, as things will become clearer once much of the AI infrastructure has been built. Until that happens, we encourage everyone to keep a wary eye on the AI industry and to take its promises with a grain of salt, but also not to panic at the slightest sign of trouble.