Why Fears of a Trillion-Dollar AI Bubble Are Growing

(Bloomberg) — For almost as long as the artificial intelligence boom has been in full swing, there have been warnings of a speculative bubble that could rival the dot-com craze of the late 1990s that ended in a spectacular crash and a wave of bankruptcies.

Tech firms are spending hundreds of billions of dollars on advanced chips and data centers, not just to keep pace with a surge in the use of chatbots such as ChatGPT, Gemini and Claude, but to make sure they’re ready to handle a more fundamental and disruptive shift of economic activity from humans to machines. The final bill may run into the trillions. The financing is coming from venture capital, debt and, lately, some more unconventional arrangements that have raised eyebrows on Wall Street.

Most Read from Bloomberg

Even some of AI’s biggest cheerleaders acknowledge the market is frothy, while still professing their belief in the technology’s long-term potential. AI, they say, is poised to reshape multiple industries, cure diseases and generally accelerate human progress.

Yet never before has so much money been spent so rapidly on a technology that, for all its potential, remains somewhat unproven as a profit-making business model. Tech industry executives who privately doubt the most effusive assessments of AI’s revolutionary potential — or at least struggle to see how to monetize it — may feel they have little choice but to keep pace with their rivals’ investments or risk being out-scaled and sidelined in the future AI marketplace.

What are the warning signs for AI?

When Sam Altman, the chief executive of ChatGPT maker OpenAI, announced a $500 billion AI infrastructure plan known as Stargate alongside other executives at the White House in January, the price tag triggered some disbelief. Since then, other tech rivals have ramped up spending, including Meta’s Mark Zuckerberg, who has pledged to invest hundreds of billions in data centers. Not to be outdone, Altman has since said he expects OpenAI to spend “trillions” on AI infrastructure.

To finance those projects, OpenAI is entering into new territory. In September, chipmaker Nvidia Corp. announced an agreement to invest up to $100 billion in OpenAI’s data center buildout, a deal that some analysts say raises questions about whether the chipmaker is trying to prop up its customers so that they keep spending on its own products.

The concerns have followed Nvidia, to varying degrees, for much of the boom. The dominant maker of AI accelerator chips has backed dozens of companies in recent years, including AI model makers and cloud computing providers. Some of them then use that capital to buy Nvidia’s expensive semiconductors. The OpenAI deal was far larger in scale.

OpenAI has also indicated it could pursue debt financing, rather than leaning on partners such as Microsoft Corp. and Oracle Corp. The difference is that those companies have rock-solid, established businesses that have been profitable for many years. OpenAI expects to burn through $115 billion of cash through 2029, The Information has reported.

Other large tech companies are also relying increasingly on debt to support their unprecedented spending. Meta, for example, turned to lenders to secure $26 billion in financing for a planned data center complex in Louisiana that it says will eventually approach the size of Manhattan. JPMorgan Chase & Co. and Mitsubishi UFJ Financial Group are also leading a loan of more than $22 billion to support Vantage Data Centers’ plan to build a massive data-center campus, Bloomberg News has reported.

So how about the payback?

By 2030, AI companies will need $2 trillion in combined annual revenue to fund the computing power needed to meet projected demand, Bain & Co. said in a report released in September. Yet their revenue is likely to fall $800 billion short of that mark, Bain predicted.

“The numbers that are being thrown around are so extreme that it’s really, really hard to understand them,” said David Einhorn, a prominent hedge fund manager and founder of Greenlight Capital. “I’m sure it’s not zero, but there’s a reasonable chance that a tremendous amount of capital destruction is going to come through this cycle.”

In a sign of the times, there’s also a growing number of less proven firms trying to capitalize on the data center goldrush. Nebius, an Amsterdam-based cloud provider that split off from Russian internet giant Yandex in 2024, recently inked an infrastructure deal with Microsoft worth up to $19.4 billion. And Nscale, a little-known British data center company, is working with Nvidia, OpenAI and Microsoft on build-outs in Europe. Like some other AI infrastructure providers, Nscale previously focused on another frothy sector: cryptocurrency mining.

“AI probably will have profound consequences on the way we all work. But Schumpeterian creative destruction being what it is, there will be some pain ahead before we all enjoy the businesses that are being built.”

— John Authers, Bloomberg Opinion. Click here for his full column.

Are there concerns about the technology itself?

The data center spending spree is overshadowed by persistent skepticism about the payoff from AI technology. In August, investors were rattled after researchers at the Massachusetts Institute of Technology found that 95% of organizations saw zero return on their investment in AI initiatives.

More recently, researchers at Harvard and Stanford offered a possible explanation for why. Employees are using AI to create “workslop,” which the researchers define as “AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.”

The promise of AI has long been that it would help streamline tasks and boost productivity, making it an invaluable asset for workers and one that corporations would pay top dollar for. Instead, the Harvard and Stanford researchers found the prevalence of workslop could cost larger organizations millions of dollars a year in lost productivity.

AI developers have also been confronting a different challenge. OpenAI, Claude chatbot developer Anthropic and others have for years bet on the so-called scaling laws — the idea that more computing power, data and larger models will inevitably pave the way for greater leaps in the power of AI. Eventually, they say, these advances will lead to artificial general intelligence, a hypothetical form of the technology so sophisticated that it matches or exceeds humans in most tasks.

Over the past year, however, these developers have experienced diminishing returns from their costly efforts to build more advanced AI. Some have also struggled to match their own hype. After months of touting GPT-5 as a significant leap, OpenAI’s release of its latest AI model in August was met with mixed reviews. In remarks around the launch, Altman conceded that “we’re still missing something quite important” to reach AGI.

Those concerns are compounded by growing competition from China, where companies are flooding the market with competitive, low-cost AI models. While US firms are generally still viewed as ahead in the race, the Chinese alternatives risk undercutting Silicon Valley on price in certain markets, making it harder to recoup the significant investment in AI infrastructure.

There’s also the risk that the AI industry’s vast data center buildout, entailing a huge increase in electricity consumption, will be held back by the realities of strained national power networks.

What does the AI industry say in response?

Sam Altman, the face of the current AI boom, has repeatedly acknowledged the risk of a bubble in recent months while maintaining his optimism for the technology. “Are we in a phase where investors as a whole are overexcited about AI? In my opinion, yes,” he said in August. “Is AI the most important thing to happen in a very long time? My opinion is also yes.”

Altman and other tech leaders continue to express confidence in the roadmap toward AGI, with some suggesting it could be closer than skeptics think. “Developing superintelligence is now in sight,” Zuckerberg wrote in July, referencing an even more powerful form of AI that his company is aiming for. In the near term, some AI developers also say they need to drastically ramp up computing capacity to support the rapid adoption of their services. Altman, in particular, has stressed repeatedly that OpenAI remains constrained in computing resources as hundreds of millions of people around the world use its services to converse with ChatGPT, write code and generate images and videos.

OpenAI and Anthropic have also released their own research and evaluations that indicate AI systems are having a meaningful impact on work tasks, in contrast to the more damning reports from outside academic institutions. An Anthropic report released in September found that roughly three quarters of companies are using Claude to automate work. The same month, OpenAI released a new evaluation system called GDPval that measures the performance of AI models across dozens of occupations.

“We found that today’s best frontier models are already approaching the quality of work produced by industry experts,” OpenAI said in a blog post. “Especially on the subset of tasks where models are particularly strong, we expect that giving a task to a model before trying it with a human would save time and money.”

So how much will customers eventually be willing to pay for these services? The hope among developers is that, as AI models improve and field more complex tasks on users’ behalf, they will be able to convince businesses and individuals to spend far more to access the technology.

“I want the door open to everything,” OpenAI Chief Financial Officer Sarah Friar said in late 2024, when asked about a report that the company has discussed a $2,000 monthly subscription for its AI products. “If it’s helping me move about the world with literally a Ph.D.-level assistant for anything that I’m doing, there are certainly cases where that would make all the sense in the world.”

In September, Zuckerberg said an AI bubble is “quite possible,” but stressed that his bigger concern is not spending enough to meet the opportunity. “If we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously,” he said in a podcast interview. “But what I’d say is I actually think the risk is higher on the other side.”

What makes a market bubble?

Bubbles are economic cycles defined by a swift increase in market values to levels that aren’t supported by the underlying fundamentals. They’re usually followed by a sharp selloff — the so-called pop.

A bubble often begins when investors get swept up in a speculative frenzy — over a new technology or other market opportunity — and pile in for fear of missing out on further gains. American economist Hyman Minsky identified five stages of a market bubble: displacement, boom, euphoria, profit-taking and panic.

Bubbles are sometimes difficult to spot because market prices can become dislocated from real-world values for many reasons, and a sharp price drop isn’t always inevitable. And, because a crash is part of a bubble cycle, they can be hard to pinpoint until after the fact.

Generally, bubbles pop when investors realize that the lofty expectations they had were too high. This usually follows a period of over-exuberance that tips into mania, when everyone is buying into the trend at the very top. What comes next is a usually a slow, prolonged selloff where company earnings start to suffer, or a singular event that changes the long-term view, sending investors dashing for the exits.

There was some fear that an AI bubble had already popped in late January, when China’s DeepSeek upended the market with the release of a competitive AI model purportedly built at a fraction of the amount that top US developers spend. DeepSeek’s viral success triggered a trillion-dollar selloff of technology shares. Nvidia, a bellwether AI stock, slumped 17% in one day.

The DeepSeek episode underscored the risks of investing heavily in AI. But Silicon Valley remained largely undeterred. In the months that followed, tech companies redoubled their costly AI spending plans, and investors resumed cheering on these bets. Nvidia shares charged back from an April low to fresh records. It was worth more than $4 trillion by the end of September, making it the most valuable company in the world.

So is this 1999 all over again?

As with today’s AI boom, the companies at the center of the dot-com frenzy drew in vast amounts of investor capital, often using questionable metrics such as website traffic rather than their actual ability to turn a profit. There were many flawed business models and exaggerated revenue projections. Telecommunication companies raced to build fiber-optic networks only to find the demand wasn’t there to pay for them. When it all crashed in 2001, many companies were liquidated, others absorbed by healthier rivals at knocked-down prices.

Echoes of the dot-com era can be found in AI’s massive infrastructure build-out, sky-high valuations and showy displays of wealth. Venture capital investors have been courting AI startups with private jets, box seats and big checks. Many AI startups tout their recurring revenue as a key metric for growth, but there are doubts as to how sustainable or predictable those projections are, particularly for younger businesses. Some AI firms are completing multiple mammoth fundraisings in a single year. Not all will necessarily flourish.

“I think there’s a lot of parallels to the internet bubble,” said Bret Taylor, OpenAI’s chairman and the CEO of Sierra, an AI startup valued at $10 billion. Like the dot-com era, a number of high-flying companies will almost certainly go bust. But in Taylor’s telling, there will also be large businesses that emerge and thrive over the long term, just as happened with Amazon.com Inc. and Alphabet Inc.’s Google in the late 90s.

“It is both true that AI will transform the economy, and I think it will, like the internet, create huge amounts of economic value in the future,” Taylor said. “I think we’re also in a bubble, and a lot of people will lose a lot of money.”

There are also some key differences that market watchers point out, the first being the broad health and stability of the biggest businesses that are at the forefront of the trend. Most of the “Magnificent Seven” group of US tech companies are long-established giants that make up much of the earnings growth in the S&P 500 Index. These firms have huge revenue streams and are sitting on large stockpiles of cash.

Despite the skepticism, AI adoption has also proceeded at a rapid clip. OpenAI’s ChatGPT has about 700 million weekly users, making it one of the fastest growing consumer products in history. Top AI developers, including OpenAI and Anthropic, have also seen remarkably strong sales growth. OpenAI previously forecast revenue would more than triple in 2025 to $12.7 billion. While the company does not expect to be cash-flow positive until near the end of this decade, a recent deal to help employees sell shares gave it an implied valuation of $500 billion — making it the world’s most valuable company never to have turned a profit.

–With assistance from Neil Campling.

Most Read from Bloomberg Businessweek

©2025 Bloomberg L.P.

(Bloomberg) — For almost as long as the artificial intelligence boom has been in full swing, there have been warnings of a speculative bubble that could rival the dot-com craze of the late 1990s that ended in a spectacular crash and a wave of bankruptcies.

Tech firms are spending hundreds of billions of dollars on advanced chips and data centers, not just to keep pace with a surge in the use of chatbots such as ChatGPT, Gemini and Claude, but to make sure they’re ready to handle a more fundamental and disruptive shift of economic activity from humans to machines. The final bill may run into the trillions. The financing is coming from venture capital, debt and, lately, some more unconventional arrangements that have raised eyebrows on Wall Street.

Most Read from Bloomberg

Even some of AI’s biggest cheerleaders acknowledge the market is frothy, while still professing their belief in the technology’s long-term potential. AI, they say, is poised to reshape multiple industries, cure diseases and generally accelerate human progress.

Yet never before has so much money been spent so rapidly on a technology that, for all its potential, remains somewhat unproven as a profit-making business model. Tech industry executives who privately doubt the most effusive assessments of AI’s revolutionary potential — or at least struggle to see how to monetize it — may feel they have little choice but to keep pace with their rivals’ investments or risk being out-scaled and sidelined in the future AI marketplace.

What are the warning signs for AI?

When Sam Altman, the chief executive of ChatGPT maker OpenAI, announced a $500 billion AI infrastructure plan known as Stargate alongside other executives at the White House in January, the price tag triggered some disbelief. Since then, other tech rivals have ramped up spending, including Meta’s Mark Zuckerberg, who has pledged to invest hundreds of billions in data centers. Not to be outdone, Altman has since said he expects OpenAI to spend “trillions” on AI infrastructure.

To finance those projects, OpenAI is entering into new territory. In September, chipmaker Nvidia Corp. announced an agreement to invest up to $100 billion in OpenAI’s data center buildout, a deal that some analysts say raises questions about whether the chipmaker is trying to prop up its customers so that they keep spending on its own products.

The concerns have followed Nvidia, to varying degrees, for much of the boom. The dominant maker of AI accelerator chips has backed dozens of companies in recent years, including AI model makers and cloud computing providers. Some of them then use that capital to buy Nvidia’s expensive semiconductors. The OpenAI deal was far larger in scale.

OpenAI has also indicated it could pursue debt financing, rather than leaning on partners such as Microsoft Corp. and Oracle Corp. The difference is that those companies have rock-solid, established businesses that have been profitable for many years. OpenAI expects to burn through $115 billion of cash through 2029, The Information has reported.

Other large tech companies are also relying increasingly on debt to support their unprecedented spending. Meta, for example, turned to lenders to secure $26 billion in financing for a planned data center complex in Louisiana that it says will eventually approach the size of Manhattan. JPMorgan Chase & Co. and Mitsubishi UFJ Financial Group are also leading a loan of more than $22 billion to support Vantage Data Centers’ plan to build a massive data-center campus, Bloomberg News has reported.

So how about the payback?

By 2030, AI companies will need $2 trillion in combined annual revenue to fund the computing power needed to meet projected demand, Bain & Co. said in a report released in September. Yet their revenue is likely to fall $800 billion short of that mark, Bain predicted.

“The numbers that are being thrown around are so extreme that it’s really, really hard to understand them,” said David Einhorn, a prominent hedge fund manager and founder of Greenlight Capital. “I’m sure it’s not zero, but there’s a reasonable chance that a tremendous amount of capital destruction is going to come through this cycle.”

In a sign of the times, there’s also a growing number of less proven firms trying to capitalize on the data center goldrush. Nebius, an Amsterdam-based cloud provider that split off from Russian internet giant Yandex in 2024, recently inked an infrastructure deal with Microsoft worth up to $19.4 billion. And Nscale, a little-known British data center company, is working with Nvidia, OpenAI and Microsoft on build-outs in Europe. Like some other AI infrastructure providers, Nscale previously focused on another frothy sector: cryptocurrency mining.

“AI probably will have profound consequences on the way we all work. But Schumpeterian creative destruction being what it is, there will be some pain ahead before we all enjoy the businesses that are being built.”

— John Authers, Bloomberg Opinion. Click here for his full column.

Are there concerns about the technology itself?

The data center spending spree is overshadowed by persistent skepticism about the payoff from AI technology. In August, investors were rattled after researchers at the Massachusetts Institute of Technology found that 95% of organizations saw zero return on their investment in AI initiatives.

More recently, researchers at Harvard and Stanford offered a possible explanation for why. Employees are using AI to create “workslop,” which the researchers define as “AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.”

The promise of AI has long been that it would help streamline tasks and boost productivity, making it an invaluable asset for workers and one that corporations would pay top dollar for. Instead, the Harvard and Stanford researchers found the prevalence of workslop could cost larger organizations millions of dollars a year in lost productivity.

AI developers have also been confronting a different challenge. OpenAI, Claude chatbot developer Anthropic and others have for years bet on the so-called scaling laws — the idea that more computing power, data and larger models will inevitably pave the way for greater leaps in the power of AI. Eventually, they say, these advances will lead to artificial general intelligence, a hypothetical form of the technology so sophisticated that it matches or exceeds humans in most tasks.

Over the past year, however, these developers have experienced diminishing returns from their costly efforts to build more advanced AI. Some have also struggled to match their own hype. After months of touting GPT-5 as a significant leap, OpenAI’s release of its latest AI model in August was met with mixed reviews. In remarks around the launch, Altman conceded that “we’re still missing something quite important” to reach AGI.

Those concerns are compounded by growing competition from China, where companies are flooding the market with competitive, low-cost AI models. While US firms are generally still viewed as ahead in the race, the Chinese alternatives risk undercutting Silicon Valley on price in certain markets, making it harder to recoup the significant investment in AI infrastructure.

There’s also the risk that the AI industry’s vast data center buildout, entailing a huge increase in electricity consumption, will be held back by the realities of strained national power networks.

What does the AI industry say in response?

Sam Altman, the face of the current AI boom, has repeatedly acknowledged the risk of a bubble in recent months while maintaining his optimism for the technology. “Are we in a phase where investors as a whole are overexcited about AI? In my opinion, yes,” he said in August. “Is AI the most important thing to happen in a very long time? My opinion is also yes.”

Altman and other tech leaders continue to express confidence in the roadmap toward AGI, with some suggesting it could be closer than skeptics think. “Developing superintelligence is now in sight,” Zuckerberg wrote in July, referencing an even more powerful form of AI that his company is aiming for. In the near term, some AI developers also say they need to drastically ramp up computing capacity to support the rapid adoption of their services. Altman, in particular, has stressed repeatedly that OpenAI remains constrained in computing resources as hundreds of millions of people around the world use its services to converse with ChatGPT, write code and generate images and videos.

OpenAI and Anthropic have also released their own research and evaluations that indicate AI systems are having a meaningful impact on work tasks, in contrast to the more damning reports from outside academic institutions. An Anthropic report released in September found that roughly three quarters of companies are using Claude to automate work. The same month, OpenAI released a new evaluation system called GDPval that measures the performance of AI models across dozens of occupations.

“We found that today’s best frontier models are already approaching the quality of work produced by industry experts,” OpenAI said in a blog post. “Especially on the subset of tasks where models are particularly strong, we expect that giving a task to a model before trying it with a human would save time and money.”

So how much will customers eventually be willing to pay for these services? The hope among developers is that, as AI models improve and field more complex tasks on users’ behalf, they will be able to convince businesses and individuals to spend far more to access the technology.

“I want the door open to everything,” OpenAI Chief Financial Officer Sarah Friar said in late 2024, when asked about a report that the company has discussed a $2,000 monthly subscription for its AI products. “If it’s helping me move about the world with literally a Ph.D.-level assistant for anything that I’m doing, there are certainly cases where that would make all the sense in the world.”

In September, Zuckerberg said an AI bubble is “quite possible,” but stressed that his bigger concern is not spending enough to meet the opportunity. “If we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously,” he said in a podcast interview. “But what I’d say is I actually think the risk is higher on the other side.”

What makes a market bubble?

Bubbles are economic cycles defined by a swift increase in market values to levels that aren’t supported by the underlying fundamentals. They’re usually followed by a sharp selloff — the so-called pop.

A bubble often begins when investors get swept up in a speculative frenzy — over a new technology or other market opportunity — and pile in for fear of missing out on further gains. American economist Hyman Minsky identified five stages of a market bubble: displacement, boom, euphoria, profit-taking and panic.

Bubbles are sometimes difficult to spot because market prices can become dislocated from real-world values for many reasons, and a sharp price drop isn’t always inevitable. And, because a crash is part of a bubble cycle, they can be hard to pinpoint until after the fact.

Generally, bubbles pop when investors realize that the lofty expectations they had were too high. This usually follows a period of over-exuberance that tips into mania, when everyone is buying into the trend at the very top. What comes next is a usually a slow, prolonged selloff where company earnings start to suffer, or a singular event that changes the long-term view, sending investors dashing for the exits.

There was some fear that an AI bubble had already popped in late January, when China’s DeepSeek upended the market with the release of a competitive AI model purportedly built at a fraction of the amount that top US developers spend. DeepSeek’s viral success triggered a trillion-dollar selloff of technology shares. Nvidia, a bellwether AI stock, slumped 17% in one day.

The DeepSeek episode underscored the risks of investing heavily in AI. But Silicon Valley remained largely undeterred. In the months that followed, tech companies redoubled their costly AI spending plans, and investors resumed cheering on these bets. Nvidia shares charged back from an April low to fresh records. It was worth more than $4 trillion by the end of September, making it the most valuable company in the world.

So is this 1999 all over again?

As with today’s AI boom, the companies at the center of the dot-com frenzy drew in vast amounts of investor capital, often using questionable metrics such as website traffic rather than their actual ability to turn a profit. There were many flawed business models and exaggerated revenue projections. Telecommunication companies raced to build fiber-optic networks only to find the demand wasn’t there to pay for them. When it all crashed in 2001, many companies were liquidated, others absorbed by healthier rivals at knocked-down prices.

Echoes of the dot-com era can be found in AI’s massive infrastructure build-out, sky-high valuations and showy displays of wealth. Venture capital investors have been courting AI startups with private jets, box seats and big checks. Many AI startups tout their recurring revenue as a key metric for growth, but there are doubts as to how sustainable or predictable those projections are, particularly for younger businesses. Some AI firms are completing multiple mammoth fundraisings in a single year. Not all will necessarily flourish.

“I think there’s a lot of parallels to the internet bubble,” said Bret Taylor, OpenAI’s chairman and the CEO of Sierra, an AI startup valued at $10 billion. Like the dot-com era, a number of high-flying companies will almost certainly go bust. But in Taylor’s telling, there will also be large businesses that emerge and thrive over the long term, just as happened with Amazon.com Inc. and Alphabet Inc.’s Google in the late 90s.

“It is both true that AI will transform the economy, and I think it will, like the internet, create huge amounts of economic value in the future,” Taylor said. “I think we’re also in a bubble, and a lot of people will lose a lot of money.”

There are also some key differences that market watchers point out, the first being the broad health and stability of the biggest businesses that are at the forefront of the trend. Most of the “Magnificent Seven” group of US tech companies are long-established giants that make up much of the earnings growth in the S&P 500 Index. These firms have huge revenue streams and are sitting on large stockpiles of cash.

Despite the skepticism, AI adoption has also proceeded at a rapid clip. OpenAI’s ChatGPT has about 700 million weekly users, making it one of the fastest growing consumer products in history. Top AI developers, including OpenAI and Anthropic, have also seen remarkably strong sales growth. OpenAI previously forecast revenue would more than triple in 2025 to $12.7 billion. While the company does not expect to be cash-flow positive until near the end of this decade, a recent deal to help employees sell shares gave it an implied valuation of $500 billion — making it the world’s most valuable company never to have turned a profit.

–With assistance from Neil Campling.

Most Read from Bloomberg Businessweek

©2025 Bloomberg L.P.