Since the artificial intelligence (AI) boom began in earnest, markets have repeatedly sounded warnings: the current speculative bubble could rival the late 1990s internet frenzy, which ultimately ended in spectacular crashes and waves of bankruptcies.
Tech companies are spending hundreds of billions of dollars on advanced chips and data center construction—not just to keep pace with surging usage of chatbots like ChatGPT, Gemini, and Claude, but to prepare for a more fundamental and disruptive shift in economic activity from humans to machines. The final bill could reach trillions of dollars, funded through venture capital, debt, and some recent unconventional arrangements that have caught Wall Street's attention.
Even AI's biggest supporters acknowledge the market feels "bubbly," yet they remain convinced of the technology's long-term potential. They believe AI could reshape multiple industries, cure diseases, and accelerate human progress overall.
However, never before has such massive capital been deployed toward a single technology in such a short timeframe. Despite AI's enormous potential, as a consistently profitable business model, it remains largely unproven. Tech industry executives who privately harbor doubts about the revolutionary AI narrative—or at least struggle to see how to monetize it—may feel they have no choice but to match competitors' investment pace, or risk being marginalized in the future AI market.
**What Are the Warning Signs?**
When OpenAI CEO Sam Altman announced the $500 billion "Stargate" AI infrastructure plan at the White House in January alongside other executives, the price tag left some incredulous. Since then, other tech giants have accelerated their spending. Meta's Mark Zuckerberg has committed to investing hundreds of billions in data centers. Not to be outdone, Altman subsequently stated he expects OpenAI to spend "trillions" on AI infrastructure.
To fund these projects, OpenAI is venturing into new territory. In September, chip manufacturer NVIDIA announced an agreement to invest up to $100 billion in OpenAI for data center construction. Some analysts question whether NVIDIA is propping up its own customers so they can continue purchasing its products.
The first Stargate AI data center is under construction in Abilene, Texas.
Such concerns have followed NVIDIA throughout this boom cycle to varying degrees. NVIDIA dominates AI accelerator production and has invested in dozens of companies in recent years, including AI model developers and cloud service providers, some of which use that funding to purchase NVIDIA's high-priced chips. The OpenAI deal is simply much larger in scale.
OpenAI has also indicated it may seek debt financing rather than relying on partners like Microsoft and Oracle, both of which have highly stable, profitable businesses spanning many years. According to reports, OpenAI expects to burn through $115 billion in cash by 2029.
Other major tech companies are increasingly relying on debt to support unprecedented spending. Meta is seeking $26 billion in bank financing for a planned data center complex in Louisiana that the company says will ultimately approach the size of Manhattan. Morgan JPMorgan and Mitsubishi UFJ Financial Group reportedly led a loan exceeding $22 billion for a Vantage Data Centers mega data center campus.
**What Returns on Massive Investment?**
Bain & Company reported in September that by 2030, AI companies would need to collectively generate $2 trillion in annual revenue to fund the computing power required to meet projected demand. However, Bain predicts their revenues will fall $800 billion short of that level.
"The numbers being thrown around are so exaggerated they're really very difficult to understand," said David Einhorn, prominent hedge fund manager and founder of Greenlight Capital. "I'm sure the returns won't be zero, but it's quite possible this cycle will see massive capital destruction."
To keep pace with the times, increasingly unproven companies are trying to grab a piece of the data center "gold rush." Amsterdam-based cloud services provider Nebius, spun off from Russian internet giant Yandex in 2024, recently signed a $19.4 billion infrastructure agreement with Microsoft. Another little-known British data center company, Nscale, is working with NVIDIA, OpenAI, and Microsoft to build European data centers. Like some other AI infrastructure providers, Nscale's previous business focus was another notably bubble-prone area: cryptocurrency mining.
"AI will likely profoundly change how we work. But in terms of Schumpeterian 'creative destruction,' we must inevitably experience some growing pains before enjoying the fruits of these emerging enterprises."
**Is the Technology Itself Concerning?**
Behind this data center investment frenzy, skepticism about AI technology returns persists. In August, an MIT study unsettled investors: 95% of organizations reported zero returns from their AI projects.
Recently, Harvard and Stanford researchers offered a possible explanation: employees are using AI to create so-called "workslop"—defined by researchers as "AI-generated content that appears well-done but lacks substance and fails to advance tasks."
The long-standing promise of AI has been that it would streamline processes, boost productivity, and become an employee "superpower" that companies would pay premium prices for. Instead, Harvard and Stanford research found that the proliferation of "workslop" could cost large institutions millions of dollars in productivity annually.
AI developers also face another challenge. OpenAI, Anthropic (Claude chatbot developer), and others have long bet on so-called "scaling laws"—that more computing power, more data, and larger models will inevitably yield breakthrough improvements in AI capabilities. They envision such progress eventually producing artificial general intelligence (AGI), a hypothetical technology that could match or exceed human performance in most tasks.
Over the past year, these developers have increasingly experienced diminishing returns while building more advanced AI at enormous cost. Some organizations have failed to meet their own promotional heights. After months of claiming GPT-5 would be a major leap, OpenAI's August release of its latest model still received mixed reviews. Around the launch, Altman acknowledged that to reach AGI, "we're still missing something very key."
Meanwhile, external competition is intensifying as companies flood the market with competitive, low-cost models.
Another risk: the AI industry's massive data center construction comes with dramatically increased power consumption, while national grid capacity is already strained, potentially slowing expansion progress.
**How Does the AI Industry Respond?**
As the current AI boom's "spokesperson," Altman has repeatedly acknowledged bubble risks in recent months while remaining optimistic about technology prospects. "Overall, are investors currently overexcited about AI? In my view, yes," he said in August. "Is AI the biggest thing for quite some time? My answer is also yes."
Altman and other tech leaders remain confident about the path to AGI, with some believing it can be achieved faster than skeptics imagine. "Today, superintelligence is within sight," Zuckerberg wrote in July, referring to a more powerful form of AI his company hopes to achieve. In the near term, some AI developers say they need dramatically increased computing power to support rapid service adoption. Altman has particularly emphasized that as hundreds of millions globally use OpenAI's service to chat with ChatGPT, write code, and generate images and videos, OpenAI remains constrained by computing capacity.
OpenAI and Anthropic have also published their own research and evaluations showing AI systems are having substantive impact on work tasks, contrasting with more stringent reports from external academic institutions. Anthropic reported in September that about three-quarters of companies are using Claude for automation. That same month, OpenAI launched GDPval, a new evaluation system measuring model performance across dozens of professions.
"We found that current strongest frontier models are approaching the work quality of industry experts," OpenAI wrote in a blog post. "Especially on task subsets where models excel, we expect assigning tasks to models rather than humans first will save time and reduce costs."
So how much will customers ultimately pay for these services? Developers hope that as AI models improve and can handle more complex tasks for users, they can convince businesses and individuals to pay higher fees for the technology.
When asked about reports that OpenAI had discussed charging "$2,000 monthly subscription fees" for its AI products, OpenAI CFO Sarah Friar said in late 2024: "I don't want to rule out any possibilities. If it can help me handle anything in the real world like a 'PhD-level assistant,' that certainly makes complete sense in some scenarios."
In September, Zuckerberg said an AI bubble was "quite possible," but emphasized his greater concern was not spending enough to seize opportunities. "If we end up spending several hundred billion dollars wrong, that's certainly very unfortunate," he said on a podcast. "But I'd say I actually think the risk on the other side of the scale is greater."
**What Constitutes a Bubble?**
A bubble refers to an economic cycle where market values rapidly climb to levels mismatched with fundamentals, often followed by sharp selloffs—so-called bubble "bursting."
Bubbles typically begin when investors become swept up in speculative fervor over new technology or opportunities, piling in due to "fear of missing out" on further gains. American economist Hyman Minsky divided market bubbles into five stages: displacement, boom, euphoria, profit-taking, and panic.
Bubbles can be difficult to identify because market prices may disconnect from "true value" for various reasons, and sharp price drops don't always occur. Moreover, since "crashes" are part of bubble cycles, bubbles often only become clearly identifiable in hindsight.
Generally, bubbles burst when investors realize previous high expectations were overly optimistic. This typically happens after a period of excessive exuberance that turns "euphoric," with everyone chasing highs. What often follows is a slow, sustained correction period where company earnings begin suffering, or a single event changes long-term market outlook, causing investors to collectively "rush for the exits."
In late January this year, some feared the AI bubble had already burst when DeepSeek launched a competitive AI model at reportedly far lower cost than leading US developers, shocking markets. DeepSeek's explosion triggered trillion-dollar tech stock selloffs, with AI bellwether NVIDIA plunging 17% in a single day.
The DeepSeek incident highlighted risks of heavy AI investment. But Silicon Valley remained essentially unshaken. In subsequent months, tech companies doubled down on high-cost AI spending plans, with investors again cheering these bets. NVIDIA stock rebounded from April lows to new records. By late September, its market value exceeded $4 trillion, making it the world's most valuable company.
**Will 1999 History Repeat?**
Similar to the current AI boom, companies at the center of the internet bubble attracted massive capital, often relying on questionable metrics (like website traffic) rather than actual profitability. There were plenty of flawed business models and inflated revenue projections. Telecom companies raced to lay fiber networks only to discover demand insufficient to cover costs. When the 2001 crash came, many companies were liquidated while others were acquired at discount prices by relatively better-performing rivals.
Current phenomena echoing the internet bubble include AI's massive infrastructure construction, sky-high valuations, and conspicuous wealth displays. Venture capitalists are attracting AI startups with private jets, luxury boxes, and large checks. Many AI startups call recurring revenue a key growth metric, but the sustainability and predictability of these expectations face questions, particularly for younger companies. Some AI companies complete multiple massive funding rounds within a year, and not all will thrive.
"I think this has many similarities to the internet bubble," said Bret Taylor, OpenAI chairman and CEO of $10 billion-valued AI startup Sierra. As then, many companies currently "flying high" will almost certainly crash. But Taylor also believes large enterprises will ultimately emerge and prosper long-term, like Amazon and Alphabet's Google did in the late 1990s.
"AI will change the economy—that's true, and I think it will create enormous economic value in the future like the internet did," Taylor said. "I also think we're in a bubble, and many people will lose a lot of money."
Amazon founder Jeff Bezos said current AI-related investments resemble an "industrial bubble," somewhat like the 1990s biotech bubble, but he still expects AI to boost "productivity at every company in the world."
However, market observers also point out key differences compared to the internet bubble period. First is the overall health and stability of the largest companies at the forefront. Most of the "Magnificent Seven" tech stocks are battle-tested giants contributing significantly to S&P 500 earnings growth. These companies have massive revenues and substantial cash reserves.
Despite persistent skepticism, AI adoption is advancing rapidly. OpenAI's ChatGPT has about 700 million weekly users, making it one of history's fastest-growing consumer products. Leading AI developers including OpenAI and Anthropic have shown exceptionally strong sales growth. OpenAI predicted its 2025 revenue would more than triple to $12.7 billion. Although the company expects positive cash flow only near decade's end, a recent transaction helping employees sell shares gave it an implied $500 billion valuation, making it the world's most valuable company that has never turned a profit.