The AI bubble: are we on the brink of a burst?

Key points on the AI bubble

• Private markets vs. public companies: the AI bubble is more pronounced in private markets, where venture capital fuels high valuations for startups like OpenAI, often based on hype rather than proven revenue. Large public companies like Google and Meta, by contrast, have diversified revenue streams, so a burst would hurt them less.
• Expert warnings: Martin Eftimoski, a former Reserve Bank of Australia economist, publicly withdrew all AI-related investments from his pension. He argues that business leaders don’t understand AI’s practical usefulness and that expectations are overinflated.
• Physical barriers to growth: shortages in data center capacity and energy supplies are constraining AI’s expansion. Global energy demand for AI could reach 200 TWh in 2025, which would push costs up and slow innovation.

Overview of the AI hype cycle

The AI boom has driven massive investments, but recent reports show a disconnect between hype and real-world returns. Studies show that while adoption is high, measurable benefits remain elusive for most organizations. Market analysts warn of corrections, yet the technology’s long-term potential persists.

Signs of overvaluation

Investors have poured billions into AI, but skepticism is growing. Private markets show speculative bubbles, while public tech giants maintain stability through broad business models.

Venture capitalists are pouring billions into AI startups, and tech giants are racing to build ever more advanced models. Beneath that frenzy is a question that won’t go away: is the AI boom a bubble on the verge of bursting? That is not idle speculation. A growing body of evidence suggests the hype around AI is far outpacing its real-world value.

Enthusiasm falters as ROI fails to materialize

A clear sign of a potential AI bubble is the growing disillusionment with the return on investment (ROI) of AI technologies. A recent study from the Massachusetts Institute of Technology (MIT), titled “The GenAI Divide: State of AI in Business 2025,” delivered a sobering statistic: 95% of organizations have seen zero measurable return on their generative AI investments. This is despite a global investment of $30 to $40 billion in the technology.

This lack of ROI is beginning to impact AI adoption rates. For the first time since the AI boom began, the US Census Bureau’s Business Trends and Outlook Survey (BTOS) has shown a decline in AI adoption among large companies. Adoption among firms with over 250 employees dropped from a peak of 14% earlier in 2025 to around 12% by late summer. This suggests that the initial excitement is wearing off as businesses confront the reality that implementing AI is not a guaranteed path to profitability.

Gartner, the technology research firm, has placed generative AI in the “Trough of Disillusionment” in its 2025 Hype Cycle. That phase means waning interest as the initial hype fades and the implementation challenges become more apparent. Gartner estimates it could take 2-5 years for gen AI to move toward productive use.

The double-edged sword of AI in coding

One of the most touted use cases for AI is software development, where AI-powered coding assistants are supposed to change how we write code. The reality is messier. While some studies show these assistants can boost productivity, particularly for junior developers, others suggest a “productivity placebo” effect. Developers may feel more productive, but the measurable gains are often marginal or even negative.

There are also growing concerns about the quality and security of AI-generated code. Research has shown that code produced by AI assistants can introduce far more security vulnerabilities and design flaws than human-written code. For instance, by mid-2025, AI-generated code was linked to over 10,000 new security findings per month in studied repositories, a 10x increase in risks. Vulnerabilities include privilege escalation issues (up 322%) and other flaws tied to 38 Common Weakness Enumeration (CWE) categories. This creates a new set of problems for developers, who must now spend more time reviewing and debugging AI-generated code, which can cancel out any initial time savings.

Source: Apiiro Ltd.

Source: Apiiro Ltd.

Source: Apiiro Ltd.

Wall Street’s worries: a correction on the horizon?

The financial markets are also showing signs of nervousness. A recent Goldman Sachs report warns that a slowdown in AI spending could push the S&P 500 down 15-20%, and that a correction is on the table. The report notes that while AI spending is currently strong, a deceleration is expected in late 2025 and into 2026, as companies shift focus from capex to other priorities like share buybacks.

A recent McKinsey report points the same way. AI adoption has increased to around 72% among companies, but only a small percentage attribute a significant portion of their earnings to the technology. Over 80% of businesses using AI report no real gains in earnings. That disconnect between investment and financial returns is a classic indicator of a bubble.

The bubble appears more acute in private markets, where venture capital and startups like OpenAI face inflated valuations driven by speculation rather than performance. Nearly 500 private AI companies are now valued over $1 billion each, with an average valuation of $5.4 billion and often no proven revenue behind it. In contrast, large, diversified public companies like Google and Meta are less vulnerable, as their revenues span advertising, cloud services, and other sectors beyond pure AI plays. This disparity suggests that a burst could hit venture-backed firms hardest, while Big Tech consolidates its advantages.

Physical barriers: hitting the wall on infrastructure

AI’s rapid growth is increasingly constrained by physical limits, especially shortages in data center capacity and energy supplies. Projections put global data center capacity growth at 15% a year through 2027, but demand, driven largely by AI, may outpace it and create bottlenecks. Energy is the harder constraint: AI’s global consumption is expected to reach 200 TWh in 2025, more than entire countries like Belgium consume. In the US alone, data center power needs could strain the grid, and 40% of AI data centers could face shortages by late 2025 without new infrastructure.

These constraints could drive costs up. Electricity for data centers is projected to more than double by 2030, which would stifle innovation as companies compete for limited resources. Water use for cooling is another concern, since massive facilities are sprouting across regions already under resource stress. Without rapid progress in efficient computing or renewable energy integration, these barriers could slow AI development and deepen bubble risks.

Acknowledging the bubble from within

Some of the strongest warnings come from inside the industry itself. OpenAI CEO Sam Altman has openly stated that he believes we are in an “AI bubble” and that investors are “overexcited.” That is an unusual admission from the head of one of the most prominent AI companies in the world.

Martin Eftimoski, a former economist for the Reserve Bank of Australia, has publicly said he removed all AI investments from his pension and moved entirely to cash. He argues that business leaders often misunderstand AI’s actual usefulness, treating it as a panacea rather than a tool with limited practical applications in many sectors. His warning adds to the broader concern that overhyped expectations are driving unsustainable investments.

OpenAI’s own financials show massive spending and no certain path to profitability. The company is reportedly burning through billions of dollars in cash and is projected to spend $115 billion through 2029 on infrastructure and operations. This high-burn, high-investment model is sustainable only as long as investor confidence remains high, but losses reached $5 billion in 2024 alone on $3.7 billion in revenue.

AI adoption and impact: a data overview

The table below collects the main adoption and economic metrics from these reports:

This data shows a wide gap between adoption hype and tangible outcomes, which reinforces bubble concerns.

The road ahead: a correction, not a collapse

While the evidence for an AI bubble is compelling, a bursting bubble doesn’t necessarily mean the end of AI. The dot-com bust of the early 2000s, for example, was followed by a period of consolidation and the rise of the internet giants we know today.

A similar scenario is likely for AI. The current hype may be unsustainable, and a market correction may be inevitable. But the underlying technology is real. The companies that survive the correction will be those that can demonstrate real-world value and a clear path to profitability. For now, proceed with caution. AI itself is not going anywhere, but the road ahead may be bumpy, especially as infrastructure limits and ROI realities set in.

Sources

AI’s Trillion-Dollar Shakeout: Navigating a Potential Bubble, Correction, Or Market Reset - Forbes
How Tech Giants Have Manipulated the AI Market to Their Advantage - AI Now Institute
Analyzing the Possibility of an Artificial Intelligence Bubble - Built In
Former RBA Economist Expresses Concerns Over an AI Bubble, Moves Superannuation to Cash - News.com.au
Can the U.S. Infrastructure Support the Growing Demands of the AI Economy? - Deloitte
Over 25 Key Statistics and Trends for AI Data Centers in 2025 - The Network Installers
Potential Power Shortages for AI Data Centers by 2025 - PCMag
AI Adoption Declining Among Large Companies, According to U.S. Census Bureau - Tom’s Hardware
Gartner’s 2025 Hype Cycle for AI: Moving Beyond Generative AI - Gartner
The Paradox of AI Coding Assistants and Their Impact on Productivity - Cerbos
AI Coding Assistants Increase Development Velocity but also Security Vulnerabilities - Apiiro
Goldman Sachs Issues Warning: A Slowdown in AI Could Negatively Impact the Stock Market - Yahoo Finance
MIT Report Finds 95% of Corporate Generative AI Pilots Are Unsuccessful - Fortune
Sam Altman’s Warning of an AI Bubble and Overvaluation Concerns - Yahoo Finance
OpenAI’s Projected Business Losses of $115 Billion Through 2029 - Reuters