When Michael Burry speaks, Wall Street listens—especially when his track record includes correctly predicting the 2008 housing market collapse. Now, in 2025, Burry has shifted his critical lens toward the red-hot artificial intelligence sector. In a recent CNBC exclusive, the famed hedge fund manager of Scion Asset Management issued a bold claim: the AI boom mirrors past bubbles in structure, scale, and sentiment—and it’s now beginning to unravel (CNBC, 2025).
According to Burry, investors are vastly overestimating the short-term gains of AI technology while ignoring fundamental inefficiencies, excessive capital burn, and growing systemic risk. His analysis points to deteriorating margins, overleveraged start-ups, and an unsustainable AI chip supply chain as signs that the market is heading into correction territory. Recent moves by Scion—including new short positions in high-profile AI companies—signal that Burry is not just hypothesizing, he’s preparing to profit from the decline.
What Burry Sees: Classic Bubble Dynamics at Play
Burry draws parallels between the current AI enthusiasm and the dot-com boom of the late 1990s. The AI narrative—fueled by ChatGPT’s meteoric rise, generative models like Gemini and Claude 3, and rapid proliferation of startups—produced a FOMO-driven investment wave. However, beneath the hype, Burry identifies troubling metrics:
- Burn rates for AI startups exceeding $1.6 billion in aggregate per quarter globally (VentureBeat AI, 2025).
- NVIDIA’s data center segment margins declining in Q3 2025 for the first time since 2023 (NVIDIA Blog, 2025).
- Corporate earnings reports showing weak ROI on AI deployments across non-tech sectors (McKinsey Global Institute, 2025).
In particular, Burry highlights how AI investments are afflicted by what he terms “deferred revenue delusions”—startups are booking future sales from trials and API agreements as today’s cash flow, a tactic that inflates valuations unsustainably. Much like Enron’s accounting tactics before its collapse, Burry argues these financial structures leave many AI firms vulnerable to liquidity crises when capital dries up.
Costs and Constraints: The AI Infrastructure Bottleneck
Another pillar of Burry’s concern surrounds infrastructure costs. AI’s compute-intensive requirements rely heavily on limited hardware: high-end GPUs, networking gear, and specialized data centers. A recent Deloitte study estimates that constructing AI-ready GPU data centers now costs up to 4.5x more than traditional cloud servers due to energy, cooling, and spatial demands (Deloitte Insights, 2025).
NVIDIA’s dominance in GPU supply has turned the company into the de facto “oil supplier” of the AI era. However, supply chain bottlenecks continue into Q4 2025, and even OpenAI has warned about “severe compute constraints” limiting model improvements and operation uptime (OpenAI Blog, 2025).
These constraints not only increase operational costs but also present geopolitical and regulatory vulnerabilities. Citing FTC inquiries and export restrictions affecting chip sales to China, Burry suggests the sector’s dependency on a single hardware provider is dangerously reminiscent of commodity shocks that triggered past recessions (FTC News, 2025).
Comparative Cost Table: AI Development vs Returns
| AI Use Case | Avg. Development Cost (2025) | Avg. ROI in First 12 Months |
|---|---|---|
| Customer Service Chatbot (Large Enterprise) | $5M | 8% |
| Autonomous Warehouse Robotics | $20M | 11% |
| Generative AI Content Suite | $8M | 5% |
This table, grounded in McKinsey and Deloitte’s 2025 analysis, shows that over-the-top AI investments often fail to justify capital outlay—especially for non-digital-natives like banks or industrial firms.
Market Psychology and Herd Behavior
Echoing lessons from the 2008 subprime crisis, Burry points to market psychology as a primary trigger of AI’s unfolding bubble. Investors are treating AI like a guaranteed disruptive force, assigning sky-high valuations despite low earnings and nascent regulations. The top ten AI firms by valuation, according to recent NASDAQ data, command an average P/E ratio exceeding 120—compared to the S&P 500 average of 27 (MarketWatch, 2025).
Notably, retail investor behavior mirrors crypto’s 2021 mania. Apps like Robinhood and Webull recorded record volumes in penny-AI stocks this summer, driven by TikTok influencers and AI-focused ETFs. This herd mentality primes the market for a sharp correction once sentiment reverses—a dynamic described in Michael Burry’s own words as “a psychological rubber band that’s been stretched too far.”
The Competitive Arms Race: Too Fast, Too Expensive
Racing to dominate generative AI, leading labs—OpenAI, Google DeepMind, Anthropic, and Meta—have entered an unsustainable escalation. Developer benchmarks show that OpenAI’s GPT-5 training cost is estimated to surpass $1.2B, even as Marginal Model Performance Improvement (MMPI) tapers off by just 2-3% in synthetic benchmarks compared to GPT-4.5 (The Gradient, 2025).
Further, OpenAI CEO Sam Altman acknowledged in a podcast this November that “scaling is no longer enough,” referring to diminishing returns from brute-force model size increases (OpenAI Blog, 2025). Despite this, venture capital funds continue pouring into foundation model startups hoping to build the next Claude or Gemini.
Anthropic’s Claude 3 recently launched with significant marketing fanfare, but its stability issues and data leaks have begun eroding enterprise confidence (MIT Technology Review, 2025). As infrastructure, data access, and talent become increasingly expensive, the runway for unprofitable innovation shortens with every quarter.
Implications for Investors and the Broader Economy
While short-term volatility may trigger panic, the long-term implications of an AI bubble burst are economic and societal. According to a 2025 report by the World Economic Forum, AI-driven automation supports over 300 million jobs globally. Sudden contraction could lead to layoffs in companies overly reliant on AI workflows (WEF, 2025).
Investor portfolios—especially those tied to thematic ETFs—may face disproportionate downside. For example, ARK Invest’s AI Innovation ETF dropped 18% in September amid technical corrections. Furthermore, enterprise adoption of AI could slow dramatically, resulting in stalled transformation initiatives across logistics, finance, and healthcare.
Still, the underlying technology may outlast the bubble. As observed with the dot-com era, infrastructure built today can lay the groundwork for durable services once hype subsides. The key, as Burry emphasizes, will be distinguishing utility from speculation—identifying companies delivering value versus those masking fragility behind buzzwords.
Conclusion: Caution as a Strategy
Michael Burry’s track record demands attention. His analysis does not discount the potential of AI but warns against its investor irrationality and over-hyped valuations. As we enter 2026, market correction is no longer theoretical—it is materializing via softening stock performance, shrinking margins, and VC hesitancy. Investors and observers would do well to remember that any innovation, no matter how transformative, is not immune to the laws of economics and risk.
Prudent players in the AI arena must temper expectation with analytical discipline. The market will reward those who identify clear ROI, build sustainable deployment strategies, and plan for volatility. Meanwhile, the unraveling AI bubble may filter speculators from builders—ushering a more mature and efficient future for true AI utility.