The AI Bubble: Is the Global Economy Doomed?
- Kawtar Rettab
- 6 hours ago
- 7 min read
From 1995 to 2000, investors poured billions into internet startups, many of which only had a website and an idea. This resulted in unrealistic valuations, and by 2000-2001, the bubble collapsed, wiped out trillions in market value and bankrupting countless companies. That is the infamous dot-com bubble referenced today, in an age where AI has become the new hype. Is history repeating itself, or is “this time different”?
Background Information
AI has consumed our daily lives, and with it, the whole business landscape. The largest tech incumbents in the industry are pouring billions of dollars into the AI race due to massive potential for future gains. ChatGPT, Deepseek, Gemini, and tons of other large language models (LLMs) are replacing online business activities, such as customer support, automating work tasks, drafting contracts, etc. The productivity gains delivered by AI save companies millions of dollars in spending on less important tasks.
AI activity is expensive and highly demanding. An LLM like ChatGPT works by predicting the next word based on prior context, through trillions of calculations from a massive dataset. LLMs thus require specialized GPUs and chips, the majority of which are designed by Nvidia and manufactured by TSMC (Taiwan Semiconductor Manufacturing Company). The largest investors in AI, such as Microsoft, Amazon, Google, and Meta, have capital expenditures exceeding $380 billion in 2025. However, concerns about an AI bubble are surfacing as the demand for chips exceeds the scarce supply, making the billions spent on investment impractical. Furthermore, investors speculate that AI companies are overhyped, valued based on uncertain future potential rather than practical revenue estimates.
The Bottleneck of TSMC & ASML
TSMC (founded in 1987) is the world’s dominant chip manufacturer, due to its massive scale and early competitive strategy. TSMC operates a unique model; it leaves the chip design to its clients, as a result, avoiding competition, building trust, and gaining specialized knowledge on manufacturing with highly skilled employees. Furthermore, computer chips require nodes. The 3 nm node is the current industry ideal, as 3 nm fits more transistors in a chip, which increases computational power for LLM usage. TSMC has a significant competitive advantage in the field of LLMs, namely, 3 nm node manufacturing with high yield and quality. Other companies, such as Samsung and Intel, are also able to produce 3 nm nodes but have yet to meet industry standards for yield and quality.
The demand for advanced semiconductors by TSMC is so large, it has increased Taiwan’s economic yearly GDP by 7.64%. This has generated a trade surplus of $92.7 billion in 12 months, with the microchip sector accounting for 13% to 15% of GDP. The demand for TSMC chip manufacturing is attributed to the high-performance computing (HPC) ability of TSMC and aforementioned advanced node production capabilities.
Furthermore, TSMC partners with ASML, the world’s sole supplier of Extreme Ultraviolet (EUV) lithography machines; the only type capable of printing ultra-dense circuits onto silicon wafers for producing 3 nm, 2 nm and lower nodes (NASDAQ). This is due to the expense, complexity, and sub-supplier parts required of the EUV systems. A standard EUV machine costs over $100 million. Chip manufacturing requires extreme precision, and performance failures such as contamination, unskilled labor, or even the slightest errors can ruin the yield.
TSMC had decades of perfecting yields, close relations with ASML, predictable delivery, and quality assurance. However, according to Zythos Business, TSMC is both highly profitable and risky due to 3 factors: first, the geopolitical tension between China and the U.S., second, reliance on supply of engineering talent, and third, due to dependence on rainfall for cooling and cleaning machinery to prevent delays. Such factors are beyond the company’s control, posing huge risks. Moreover, the political relationships among Taiwan, the U.S. and China are strained, as the U.S. had proposed a “50-50” split of chip production, which Taiwan refused. China is building its own chip supply chain and has even attempted to poach Taiwanese engineers, which caused recruitment bans and opened criminal investigations.
Geopolitical Risks
Investors fear that China may invade Taiwan for geopolitical or economic reasons, which would disrupt chip supply and AI advancements. On the other hand, an invasion is unlikely for several reasons: 1. The U.S. may protect Taiwan to prevent China reaping the benefits, which risks nuclear escalation, 2. The economic consequences for China would be severe, through sanctions, restrained global shipping, etc., 3. As Taiwan is an island, it may be highly complex to coordinate and establish a foothold in their territory.
Furthermore, as of 2024, Taiwan’s largest trading partners are China, which accounts for 20.4% ($96.8 billion) of total exports, and the United States with 23.5% ($111.3 billion).
Conclusively, supply restrictions from TSMC and ASML will likely slow the progression of AI, and it will take a long time to materialize. Furthermore, geopolitical tensions exist and the future is uncertain; if a conflict does arise, AI innovation will take a massive hit due to restrictions on TSMC. Although TSMC has fabs located or being built in China, United States, and Japan, none have the same capabilities of the Taiwan fabs.
Overvaluation Due To Unrealistic Projections
Another increasing concern for many investors is the overvaluation of companies based on future potential that may be unrealized, presenting a significant monetization lag risk. Several real-world examples are presented by the research article, “Anchoring AI Capabilities in Market Valuations: The Capability Realization Rate Model and Valuation Misalignment Risk”.
Investor optimism in Adobe has outpaced its ability to monetize its AI tools, and OpenAI’s valuation assumes breakthroughs in LLMs that may never be achieved due to resource constraints. Meta’s valuation can be supported by its advertising and engagement performance rather than AI innovation alone, and Microsoft’s OpenAI investment has allowed for early monetization across Azure and Microsoft 365 however similarly faces competitive pressure and uncertainty. The article argues that Nvidia valuation is tied to actual earnings growth, with proven demand for its AI chips, record-breaking revenue growth, and years of R&D that allowed Nvidia to dominate the computer chip industry. Its valuation is thus justified; however, as aforementioned, Nvidia is reliant on TSMC and is subject to bottleneck pressures.
The publication concludes with several policy recommendations to prevent inflated valuations. Firstly, make companies disclose standardized AI metrics such as R&D expenditures or revenue from AI products. Secondly, regulators should enforce fairer valuation and curb misleading AI hype through accurate equity research. Thirdly, governments should invest in AI research to realize its benefits for smoother societal integration, which can justify high valuations. Fourthly, managing labor transitions for successful AI adoption through retraining and education programs. Lastly, building international cooperation through established AI access, safety standards and export cooperation. Considering the possible overvaluation of companies, is the dot com bubble really repeating itself?
Comparing the Dot Com to the AI Bubble
The world expects AI to generate huge profits as a revolutionary technology. The dot-com bubble was fueled by excessive investments without evidence of future profitability. To avoid a similar scenario, a safer approach is to wait until AI companies demonstrate profits that justify their massive investments and stock market valuations.
On the other hand, perhaps it really “is different”; as the companies being invested in are not startups, but rather larger incumbents such as Google or Meta. Such companies have consistent free cash flows from other services as a cushion, and have released successful LLMs such as Gemini or LLaMA. For other companies such as OpenAI, Anthropic or Mistral, AI is the primary product for companies meaning greater risk due to the lack of fallback revenue. However, according to the Stanford AI report, up to 78% of companies use AI, and a study by KPMG revealed that 66% of individuals use AI frequently. This evidences customer interest and demand (unlike with the dot com bubble), signaling lower risk due to future monetization potential. Additionally, the aforementioned companies have successfully monetized their service; for example, OpenAI received $13 billion in 2025 according to the Financial Times.
The risk may instead lie in smaller startups that receive large investments just for offering a specific AI service. For example, CodeParrot received $500,000 in funding as a YC-backed coding tool; however failed to gain sufficient revenue and was shut down. These companies often face high R&D costs, uncertain user willingness to pay, fraudulent practices and limited competitive advantages. Yet the comparison with the dot-com era is not perfect: the dot-com crash was amplified by temporary, artificial demand, such as the Y2K spending surge, as well as fraudulent capital expenditure records from firms like WorldCom. Contrastingly, today’s AI demand is structural and rising. It is expected that in the end, only a few dominant LLMs will power broader AI services; and as with the dot-com era, the real risk may lie in overfunded smaller startups with an AI service idea that cannot justify their valuations.
Conclusion
Just like with the value expected from the internet, we expect great value from AI. However, excessively overvaluing AI due to hype told by important investors and CEOs is a massive risk. Perhaps the same story is unfolding; a new technological breakthrough, where only a few emerge victorious in the AI race. Just like the internet, AI can be improved over time and increase productivity. LLM improvements may take years to materialize, and considering the monetization lag, scarcity of chip production, and geopolitical risks, fears of investors are justified.
References
Business Insider. (2025, April 26). Wall Street’s fear of an AI slowdown is ‘laughable,’ Morgan Stnley says. https://www.businessinsider.com/ai-stocks-outlook-slowdown-spending-chips-gpu-nvidia-deepseek-tariffs-2025-4
Business of Apps. (n.d.). DeepSeek statistics. https://www.businessofapps.com/data/deepseek-statistics/
Finance Yahoo. (n.d.). Nvidia CEO clarifies remarks on China. https://finance.yahoo.com/news/nvidia-ceo-clarifies-remarks-china-150713825.html
IESE Insight. (2025, September 1). TSMC: Lessons in strategy and operational excellence from the world’s chipmaker. https://www.iese.edu/insight/articles/tsmc-geopolitics-operations-strategy
Janus Henderson Investors. (n.d.). AI versus the Dotcom Bubble: 8 reasons the AI wave is different. Janus Henderson. https://www.janushenderson.com/corporate/article/ai-versus-the-dotcom-bubble-8-reasons-the-ai-wave-is-different/
KPMG & The University of Melbourne. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. KPMG. https://kpmg.com/xx/en/our-insights/ai-and-technology/trust-attitudes-and-use-of-ai.html
Moomoo. (n.d.). Nvidia supply chain in one chart: Who are the suppliers of the AI giant? https://www.moomoo.com/community/feed/nvidia-supply-chain-in-one-chart-who-are-the-suppliers-111805778165765
Nasdaq. (2025, May 18). Why ASML and TSMC are the chokepoints in global chipmaking. https://www.nasdaq.com/articles/why-asml-and-tsmc-are-chokepoints-global-chipmaking
Stanford HAI. (2025). The 2025 AI Index Report. https://hai.stanford.edu/ai-index/2025-ai-index-report
Stimson Center. (2025). Rethinking the threat: Why China is unlikely to invade Taiwan. https://www.stimson.org/2025/rethinking-the-threat-why-china-is-unlikely-to-invade-taiwan/
Superlines. (n.d.). ChatGPT statistics 2025. https://www.superlines.io/articles/chatgpt-statistics-2025
Taiwan Semiconductor Manufacturing Company. (n.d.). About TSMC. https://www.tsmc.com/english/aboutTSMC
World’s Top Exports. (n.d.). Taiwan’s top import partners. https://www.worldstopexports.com/taiwans-top-import-partners/
Zythos. (2025, November). Economic and financial analysis of Taiwan’s technological double-helix dominance in microchips and the emerging robotics industry (2024–2025). https://zythos.es/en/2025/11/economic-and-financial-analysis-of-taiwans-technological-double-helix-dominance-in-microchips-and-the-emerging-robotics-industry-2024-2025/