Researchers at Stanford University have warned of the risk of an investment bubble collapse, noting that artificial intelligence (AI) infrastructure investment and market valuation have increasingly supported the U.S. economy and financial markets. They diagnosed that if a shock occurs while AI has yet to produce clear economic results sufficient to recoup investment costs, it could spread to the global economy.
The AI100 project under the Stanford Institute for Human-Centered AI evaluated in its 2026 report, released on the 10th, that AI infrastructure investment is at an unprecedented level in both scale and speed. The analysis suggests that AI investment is fueling bubble concerns as it simultaneously sustains parts of the U.S. economy and drives trends in financial markets.
The report pointed out that NVIDIA's market capitalization exceeded $5 trillion at the time of writing. It analyzed that more than 15% of the U.S. stock market's gains in 2025 can be explained by the rise in NVIDIA's stock price, and when combined with the technology and telecommunications sectors, they account for 60% of total gains.
This means that as market growth is concentrated in specific companies and sectors, the scope of a potential shock could widen if AI investment sentiment weakens.
The researchers diagnosed that, excluding financial market value and capital expenditure, the economic effects of AI have not yet appeared clearly in macroeconomic indicators or corporate return on investment. While massive capital has been poured in, AI has not yet reached the stage of earning back its investment costs. The investment structure of data centers was also pointed out as a risk factor.
They explained that while 19th-century railroads could recoup investment over a long period, computer equipment in data centers must be replaced every few years, resulting in a shorter payback period. They also viewed the fact that a significant portion of AI infrastructure investment relies on borrowings and external capital as a factor that could increase vulnerability to market volatility.
AI infrastructure is also placing a burden on energy and water supplies. The report pointed out that the expansion of data centers is affecting global increases in carbon emissions and energy policy changes in each country.
As computing resources needed for AI model development and data center construction costs rise rapidly, a trend is also emerging where investment scales are concentrated among a few companies and nations.
The report diagnosed that AI competition is being divided primarily between the U.S. and China. It analyzed that while the U.S. leads in developing frontier models that deliver the highest performance, China is ahead in open-weight models that disclose the core parameters of models so they can be utilized by other companies and developers.
Consequently, it explained that different AI ecosystems are forming around the U.S. and China, and other countries are in a position where they must choose which of the two systems to utilize. This means that each country's technological and industrial strategy could be divided between the U.S.-style closed-source model and the Chinese-style open-source model.
Options for AI competition were also presented to South Korea. Explaining that cutting-edge semiconductor production capacity is effectively concentrated in Taiwan and South Korea, the report stated that each country must decide whether to use the U.S. model, utilize…
