On October 22, 2026, Xinmin Evening News published “Shenhai Commentary | Why Are Capital Markets Divided as the U.S. Ramps Up AI Investment?”, an article by YAO Xu, Secretary-General of the Center for Global AI Innovative Governance and Associate Professor at the Fudan Development Institute, and GUO Haobo, Research Assistant at the Center for Global AI Innovative Governance. The authors argue that assessing the U.S. AI boom requires a closer, longer-term examination of the tensions that AI is generating within American markets, society and the political system.
On October 17, at the 2026 Annual Meetings of the International Monetary Fund (IMF) and the World Bank Group, Clark, a co-founder of U.S. technology company Anthropic, said that artificial intelligence (AI) capabilities were advancing rapidly. According to the company’s internal modelling, AI could help increase U.S. GDP by 50% by 2030. Yet despite this enticing prospect, U.S. capital markets are displaying two distinct sentiments: continued increases in investment alongside concerns about overheating and a market bubble.
Are Investment Returns Being Overestimated?
U.S. investment in AI has maintained strong growth in recent years. Figures suggest that capital expenditure by the five hyperscale cloud providers is expected to approach $800 billion this year. Behind this spending are high expectations for AI’s prospects and a belief that the technology is creating a vast new market. Those who remain bullish argue that large-scale investment today is necessary to secure a competitive position in tomorrow’s industries. U.S. asset manager BlackRock continues to rank AI among its highest-conviction investment themes. It argues that although the computing infrastructure and research and development systems now being built will take considerable time to generate returns, they could provide an important foundation for future economic growth.
At a conference in New York on September 20, IMF Managing Director Kristalina Georgieva said that although the world economy had proved more resilient than many had expected, it still faced risks from persistent inflation, rising debt-servicing costs and overheated AI investment.
Source: Xinhua
Others, however, believe that this continued investment is pricing in future gains too far in advance. Ray Dalio, founder of U.S. investment firm Bridgewater Associates, has described the AI boom as a “classic bubble”, arguing that heavy borrowing combined with rising interest rates is bringing conditions closer to those associated with a bubble bursting. Bank of America’s October Global Fund Manager Survey found that nearly half of respondents expected a bubble to emerge next year. Some companies have already taken on costly debt to secure a share of the computing market. Computing infrastructure provider CoreWeave recorded net interest expenses of $640 million in the second quarter of this year, equivalent to roughly a quarter of its revenue. Credit-market pricing implies a 63% probability that CoreWeave will default within the next five years, while its shares have also suffered a sharp recent decline.
Despite these divergent assessments, fund managers’ net overweight allocation to technology stocks has risen to its highest level since August 2020. This apparent contradiction reflects investors’ concern that AI is being overhyped, coupled with their reluctance to miss out on the growth generated by this technological shift. Neither side disputes AI’s technological prospects. Their disagreement concerns whether actual returns can justify today’s massive spending and high valuations. Understanding the U.S. AI boom therefore requires examining both industry supply and end-user adoption.
Overcapacity May Lie Ahead
In the short term, U.S. computing infrastructure remains in short supply. Existing construction projects are supported by real demand, and customers have already committed to much of the new capacity before it is completed. According to a report by real estate services firm CBRE, supply in North America’s primary data centre markets grew by 33.7% year on year in the first half of this year. Yet the vacancy rate stood at just 1.4%, and 80.4% of capacity under construction had already been pre-leased.
A shortage of computing capacity today can nevertheless coexist with the prospect of excess capacity in the future. Both may occur within the same investment cycle. Data centres are capital-intensive projects that take several years to build, making it difficult to change course as market demand shifts. The Bank for International Settlements noted in its Annual Economic Report 2026 that current bottlenecks in electricity, advanced chips and grid equipment were prompting companies to secure future capacity through long-term contracts for fear of falling behind.
These forward commitments may, however, become disconnected from actual future demand. Some projected demand has already weakened. The market monitor for PJM, the largest regional power grid in the United States, reported that approximately 42% of previously forecast additional electricity demand from data centres had been cancelled or delayed.
The large-scale expansion of U.S. computing infrastructure also faces constraints from social costs and local policy. In February, hundreds of people gathered outside City Hall in San Marcos, Texas, to protest a proposed $1.5 billion data centre. Residents voiced opposition at a public hearing, citing concerns about water consumption, electricity use and long-term environmental pressures. The city council subsequently voted against the land-use planning amendment required for the project. This was not an isolated case. In July, New York State also announced a pause on issuing state environmental permits for the construction of data centres with power capacity of 50 megawatts or more.
Users Take a More Measured View
In end-user markets, AI is being widely adopted in software development, workplace collaboration and customer service. More businesses are willing to pay for AI-driven productivity, while leading model developers are seeing rapid revenue growth. Reports suggest that OpenAI’s annualised revenue is expected to exceed $50 billion this year. Its research on individual ChatGPT users indicates that, six months after registration, users send approximately 50% more messages per day than they did in their first month.
As AI models become more deeply integrated into established workflows, however, businesses are taking a more measured approach. An EY survey of U.S. executives found that 82% of respondents were concerned about token usage and related costs. An S&P Global survey found that the proportion of companies abandoning AI model projects midway had doubled since its previous survey.
Meanwhile, improvements in model efficiency and intensifying competition are continuing to drive down model usage prices. Technological progress is thus narrowing profit margins to some extent. Venture capital firm a16z estimates that, driven by algorithmic improvements and the spread of open-source models, inference prices for large models delivering equivalent performance are falling by approximately 90% a year on average. This puts greater pricing pressure on model service providers and may also affect margins across the industry.
On July 23, 2025, the White House released an AI action plan outlining dozens of federal policy actions to be pursued by the Trump administration. Some media analyses interpreted the move as reversing several Biden-era policies and easing regulatory restrictions on AI development.
Source: Xinhua
The American public’s perception of AI risks is also shifting. In the Future Risks Report published by AXA and Ipsos, surveyed experts ranked AI and big data risks above climate change as the leading risk over the next five to ten years. The Pew Research Center’s latest tracking survey found that 52% of U.S. adults felt “more concerned than excited” about the use of AI models in daily life, while 71% expected AI to reduce the number of jobs over the next twenty years.
A national poll by Quinnipiac University further found that 74% of Americans did not trust the leadership of AI model companies. Another 77% believed that the development of powerful AI systems should be paused or slowed until their safety had been assessed, and 86% supported mandatory independent safety standards even if these slowed research and development. These widely held attitudes provide a social basis for those who remain bearish on AI or advocate a slower pace of development.
Assessment Requires Multiple Perspectives
The Trump administration, however, remains committed to a race-driven approach and has consistently pushed back against calls to slow down. In response to recent appeals for a slower pace from Anthropic CEO Dario Amodei and others, the administration has taken a firm opposing stance, anchoring its policy in greater investment and winning the AI race. Trump himself has described “so-called AI doomsday warnings” as an exaggerated hoax. Presidential technology adviser David Sacks, formerly the White House AI “czar”, was also quick to publicly mock leading companies for seeking “regulatory capture” and industry monopolies in the name of safety.
This race-driven approach has shaped a series of Trump administration policies since 2025. Through Winning the Race: America’s AI Action Plan, the White House has sought to remove regulatory barriers, accelerate approvals for data centres and power infrastructure, and promote exports of the full U.S. AI technology stack. It has also introduced a National AI Legislative Framework, urging Congress to replace fragmented state-level restrictions with uniform standards and accelerate the establishment of a nationally consistent federal AI policy framework that imposes the least possible burden. Through broad deregulation and faster infrastructure development, the administration aims to secure an overwhelmingly dominant position for the United States.
Assessing whether the U.S. AI surge is a boom or a bubble therefore requires more than observing any single dimension. It calls for a closer, longer-term examination of the tensions that AI is generating within American markets, society and the political system.
Original Link：https://www.shobserver.cn/staticsg/res/html/web/newsDetail.html?id=1199412
