Is the AI (Artificial Intelligence) craze cooling down?
According to a biweekly business survey conducted by the U.S. Census Bureau involving tens of thousands of companies, the application of AI in American businesses appears to have experienced a temporary decline. In large enterprises with more than 250 employees, the application rate of AI in production and services peaked at 15% in June, but fell to about 11% by the end of August. Small and medium-sized enterprises are also showing a similar downward trend.
Nevertheless, the current actual usage rate of AI is still significantly higher than the same period last year, more than double that of last year. Particularly in industries such as finance, technology, and legal services, the penetration of AI continues to deepen.
Denis Depoux, co-chairman of Roland Berger’s global management committee, recently stated in an interview with Yicai that a normal technology cycle typically experiences a phase of rapid growth, followed by a market contraction, which can sometimes appear quite severe. However, he believes that there are currently no signs of a dramatic burst of the AI bubble.
He stated: “The current trajectory of AI development is not fundamentally different from past technology booms. The application potential of AI in the service industry is particularly vast, and it is expected to enter a phase of accelerated development. More importantly, AI is also gradually achieving key breakthroughs in the manufacturing sector. Integrating AI into fully digitalized and automated production systems will significantly boost productivity. Therefore, I do not believe the AI craze has ended.”
Some companies are losing patience with AI applications.
However, there are also doubts in the market regarding the sustainability of corporate AI applications. Torsten Sløk, chief economist at investment firm Apollo, recently wrote that the data from the U.S. Census Bureau may indicate that companies relying heavily on AI to support high valuations may face pressure.
Harrison Kupperman, founder and chief investment officer of Praetorian Capital, raised concerns from an investment return perspective. He believes that large-scale technology companies and AI leaders like OpenAI plan to invest heavily in building AI data centers within a year, which means they will need to generate an additional $40 billion in revenue each year over the next decade just to cover their depreciation costs.
On the other hand, companies still face significant challenges in implementing AI. A study from MIT shows that while many companies are satisfied with off-the-shelf generative AI tools launched by OpenAI and Microsoft, the failure rate of pilot projects when attempting to develop customized AI systems is as high as 95%, and these projects often yield substantial returns. The study states that corporate users are generally skeptical of customized or vendor-recommended AI tools, believing they are fragile, over-engineered, or misaligned with actual workflows.
Lane Shelton, vice president of IT consulting firm SHI, added that companies typically take months or even years to transition from adopting AI to realizing returns. The company assists clients in negotiating contracts with software vendors like Microsoft and Salesforce, but it has observed that some companies are gradually losing patience and choosing to halt investments after failing to see measurable benefits. “If you are pushing an AI proof of concept, it may take a year to see initial results. This is not only costly but also requires a leap of faith,” Shelton stated. “The longer you persist, the more likely you are to see unexpected returns, but not every company has that kind of patience.”
Reports indicate that some companies have canceled their subscriptions to AI tools like ChatGPT. OpenAI itself is also adjusting its revenue expectations for some of its products, having recently updated its financial forecasts for investors, raising revenue expectations from ChatGPT while lowering revenue outlooks for AI agents and application programming interfaces for enterprise clients. It is important to note that ChatGPT’s revenue primarily relies on individual consumers.
Goldman Sachs also pointed out potential risks in a recent report, noting that capital expenditures from tech giants like Amazon, Alphabet, Meta, Microsoft, and Oracle are likely to inevitably slow down. The report further proposed an extreme scenario: if these companies cut capital expenditures back to 2022 levels, the loss of AI-related revenue would equate to 30% of the expected total sales growth for S&P 500 companies next year. Consequently, “if long-term growth expectations revert to levels seen in early 2023, it could lead to a 15% to 20% decline in S&P 500 valuation multiples.”
Will the so-called “AI bubble” really burst?
Despite concerns about over-investment in AI, many industry leaders point out that AI is gradually realizing its commercial value.
Ali Ghodsi, CEO of enterprise software company Databricks, stated earlier this month that while AI was once overhyped, an increasing number of customers are deriving actual benefits from AI agents and other AI-driven services. He noted that Databricks provides databases and tools for AI developers like OpenAI and companies across various industries, including automotive manufacturing and oil exploration, achieving a 50% increase in annual revenue, reaching $4 billion.
UBS also released an investment outlook at the end of last month, stating that while investors need to be cautious of the “capital expenditure indigestion” risk that tech giants may face after several years of high capital investment, they remain confident that AI investment will become a key driver of portfolio growth in the medium to long term. UBS emphasized that AI solution providers have made positive progress in converting technology usage into actual revenue and believe that monetization potential continues to expand. For example, tech companies have begun charging retailers for AI personalization tools to enhance customer experience and have implemented subscription fees for tools enhanced by AI capabilities. These commercialization attempts have driven strong revenue growth for cloud service providers, with Amazon, Microsoft, and Google all experiencing year-on-year growth exceeding 25%. As AI application scenarios continue to broaden and integrate deeper into business processes, its monetization space is expected to further expand.
UBS further analyzed that they are optimistic about tech companies ultimately achieving substantial returns from their AI investments. The institution stated that the economic opportunities brought by AI can be understood as a function of three variables: the proportion of tasks that can be automated in the economy, the labor share of these tasks, and the proportion of value that AI suppliers can capture. With a global economic scale of approximately $100 trillion, assuming about one-third of tasks can be automated, and these tasks account for about half of labor costs, and AI suppliers can capture 10% of that value, the potential annual AI revenue opportunity could reach approximately $1.5 trillion. In this context, UBS believes that the projected global AI capital expenditure of $780 billion between 2022 and 2025, and the forecast of capital expenditure rising to $500 billion in 2026, is not overly exaggerated.
For the stock market, UBS believes that the current bull market trend will continue. The AI-driven rally has raised questions about whether the market is in a bubble. Although valuations are above long-term historical averages, the current price-to-earnings ratios of tech giants are still far below the levels seen at the peak of the internet bubble. UBS emphasized that current valuations are supported by strong earnings growth, with companies consistently exceeding market expectations. Furthermore, the key factors that historically triggered bubble bursts (such as significant interest rate hikes) are unlikely to occur in the short to medium term, and investor sentiment does not show signs of excessive optimism.
Goldman Sachs echoed this view in a recent report. The report stated that AI concept stocks are expected to rise 32% in 2024, with a cumulative increase of 17% this year, prompting more investors to inquire whether the current U.S. stock market reflects overly optimistic expectations. However, the Goldman Sachs team pointed out that tech stocks’ current valuations are still below historical bubble levels: “The price-to-earnings ratio of the five major tech stocks (NVIDIA, Microsoft, Apple, Google, Amazon) is 28 times, compared to 40 times at the peak in 2021 and as high as 50 times during the 2000 tech bubble.”
Goldman Sachs believes that a key factor in maintaining valuations within a reasonable range is the enormous revenue generated by companies serving large-scale enterprises. Goldman Sachs estimates that these companies’ capital expenditures have reached $368 billion this year. As for when AI spending will slow down, Goldman Sachs stated, “Analysts generally expect a significant slowdown to occur between the fourth quarter of 2025 and 2026,” but with large-scale companies continuously raising their capital expenditure budgets, this slowdown may be pushed back.
