Have investors already priced AI assets well above their potential to benefit? Will AI data centers built today prove to be the “dark fiber” of the 2020s, in place many years before sufficient demand arises to utilize them?

Trends in Artificial Intelligence (AI) have many of the ingredients for a bubble. It is a transformational technology with applications throughout the economy. It could be as important for economic development as the steam engine, electricity, or the Internet. Most bubbles begin with an element of truth, and AI has that. In our view, investors then get carried away and push prices to unsustainable levels relative to rational analysis. Those prices tend to distort the real economy, leading to overinvestment in the bubble area. The dotcom bubble may be a good analogy. The Internet was transformational, but high prices for anything Internetrelated led to situations such as speculative investment in what became known as “dark fiber,” or fiber optic capacity that was installed many years before the demand from Internet traffic to use it. Much of the debt raised to finance the “dark fiber” buildout eventually defaulted.

In our view, there will eventually be an AI bubble. The question is whether we are in one already. Have investors already priced AI assets well above their potential to benefit? Will AI data centers built today prove to be the “dark fiber” of the 2020s, in place many years before sufficient demand arises to utilize them? These are very important questions to us because 1/3 of CI Segall Bryant & Hamill Asset Management’s All Cap Core Thematic strategy is currently either invested in the AI theme or otherwise highly influenced by AI-related developments. We believe the most important determinant of the strategy’s performance over the next five years will be getting this right: staying invested in AI when it makes sense and exiting when it makes sense.

Our current assessment is that we are not in an AI bubble. We believe this for several reasons. One is that the hyperscale AI cloud companies base their decisions on whether to expand capacity on short-term demand signals. In some cases, they sign contracts with customers that require them to deliver AI capacity and then invest in that capacity. In others they measure real-time activity and see growth that suggests additional investment is warranted. They also have important businesses outside of cloud that can utilize the capacity if third-party demand slows. For example, Amazon can use cloud capacity in its e-commerce business¹, while Google can use it in its search business². CoreWeave is a specialty AI cloud provider that provides additional capacity for hyperscale customers like Microsoft and Google. It may act as a canary in the coal mine (canaries would stop singing or fall unconscious if encountering toxic gases in a mine), as its capacity would likely go unused prior to a softening in demand for capacity that is owned and operated by Microsoft or Google directly. CoreWeave recently raised its revenue guidance because its customers are contracting for additional future capacity, indicating the canary is still singing, suggesting the coal mine is safe for now.³

We also track funds raised or earmarked to be spent from three sources. The first is hyperscale capital expenditure. Each quarter when Amazon, Microsoft, Alphabet, and Meta report, they continue to maintain or increase guidance for future capital expenditure. The second is venture capital funding for large-language model (LLM) developers. Leaders in LLM development such as OpenAI, Anthropic, and their peers continue to raise more funds at higher valuations. The third is private equity and sovereign wealth. They continue to raise and pledge larger funds for AI infrastructure - data center shells and electric power production as well as the servers and network equipment within them. For example, in May 2025, President Trump visited the Middle East and both the Kingdom of Saudi Arabia and the United Arab Emirates announced investment pledges, including for a $10 billion data center in Abu Dhabi that would be the world’s largest outside the U.S.⁴ We believe funds raised from these sources will be spent on AI infrastructure. So long as the funds raised continue to grow, we anticipate more future demand for AI infrastructure. Presently all three are green lights, suggesting it is appropriate to remain invested in AI, in our opinion.

However, trees do not grow to the sky, every business has cycles, and eventually this will end. Investors currently expect Nvidia to continue to grow its sales and earnings. At some point investors may come to believe the cycle has peaked and reprice Nvidia to a depressed level that is more indicative of peak earnings. We believe that may be a “moment” for the global capital markets; if Nvidia falls, we believe so will the Nasdaq, the U.S. stock market, global stock markets, and other risky assets like cryptocurrency and high yield debt. In our view, the most likely driver will be a buildout of AI infrastructure that clearly exceeds demand growth, sparking a pause for digestion. It could also come from a slowdown induced by the Federal Reserve in response to inflation, although this seems less likely in the near term since the central bank appears poised to resume its rate reduction cycle. Alternatively, the trigger could come from an act of war by China against Taiwan, such as a naval blockade, which would significantly impair the ability of the AI ecosystem to meet global demand. But the U.S.-China rivalry has moderated in recent months, with the two sides possibly moving closer to a trade agreement in the fall, reducing the likelihood of this scenario as well, in our opinion.

“The path won’t be smooth--for one thing, we haven’t banished the business cycle--but the fundamentals are in place for bounty that vastly exceeds anything we’ve ever seen before.” – Eric Brynjolfsson