With AI investment surging and valuations elevated, comparisons to past market bubbles are inevitable. We examine what’s fueling those concerns and why today’s AI boom may be different.
As investment in artificial intelligence (AI) continues to accelerate, comparisons to prior market bubbles have become increasingly common. These concerns are not frivolous. Valuations are elevated, the U.S. stock market is unusually concentrated, and last week’s election results could slow the pace of data center development. In addition, a growing share of AI investment is financed through debt and complex arrangements among AI companies.
History gives investors good reason to be skeptical. The internet transformed the global economy, but that did not prevent enormous amounts of capital from being destroyed during the dot-com bust.
Today, there are certainly pockets of speculation within the AI ecosystem, but the companies driving both AI investment and the public equity market look very different from the speculative businesses that defined previous bubbles.
Are Concerns About Valuations, Financing, and Leverage Justified?
The largest AI-related stocks command healthy valuations, but these premium prices should be considered alongside growth, profitability, balance-sheet quality, and reinvestment opportunities.
Today’s largest technology companies generate substantial earnings and cash flow, produce high returns on equity, and―outside of more concentrated semiconductor businesses such as NVIDIA―operate highly diversified businesses. Just as importantly, they now have a generational opportunity to reinvest significant capital with high expected returns.
Much of the recent scrutiny has focused on how that investment is financed. AI companies increasingly invest in one another, and that capital often flows back to the investor as purchases of its chips or cloud services. This circular financing sounds alarming, but the funding is generally small relative to the scale of the companies making the investments.
As AI-related spending rises, hyperscalers (e.g., Amazon, Microsoft, and Alphabet) are increasingly pairing their own cash with debt, equity, and other financing. Yet these companies retain exceptionally strong balance sheets. For these companies, adding modest leverage can enhance returns to equity holders without meaningfully compromising financial flexibility.
Most importantly, these companies are not locked into today’s capital-spending trajectory forever. If demand disappoints or expected returns deteriorate, management teams can scale back future projects, redirect data center capacity to alternative workloads, or focus more heavily on extracting greater productivity from existing infrastructure.
Not Every AI Company Is Created Equal
These hyperscalers are fundamentally different from neocloud providers such as CoreWeave and private AI model developers (often called frontier labs) such as OpenAI and Anthropic. Concerns about those businesses do not necessarily apply to the diversified, highly profitable hyperscalers that dominate public markets.
Neoclouds tend to have narrower businesses, greater dependence on external financing, and higher costs of capital. Frontier labs, despite their size, are still unprofitable and also continue to depend heavily on outside investors.
In contrast, Amazon, Microsoft, and Alphabet have diversified businesses and enormous existing revenue streams. Their ability to withstand disappointing AI returns is dramatically different from that of a highly leveraged infrastructure provider dependent on one source of demand.
Ultimately, the AI Debate Is about Demand
AI is already reshaping knowledge work such as coding, writing, and research while making meaningful inroads into fields like cybersecurity, finance, and legal services. And these are only the early use cases.
The next wave of innovation will include AI agents capable of taking actions on behalf of users. With increasingly sophisticated mathematical and scientific reasoning, there are also opportunities for health care applications, autonomous vehicles, and robotics. Beyond those, lie further advancements that are difficult to envision today.
The relevant question, therefore, is not whether today’s AI applications can justify tomorrow’s compute capacity, but whether demand for increasingly capable and affordable AI is approaching saturation. We believe the answer is no.
What about Overbuilding?
Transformative technologies tend to attract too much capital too quickly, which will almost certainly cause some AI infrastructure to earn disappointing returns. Interestingly, however, physical constraints may help moderate that risk.
Data centers cannot be built overnight. Each requires land, power, and specialized equipment, along with years of planning and construction. These hurdles prevent supply from growing too rapidly, while providing more time for underlying demand to grow. It also gives developers opportunities to reassess investments throughout a multiyear project, such as when to purchase the most expensive component—cutting-edge semiconductors—which do not need to be purchased at the beginning of the development process.
Political opposition to data centers presents another obstacle, but one that should create winners and losers rather than invalidate overall AI demand. Restrictions on new development could hurt businesses tied to semiconductors and AI infrastructure while simultaneously increasing the scarcity value of existing data centers.
Much of the data center capacity scheduled to come online in 2027 is already under construction, which should limit the near-term impact of potential new restrictions. Any delays to future development could still weigh on earnings, but we expect these headwinds to be short-lived. Ultimately the long-term trajectory of AI investment will be driven by the value the technology delivers, not state and local regulations.
Reframing the Risk
With an opportunity this large moving this fast, some capital misallocation is inevitable. Some AI companies will fail, some infrastructure will be overbuilt, and some frontier labs may never justify their valuations. But pockets of speculation are not the same thing as a system-wide bubble.
We believe we’re still in the early stages of applying increasingly capable, affordable, and accessible intelligence to an almost limitless collection of problems. The path will be volatile, and the ultimate scale of AI demand remains uncertain. But today’s AI boom looks less like overzealous speculation and more like a historic, yet economically rational, capital investment cycle. Ironically, fear of a bubble may be the best safeguard against one, keeping investors and companies disciplined in their underwriting of investments.
Looking for personalized guidance?
Our advisors bring financial expertise and biblical wisdom together to guide the decisions shaping your future.
