It’s hard to scroll through any financial news feed these days without bumping into talk of an "AI bubble." The sheer scale of investment pouring into data centers and artificial intelligence is staggering, and since ChatGPT burst onto the scene, it's become undeniable that AI is a transformative, general-purpose technology. In fact, without this wave of investment, the US economy’s growth would look considerably weaker, with a noticeable shift from consumption towards technology-led investment activity.
This rapid acceleration naturally sparks comparisons to past manias. But are we truly on the precipice of a bubble pop? Let’s take a closer look.
The Technology Itself: A Rapid Ascent
AI models are evolving at an astonishing pace. Their capabilities are rapidly approaching, and in some areas, exceeding, industry-expert levels. The complexity and length of tasks AI can handle are growing exponentially, reportedly doubling every seven months. CEOs of major tech companies, the "hyperscalers," are openly discussing their pursuit of Artificial General Intelligence (AGI) – AI that can operate at or above human intelligence across a broad spectrum of tasks – and they believe it's within reach. Their perspective is that the risk of underinvesting in this critical technology far outweighs the risk of overbuilding, especially since much of this investment is being funded through equity and cash flow, not debt.
This leads to a couple of key takeaways. First, given the ongoing progress and future potential, this investment surge is likely to continue for some time. We're still relatively early in this capital expenditure upswing, only about three years in. Second, the hyperscalers’ assessment of the asymmetric risks they face means the potential for an eventual overbuild is indeed growing.
Demand Outpacing Supply?
On the demand side, the acceleration is equally remarkable. Google reported a 5,000% year-over-year growth in inferencing tokens, which then doubled again in just a few months. Microsoft cited around 500% growth. Despite hyperscalers and chipmakers showing no signs of slowing their capital expenditures, demand is already outstripping capacity, leading to supply constraints. This doesn't exactly scream "bubble-like setup" at this point.
While much of the conversation has been US-centric, China is actively working to catch up. Their approach tends to focus on quicker, more cost-effective use cases, whereas US efforts are broader and more ambitious. Currently, there's little evidence to suggest China is ahead in terms of productivity or use cases, with most observers estimating a one- to two-year lag. Nevertheless, Chinese hyperscalers are also significantly ramping up their investment plans; Alibaba, for instance, announced a $53 billion capital expenditure plan over three years, only to revise that forecast upward recently.
Potential Roadblocks Ahead
So, what could potentially derail these ambitious investment plans? In the US, power availability is emerging as a significant bottleneck. Regulators are increasingly pushing data centers and hyperscalers to bear a larger share of energy costs. From a credit perspective, this introduces new risks, particularly concerning stranded assets. Balance sheets are starting to be leveraged in anticipation of sustained demand growth. While this might work in the short term, delivering new supply will become more challenging in the medium term due to turbine shortages, regulatory hurdles, and the higher cost of new builds compared to acquiring existing infrastructure. However, the fact that most of these investments are being financed from free cash flow, rather than debt, does mitigate some of the broader macroeconomic risks associated with this particular capital expenditure cycle.
Comparing to the Dot-Com Era
The sheer scale of these technology-driven investments inevitably invites comparisons to the late 1990s and the dot-com bubble. Yet, the differences are quite striking. Valuations today are roughly half of what they were back then, and crucially, we are already seeing tangible, real-world use cases for AI. Companies like Meta and C.H. Robinson are reporting measurable improvements in margins and revenues directly attributable to deploying AI at scale, demonstrating exceptional returns on investment. Even if only a handful of companies achieve this level of success in the near future, competitive pressures will undoubtedly drive widespread adoption across industries. While supply constraints might temper the pace, cumulative spending in the trillions isn't irrational against a $130 trillion global economy if the use cases are real and imminent.
Exuberance is certainly in the air, but whether it tips into irrationality will depend on how quickly these AI applications can scale. And unlike in 2000, some of the most promising use cases could be just a year or two away, given the current pace of progress. Ultimately, while an eventual overbuild and market correction are certainly possibilities, neither the current pace of investment nor today's valuations suggest that anything of the sort is imminent. Navigating periods of exceptional technological change is always a challenge, requiring a dynamic and active investment approach.
