Addressing Big Questions: Is the AI Market a Bubble About to Burst?

So, is the AI market bubble about to burst?

Before we answer that question, we have to interrogate it properly. That means asking the following preliminary questions: First, what is an economic or market bubble? Next, are we currently experiencing an AI market bubble? If the requisite criteria are satisfied and the answers are affirmative, we can consider whether an AI bubble is about to burst like a festering pimple on the acne-scarred face of an oleaginous adolescent.

The definition of an economic bubble on Wikipedia is serviceable enough for our purposes today:

An economic bubble (also called a speculative bubble or a financial bubble) is a period when current asset prices greatly exceed their intrinsic valuation, being the valuation that the underlying long-term fundamentals justify. Bubbles can be caused by overly optimistic projections about the scale and sustainability of growth (e.g. dot-com bubble), and/or by the belief that intrinsic valuation is no longer relevant when making an investment (e.g. Tulip mania). They have appeared in most asset classes, including equities (e.g. Roaring Twenties), commodities (e.g. Uranium bubble), real estate (e.g. 2000s US housing bubble), and even esoteric assets (e.g. Cryptocurrency bubble). Bubbles usually form as a result of either excess liquidity in markets, and/or changed investor psychology. Large multi-asset bubbles (e.g. 1980s Japanese asset bubble and the 2020–21 Everything bubble), are attributed to central banking liquidity (e.g. overuse of the Fed put).
In the early stages of a bubble, many investors do not recognise the bubble for what it is. People notice the prices are going up and often think it is justified. Therefore bubbles are often conclusively identified only in retrospect, after the bubble has already "popped" and prices have crashed.

Behavioral finance theory attributes stock market bubbles to cognitive biases that lead to groupthink and herd behavior. Bubbles occur not only in real-world markets, with their inherent uncertainty and noise, but also in highly predictable experimental markets. Other theoretical explanations of stock market bubbles have suggested that they are rational, intrinsic, and contagious.

Economic bubbles are complicated phenomena, all the more so because we cannot conclusively diagnose them until after they’ve become history.

Overhyped, Overvalued, but a Bubble?

Some of you might recall the notorious Dutch Tulip Mania of 1630. No? I suppose that was before your time. You probably don’t remember the South Sea Bubble of 1720 either. But I wager that you remember the Dotcom bubble, which burst spectacularly in 2000 and dramatically reconfigured the tech industry for years afterward. Many companies met their demise when that bubble popped, while others emerged in the aftermath of the market carnage to become linchpins of a resurgent industry in the decades that followed. Among those post-bubble winners were (and are) Google, Facebook (Meta), and Amazon. Microsoft has been around a lot longer, making its initial fortune on the considerable franchises of Windows and Office.

Now that we’ve defined a market bubble, and we even have experience of how they look and feel, let’s look at the state of the AI market. Does it qualify as a market bubble? Well, as we’ve already determined, we won’t have a definitive answer unless it actually bursts.

Nevertheless, astute observation allows us to note that AI-related valuations appear unduly inflated on both private and public markets. Some investors, like sherpas on steep mountain ascents to dizzying altitudes, pause their progress and wonder whether they should persist in their upward journey. In the camps above, they see feral outbreaks of irrational exuberance. Do you join the party or turn back?

Perhaps the market is not bubbling so much as frothing. Maybe a correction, involve a turning down of the heat so that the market simmers rather than boils, is all that we need. That’s where I believe we are, and what I think will happen. My supposition, unproven, is that the market will undergo an attitudinal reassessment, sobering up after a wild bacchanal. I don’t think we’re destined for a bursting bubble on the horrific scale of superabundant Dutch tulips or redundant dotcom businesses.

AI has been hyped and oversold, but if we view it as a means of enhancing the automation of business processes and job functions, it already possesses practical value. It’s not a panacea for all that ails us, we shouldn’t use it as a life coach, and it won’t render humans economically superfluous. Even so, it should prove to have technological staying power.

At first, we might be disappointed in the results because the hype was (and is) truly beyond the pale. We won’t perceive AI as revolutionary. We’ll just see it as another technological mainstay, part of our daily existence, like the web and smartphones. It’s only in retrospect, as with the bursting of market bubbles, that we will finally be able to make sense of the evolutionary breadth and scope of AI’s impact. At first, we’ll conclude that it fell short of expectations, but that’s only because expectations were wildly unreasonable, set by people who foment market hysteria for a living.

How Can We Tell?

Let’s pretend, for the sake of our mutual amusement, that we are confronting an AI market bubble.

Some people say they can see the signs. I’ve already mentioned the distended market valuations of companies directly involved in AI, providing the necessary AI infrastructure or developing and delivering AI applications and services.

Unfortunately, they aren’t the only ones benefiting from an AI market halo. Many companies only indirectly or peripherally involved with AI are managing to paint themselves in AI colors, boosting their stock prices by association. The irony is, depending on whether the AI market continues to ascend or plummets like a theme-park thrill ride, the modest connection those companies have to AI might be a blessing or a curse.

If a perceived bubble does burst and AI suffers from a sustained backlash, AI pretenders can shed their cosplay identities and clamber to the refuge of more conventional technology markets. Companies that materially depend on AI for revenue and profitability will not have the same option.

Some AI market mania is receding, but we don’t yet know what that means, if anything. Just a short time ago — I’m talking days rather than weeks — Meta’s Mark Zuckerberg was throwing superstar-athlete money at data scientists and researchers who might help his company expedite the realization of its belated AI ambitions. Suddenly, however, he’s hit the brakes, declaring an AI hiring freeze. Zuckerberg is a man of enthusiasms. Remember the grandiose “metaverse”? Zuckerberg probably hopes you’ve forgotten about it. He piled into it with wild abandon, first hiring and then shedding staff when he realized he’d made a major strategic blunder.

Zuckerberg chases the shiny object, whatever shines brightest at the time, and he decided recently to run hard toward the refulgent AI beacon. Now, he’s decided to pause his pursuit, temporarily halting Meta’s AI blitzkrieg. Maybe it’s a brief respite, as Meta says, or perhaps it will be more like the company’s ungainly retreat from the metaverse. Until 2014, Facebook’s motto was “move fast and break things,” subsequently supplanted by the more circumspect “move fast with stable infrastructure,” which compensates for its lack of irreverent frisson with a sensible-shoes approach to progress. Somehow, though, as its lavish AI hiring spree attests, the company retains a predilection for lavish excess.

That’s precisely why we shouldn’t extrapolate too much about the precarious nature of an AI bubble from recent mood swings at Meta. Enthusiastic advances and chastened retreats are second nature at Meta, practically standard operating procedure. Such events, on their own, reveal nothing substantive about the health of the broader AI market.

So, if what’s happening at Meta is a false signal, where can we look for useful diagnostic data?

Let the Numbers Do the Talking

Growth and expansion in the AI market have been driven mostly by AI infrastructure, which is why Nvidia now wears the crown as the publicly listed company with the largest overall market capitalization. Continued procurement of AI infrastructure is a leading indicator of the health of the overall AI market.

Cloud giants, sovereign AI operators, and enterprises aren’t buying AI infrastructure just to park it in storage, something telcos did all the time in the first wave of the internet era. That cloud provides and others to order and consume AI infrastructure tells us that they remain motivated by the prospect of AI revenue generation and cost savings. There’s some AI revenue out there, predicated on a limited but growing spectrum of use cases, but is there enough adoption and persistent growth to sustain the market’s ebullience?

That’s the real question, and we don’t have enough evidence to render a definitive verdict one way or the other. We have some Wall Street analysts who are eternal optimists, and still others who seem to be suffering from a variant of Stockholm Syndrome. We can’t trust their sunny assessments because it’s apparent they are subject to the analyst world’s answer to regulatory capture.

Do you want to know how to tell whether an AI bubble is on the verge of bursting? Monitor the results of AI picks-and-shovels purveyors. If Nvidia, AMD, Broadcom, the contract manufactures, Dell, HPE, Cisco, and Arista begin to miss the mark in revenue and earnings, citing a softening of AI-related demand, you’ll know the wheels are coming off the AI limousine. You’ll have to be quick on the draw, though, because a trickle of bad news might suddenly become a torrent.

Another salient indicator of a bursting bubble is a sharp reduction in capital spending by the cloud giants. If the AI component of infrastructure expenditures — until now punching above its weight of revenue generation— contracts significantly or even decelerates, you’ll have a reliable presentiment that winter has come early.

There should, of course, be a correlation, perhaps even a causal relationship, between decreased AI infrastructure spending and reduced revenue and profitability at the AI infrastructure vendors. Don’t watch the skies, watch the numbers.

Even if an AI bubble does burst, it won’t mean that AI is dead. The dotcom bubble burst spectacularly, as we all know, but the Internet and the web persevered, with fewer players, and went from strength to strength in subsequent years. Fortunes were made, careers blossomed, companies grew and flourished. That might happen again, but we have learned that history prefers creative variations on a theme rather than mindless repetition.

As AI becomes real and substantive, with practical applications and compelling value propositions, it will lose its ethereal halo. AI isn’t a life coach, a psychologist, or a creative writer (not even close) — it’s just software, reams of data, and inanimate machinery — but it can provide automated efficiency and greater productivity for many functions and processes. That’s a market opportunity of significant heft, but it’s not the shimmering hill of magic beans that we’ve been sold.