Is AI a bubble?

Please subscribe to my newsletter and/or listen to my podcast.

For my sins, I’m a West Ham fan. West Ham, if you don’t already know, is a football (soccer) team based in the East End of London. At each home game, the players come out, for reasons somewhat lost to time, amidst a sea of bubbles. And a song echoes out, around all corners of the stadium. “I’m forever blowing bubbles,” we sing, in unison. “They fly so high, nearly reach the sky — and then, like my dreams, they fade, and die.”

The reason we sing this song is a simple metaphor. The life of a sports fan is one of perpetual disappointment. It’s coded into the way that the game is played: even if you win one week, you might lose the next. There’s no way to keep winning indefinitely. Success has to be followed by failure, sure as failure must, ultimately, become success. Bubbles are blown, bubbles pop. Rinse, repeat.

But economic bubbles aren’t quite so easily accepted as part of the natural way of things. The phrase — which we all know — dates back to the early 18th century, when the South Sea Company was formed in response to protracted wranglings between the British government and the Bank of England. The creation of the South Sea Company afforded the new venture a monopoly over the trade of slaves to islands in the south seas (the southern Pacific, places like Polynesia, as well as Hawaii and New Zealand) which made the shares potentially hugely valuable. They were snapped up with great alacrity, even though the war of the Spanish succession made trade an absolute nightmare, and the bubble went *pop* before anyone could make a profit.

It’s hard to find exactly when the word “bubble” came to mean what it does today, but contemporaneous with the decline of the South Seas Company was the widespread usage of the term. Indeed, it precipitated a specific act of Parliament called the Bubble Schemes (Colonies) Act 1740. Part of the reason why the 18th century experienced so many of these “bubbles” is because of the changing nature of globalisation. Explorers had opened up trade routes across the world, and the possibility to head out and secure a fortune had never been stronger. It also meant that there was a systemic failure of oversight. While the claims of a mariner heading across the English channel were easily verifiable by local merchants, once the distances had expanded to traverse half the world, that sort of reputational security was eroded. Most domestic investors barely knew what was happening in the next county, let alone in Asia, America or Oceania. And so unreliable businessmen were able to float their companies, domestically, as great big exciting bubbles, that would sail off around the world.

There have been many bubbles since the 18th century, largely defined by a simple premise. They grow beyond their structural security. They take on capital that there is no realistic chance of them returning, and they continue to grow, even though they grow weaker, less secure, as they do so. The fact is, there’s no way to unblow a bubble (just ask a toddler), you blow and blow and it either pops in your face, or survives long enough to float across the room. And then pop. There’s no getting the suds back in the soapy water.

The subsection of commodities bubbles predate the formation of the South Seas Company by the best part of a century. The Dutch “tulip fever” is perhaps the best known — and most insane — of all the commodities bubbles, even if the word “bubble” (or bubbel, to use the Dutch word) wasn’t used back then. From there, all sorts of assets have come under the bubble treatment: British railways in the 1840s, internet sites in the late-1990s, Beanie Babies through my childhood, and in the present moment, cryptocurrency. The question of whether crypto is a bubble is live and political. Evangelists for the coins would never concede that they are part of a speculative bubble, but there is little doubt that a bubble, of sorts, went pop in 2017. There have been other bubbles within crypto world that have burst, but, unlike tulip bulbs or stock in a Georgian trading company, crypto bubbles have had a tendency of reinflating themselves.

Which is all to arrive at the, laboured, point: bubbles are all around us. NFTs are perhaps the most egregious recent example of bubbles, where some people decided to use simple artworks as a way of moving large sums of money around. But, along the way, the idea of an NFT as a value generating object came into the mainstream, and a bunch of people (who couldn’t afford it) made dumb bets on objects that had little intrinsic worth, and subsequently lost almost all of their value. It was a salient reminder that most people cannot afford the sort of risky speculation involved in commodities trading.

For those of us who’ve grown bored of speculating about whether crypto is a bubble, there’s a new game in town: Artificial Intelligence (AI). In the past year, AI has seen enormous investment. AI start-ups raised $42.5bn in 2023, which was actually down 10% year-on-year. But a Goldman Sachs report suggested that, in the wake of the success of OpenAI, we could see $200bn invested in the industry by 2025. This would represent an almost unprecedented influx of investment. It would also, undoubtedly, raise the question of whether AI is a bubble.

On the support side, for this proposition, is the behaviour of the markets. The fear of exclusion from this is very real, and very powerful. We’ve seen a bunch of companies, from Meta to Musk Industries, pivoting towards AI research and development. To be a major tech player in 2024 and have no autonomous stake in AI is to look like an ostrich with its head between its legs. But this means there is definitely an overinvestment in AI currently. A large percentage of the companies being created in the AI space, and which are acquiring millions of dollars of investment, are not going to meaningfully exist in 5 years time. There are simply too many companies doing too many similar things, and too much potential to be gazumped by the biggest, most monopolistic players. And so all this capital rushing into an industry which definitely cannot accommodate such plurality of business… well, that’s bubble-adjacent.

There’s also the question of how realistic revenue withdrawal is from the world of AI. The problem with tulip mania was that even though scarcity is a powerful tool in the seller’s arsenal (not unlike complexity, which tends to be the AI equivalent), your options, once you’ve acquired the bulb, are to either trade it again or turn it into a tulip flower. And so the revenue potential is really only in the ability to sell it on, at an inflated price, to another buyer — just as many start-ups are only ever able to generate money, for their founders, through an exit. Then there’s the South Seas example, where geopolitical and economic instability made it very unlikely that a potentially lucrative operation would be able to generate revenue. And so these bubbles didn’t survive the turning point where the desire to input capital is replaced by the desire to output revenue. Most of the big AI companies have fairly limited ambitions about revenue generation in the short-to-medium term; for the thousands of smaller AI firms, it’s even harder to see where the revenue comes from.

But there are also things that aren’t bubblelike. When we talk about the Dot-Com Bubble, we aren’t talking about the whole internet as a bubble, just a specific form of, predominantly e-commerce, that was gassed up by speculative, overexcited investment. Similarly, the South Sea Bubble was not the pricking of the maritime trade bubble, just a specific monopoly. The argument thus follows that AI is too big to be a bubble. There will be bubbles within it — like an Aero bar — but AI itself is foundational. LLMs as an educational aid, for example, might lead to a bunch of broke investors, but the train of machine learning and development will keep chugging along. AI’s proponents are always quick to say that AI is the new internet — a fundamental new architecture for technology — and, if that’s true, there’s no way it can be a bubble.

All the same, I look at AI and these bubbles through history and I see more similarities than differences. What causes a bubble to pop? Usually either the water evaporating or a finger poking it, causing the molecules to shrink and the water to escape. i.e. the cause is either endogenous and a flaw in the product itself, or exogenous and outside of the bubble’s control. Often the panic is precipitated by the fact that people want to get their money out of an asset, either because they have lost faith in that asset or because they have a newfound need for their money. I do not see either of these possibilities as plausible for AI: people who proselytise for the AI technologies are unlikely to be convinced to change their mind on its potential import, nor are the sorts of companies and funds investing in AI likely to need to suddenly pay an unexpectedly large electricity bill.

No, the current state of AI doesn’t really look like a bubble to me. Instead, I would liken it to a balloon. A helium balloon, maybe in the shape of SpongeBob. Bubbles are defined by their frailty, by something innate within the asset that makes it vulnerable to complete implosion, whereas AI is a strong asset. It is an investible proposition, and one that could be hugely profitable. Yes, there is the threat of it going POP! but it’s a controllable threat. A balloon shouldn’t burst unless it is deliberately mismanaged, whereas a bubble is pre-ordained to burst from the moment its blown.

But where AI is really like a balloon is because the real danger is not popping, but disappearing off into the sky. The greatest threat to AI is, for me, neither a weakness in its conception nor the threat of regulatory wing-clipping, but the risk that AI poses to itself as its growth becomes exponential. Because a technology that can self-replicate, can iterate and improve itself, can even develop adjacent (and less adjacent) technologies, isn’t one that fits easily with a modern view of investment capitalism. As an investor in an AI product, I would want to have a clear idea of, not just the technology’s market position, but also how it will be scaleable, reaching towards profitability, and avoid gazumping itself. I do not want to invest money in a technology which has a brain that is focused on things other than “revenue generation”, especially when the industry seems uninterested in bending that brain to its will.

And then there’s the question of what happens as more AI is trained by AI. We are already witnessing this with Large Language Models (LLMs), where chatbots are starting to become trained on text generated by AI. Input shit, output shit. It is a form of cannibalism that produces bad results, and results that could become exponentially worse. Again, these are impacts of ubiquity, rather than flaws in design. No doubt top AI bods are already working on solutions but the risk is the same: you reach a point where the genie exits the bottle and the technology becomes too widespread for capitalism to stamp its impression. Instead the technology becomes omnipresent, like the architecture beneath the internet, and a quarter of a trillion dollars in investment disappears into the ether, SpongeBob streaking away into the sunsetting sky.

If the South Sea Bubble came about because it was a new era of globalisation, with uncharted frontiers suddenly available for exploitation and colonialism, then the Artificial Intelligence SpongeBob Balloon is guided by a similar logic. So much of the current question about AI stems from the unknown. The rush to throw good money after bad in pursuit of this project is, perhaps, a so-called “market delusion”, but it’s also a rational answer to an irrational question. What is the impact of AI going to be on human civilisation? There is no-one on earth who can answer this question with more than a sliver of confidence. Money is a rationalisation of the uncertainty of this question, a way of commodifying the unknowable. Cents and dollars are a metric we can all parse, even if the philosophy of both technology and market remains elliptical.

“Fortune’s always hiding,” the West Ham fans sing. “I’ve looked everywhere. I’m forever blowing bubbles, pretty bubbles in the air.” I sometimes hear this song — sung by 60,000 odd predominantly burly men — and then look over and see someone who sits near me and spends much of the match checking the price of Ethereum in his crypto wallet. The temptation to infalate things — bubbles and balloons — is exactly because of this impulse to find that elusive fortune. Money, success, happiness. All of these bubbles are born out of a belief that there’s a quick and easy way of becoming a better, more successful you. Just don’t get carried away.

Interested in sceptical thoughts on technology? Give my podcast, The Ned Ludd Radio Hour, a quick listen.