Is AI in a bubble? A historical perspective

Ever since the public release of ChatGPT in November 2022, excitement around the possibilities of artificial intelligence (AI) has grown in both financial markets and boardrooms across the globe.

Billions are being spent on building out data centres with the latest high-performance computer chips. Salaries for expert AI programmers have become mind-boggling. Valuations for companies associated with the AI boom have grown both in public and private markets.

Understandably, it’s led to the obvious question: is the AI boom economically justified or merely another bubble in the making? If it’s a bubble, how close are we to the top and what will be the potential aftermath? While colleagues have explored the contours of the current boom from both an equity market and credit perspective, this note considers the issues from a more theoretical and historical context.  

The conclusion: the AI boom is likely just the latest in a long line of major technological advances that will eventually benefit the wider economy. History suggests, however, that such advances often result in disruptive boom-bust dynamics in both financial markets and the economy before these wider benefits – and eventual corporate winners – become evident.

Given human psychology and the basic economics of technological advancement haven’t changed, there’s little reason to suggest this time will be any different. Beyond the ultimate good news of what this new technology will bring to the economy, the short-run solace is that we’re likely only around mid-way within the latest boom/bust cycle, meaning it should likely play out for a while longer before the shake-out eventually happens. But as investors we need to be prepared, and simply not view this boom through rose-coloured glasses. 

Why do new major technologies lead to boom-bust cycles?

To my mind, and after reviewing the history of past technological breakthroughs – such as railways, electricity, cars, radios and even the internet itself – there seem to be three fundamental reasons why boom/bust dynamics happen. 

Fast supply ramp up due to ‘winner take all’ markets resulting from scale and network economies

For starters, many new technological innovations offer ‘first mover advantage’ to those companies that seek to exploit the new opportunity. One reason is that due to the large fixed costs of building out the required supply infrastructure, there are benefits in getting big fast thanks to economies of scale, meaning your per unit costs of delivery fall.

This has been true in building the first railroads late last century, the poles and wires needed for electricity, car and radio factories along with the fibre optic cables needed for internet transmission. Today, there appear to be scale economies in building out data centres and developing expertise in chip design.  

A second reason is that there are also often network economies in getting to customers first – they get comfortable with your services and so are less eager to switch, and the value of a network of users grows in the eyes of each user when more are using it (also known as Metcalfe’s Law). Such network effects were most evident in the case of telephones, core computer software, online marketplaces and social media platforms.

At this stage at least, there’s little obvious network effect in choosing one AI platform over others (e.g. ChatGPT vs Gemini) apart from user familiarity or habit, but that could change as more consumer and business AI applications are brought to market.

A third reason is that tech-savvy suppliers entering a new market usually assume overly rapid adoption by the broader market. That’s because they themselves see the clear benefits and assume most others will also. They’re also lulled into early over-optimism by the initial uptake of their services by the small minority of consumers who are experimental ‘early adopters’.  

Given the alluringly high ‘winner-takes-all’ stakes at play, there’s rarely a shortage of hopefuls entering a new industry – and associated financing – despite the likely high failure rate. Much like buying a lottery ticket, the downside is capped while the upside could potentially be huge.

All up, the nature of the first mover advantage and overoptimism with regard to early rewards suggests a ‘gold rush’ mentality among potential suppliers, which may encourage excess investment and overcapacity. A stylised example of a boom-bust capex cycle is provided below.

Source: Betashares

It’s worth noting, however, this overinvestment is not due to irrationality. It’s due to the winner takes all stakes at play and the asymmetric lottery-sized returns if a company ends up among the winners.   

Buyer hesitancy and slower than hoped initial consumer uptake

Despite the rush of suppliers into a new market, history suggests demand – beyond the small core of enthusiastic early adopters – is usually slower to emerge than initially hoped. The vast bulk of consumers are not tech-savvy and take time to adopt new technologies.

Why? Early supply offerings can be expensive and not easy to use. Suppliers of existing products – at risk of disruption – may heighten consumer uncertainty by warning of safety concerns (such as having an electric current running through your home or using your credit card online). There’s usually a lack of especially useful consumer-friendly applications (e.g. spreadsheet software for personal computers) and/or supporting infrastructure (roads and petrol stations for cars). 

In marketing parlance, the time lag before demand moves from early adopters to the majority of consumers is known as the S-curve. Once the majority get comfortable with a new product, adoption can be rapid, but it usually takes longer to reach this broader market than initially hoped. 

A stylised example of the consumer S-curve is provided below.

Source: Betashares

The time of reckoning and the aftermath

Combining excessive early supply/investment and slower-than-hoped consumer adoption usually leads to an industry shake-out. Finance for the weakest of the many early hopeful suppliers is cut back and the investment build-out tapers off. Elevated valuations for listed companies offering blue sky potential are brought back to earth – and many fail.

Source: Betashares 

But the companies that survive often get to take advantage of a glut of supply capacity just as demand from the majority of consumers finally begins to ramp up. Society is eventually better off with new forms of transport, energy and information exchange. And for those producers still in the game, winners do often take all until either pro-competition laws or new disruptions limit their advantage.  

The AI boom – where are we today?

As noted in the introduction, there seems little reason to think the current AI technology boom will end much differently from those in the past – suppliers will rush to provide products and consumers may be slow to respond. There will eventually be a shakeout, with the few winners remaining and society left to enjoy the widely dispersed productivity benefits. 

But there are also good grounds to think we’re not yet in the final innings of the boom stage.

• The investment boom is still ramping up with the major hyperscalers still planning to invest even more, at this stage, for the next few years. 
• Valuations of AI-related companies – apart from perhaps a few exceptions – are not yet at nosebleed levels.
• There’s yet be a major rush of AI hopefuls listing on the share market.     

Based on the last stylised diagram above, I’d judge us somewhere between 2 and 4 on the timeline – perhaps a 3 (out of 5)!