The Market Path of AI

As with all forecasts of the future, most people know absolutely nothing, and those who do “know” are often wrong. Economic forecasting is perilous. Technology forecasting is tricky. Combining the two is perilous and tricky.

There have been some famously dire predictions. In 1998, Paul Krugman said, “By 2005 it will be clear that the internet’s impact on the global economy has been no greater than the fax machine.” In 2007, Steve Ballmer: “There is no chance of the iPhone ever gaining significant market share.” Now it’s my turn. At the risk of looking like a right plonker (and the internet never forgets), here’s my take on the market path of AI. Will AI lead to the greatest economic boom ever seen?

The economic boom side of the argument has history on its side, so let’s do a bit of history. The Industrial Revolution, driven by technological changes and supported by an economic system that enabled growth, brought about massive economic prosperity and productivity. Key technology drivers included new materials such as iron and steel; new energy sources such as coal; new machinery like the steam engine and power loom; the creation of factories, and the application of science to industrial processes and business in a way that had never been seen before.

The Internet Age, driven by technological changes and supported by an economic system that enabled growth, has brought about massive economic prosperity and productivity. Key technology drivers include the World Wide Web, increased prevalence of open source software, substantially increased volumes of structured and unstructured data, mass communication and the application of the scientific method to industrial processes and business in a way that had never been seen before.

I don’t know exactly what form AI diffusion will take, but I think the following is a pretty safe bet as a broad outline. Capital investment into AI will continue apace as market participants drastically increase.

Barriers to entry to building and deploying AI systems will be reduced. As new AI pioneers disrupt industries, agile incumbents will adapt well, and others will struggle, eventually succumbing to market pressures and going out of business. AI will present economic opportunities across the entire supply chain: whichever industries we think will be at the most risk of disruption, there will be more.

But not all industries will be disrupted equally. AI adoption will vary significantly by industry and by country. Yes, innovation is diffusing more quickly over time, but if you travel the world (and you should), you will see innovation is patchy.

Ultimately, whilst capital investment and tech visionaries determine the early stages of the AI industry, consumer demand sovereignty will reign supreme and will take over.

AI will be demanded on the production side as firms increase productivity and reduce costs. AI is a step change in automation infrastructure, but most technology companies have used statistical inference, machine learning and deep learning for years. The application of deep science in industrial processes is remarkable and something to be awed by.

We’d see some sort of S-curve of innovation. AI innovation will increase rapidly before tailing off as the industry matures and innovation gets harder and harder.

Whilst all this happens, regulatory pressure will increase. Overbearing and misguided regulation may eliminate potential benefits of AI. Whilst consumer protection regulation is welcome and often well-meaning and noble, consumer protection regulation often manifests as incumbent protection regulation by erecting insurmountable barriers to entry; regulatory capture is a persistent risk. Good regulation exists: airworthiness directives and the Bank of England having the secondary objective of facilitating effective competition between firms are two examples.

Existing companies can react very badly by resurrecting regulatory barriers to entry; they can react badly by doing nothing; they can react well by adapting and pivoting their business model. See how Apple and Microsoft moved from the personal computer to the Internet Age well. Speaking of incumbent companies, AI is, perhaps, slightly different in that the existing large incumbents (Google, Meta, Microsoft) will also lead the charge into AI, so the market dynamics of AI are fascinating. Microsoft spent $3 billion on Nvidia’s AI chips in one quarter in 2023!

After a few years, the previously new AI firms are now AI incumbents. They will provide a platform for other firms to build new markets and product extensions; new businesses will crop up. When once the new companies were the darlings of innovation, they’ll be (wrongly) pejoratively known as “Big AI”.

Eventually, as competitive pressure bites, the market will see a consolidation in the number of AI firms. At this point, it’s worth pontificating whether we’ll be able to pre-determine which firms will be successful. I argue yes, but it’ll be mightily difficult and require extensive industry knowledge. Even the most acute observers will have a low hit rate. As AI market consolidation occurs, the market will become an oligopoly and display oligopolistic tendencies while still driving vast increases in global optimisation. These big players will fight to become an AI super conglomerate. Products will be of high quality, but interdependence will occur, innovation will slow, (financial) barriers to entry will be increased, and there will be a small number of price setters. A long tail of AI firms will also compete over specific niches.

Through mergers and acquisitions and bankruptcies (providing lots of welcome revenue for consultants), the market will consolidate around a few players. Those few big players will be the ones which focus on the customer: they will get the product, the technology and the business model right and will generate a shedload of free cash flow.

On a macro level, the global AI market will provide opportunities for innovation worldwide. Those markets and countries which don’t lead with AI technology will soon pick it up - Grab and Alibaba are larger in Asia than Uber and Amazon. AI technology will be localised for different markets (though markets will become globally homogenised).

If we think about it in terms of expected value, there is a risk that AI will lead to massive harm and economic contraction, just like there was with nuclear. But the probability of this outcome is small; I’d go with < 0.001%. Whereas the probability of economic and industrial benefit is substantial, I’d go with > 75%. A simple expectation calculation will show a beneficial outcome (I’m not going to try to quantify the economic value of AI because it’s guesswork).

I think an underappreciated aspect of crazy successful entrepreneurs is that they have both the technological and engineering acumen and the economic and commerce acumen. Musk knew he had to make rocketry cost less through reusable rockets and how to engineer that; Gates knew how to code and build a software business model; Bezos is a computer scientist but intricately understood the unit economics of retail and online commerce.

The first AI super conglomerate will need to get the technology and economics of AI right. Which naturally leads to the question of how to do that. To prove that I’m not a hopeless writer (say what), I’ll leave on a cliffhanger, and we’ll revisit that question later.

Thanks for reading, I’ll see you next time!