Productivity increases do not always generate increases in profits. We do not know when the economic boom generated by AI will be over, but the overstretched industry is increasingly looking shaky.
When I last wrote about financial booms and busts, I equivocated over whether the next financial bust – there will be one, we just don’t know when – would be precipitated by a collapse in the cryptocurrency market or the AI market (or perhaps something else). I’ve come to the conclusion that despite their $US2+tr magnitude (an increase of about $US130b a year since bitcoin started), the cryptocurrency markets are too peripheral from the core financial system for their collapse to threaten the core institutions. Most of them have not sufficiently trusted the cryptocurrency market to excessively expose themselves to it.
On the other hand, there are increasing economic concerns about the Artificial Intelligence industry (AI). (There are also non-economic concerns not covered here.) They are not about the long-run effects of AI on the economy, although there are predictions of more than the technology will deliver or earlier than it will deliver – perhaps to enhance the hype which is driving the market. The following analysis does not doubt that AI will have considerable impacts on the economy in the long run.
Rather, the concern is that a productivity increase – even a large one – may not always generate long-run profits for the innovator. That was Adam Smith’s point. The energies of his butcher and baker ultimately benefited consumers. Firms may lift their productivity, but where there is competition the benefits go to consumers. Yet each firm must, for if they don’t they will be drowned by firms that do.
A good example is the farm industry. Farmers are continually trying to lift their performance, but when they all do, consumers get lower prices, while farms that don’t bother go bankrupt. This does not mean New Zealand farmers would benefit if they gave up innovating. Offshore farmers are innovating too and that will drive down the international prices of farm products. (In any case there is no way we can stop our farmers striving.)
The strivers may benefit in the short run before the market settles down, and a monopoly which boosts its productivity may be able to retain some of the proceeds. But currently no AI firm is a monopoly. Every innovation by one is soon matched by others. The Chinese AI innovators are only months behind the US ones. Every firm has to keep innovating in order not to end up a laggard.
The AI market is further complicated by the free and ultra-cheap programs which most personal consumers seem content to use, even if they are not as advanced as the frontier products. The real money will have to be made in selling to businesses.
The evidence is that, with some exceptions, businesses (and bureaucracies – the military are likely to be guzzlers) have been taking up commercial AI programs very cautiously. I won’t go through the data, which is fragmentary, nor the anecdotes, which are numerous. In summary, it seems that AI requires a cultural shift within firms and that is happening only slowly. Recall that it took decades for manufacturers to take up the opportunities that electricity offered them. (If I had to bet on who will make a profit from AI, it will be consultants who help firms accelerate the cultural changes – not be simply offering them fancier AI programs but also training staff to use those programs effectively.)
A realistic assessment of the prospects of the AI industry is that it will get its revenue from businesses but not that quickly and quite possibly not enough. Probably there will be little ‘supernormal’ profit (the abnormal profits which monopolies make) because prices will be driven down by competitors including Chinese ones and free and ultra-cheap AI models which are not on the frontier. Yet each AI firm must keep progressing the frontier to remain competitive.
To do this, the market leaders are heavily investing. One estimate is they will spend this year around $US900b on chips, data centres, power generators and so forth, with $US1.4tr expected to be spent next year. (To give a scale, recall that the cryptomarket has grown an average of $US130b a year for the last 17 years – a tenth of that.)
Some of the funding from the expansion will come from revenue, thought to be about $US150b to $US200b this year, some from borrowing which may be around $US400b this year and the rest – more than $US200b – from equity investment; perhaps the majority coming from private equity firms (some of which may be borrowing). I have not seen similar projections for breakdown of funding the next year’s $US1.4tr. It is unlikely that the extra will be covered by a surge in revenue.
Not incidentally, lenders are likely to secure their advances on tangible assets like chips, data centres, power generators. The equity injections are largely, based on the investors’ hopes of profit. It follows that many are likely to be disappointed.
Presumably, the more optimistic will continue to invest until they run out of funds. Just how long will they will remain optimistic? Long enough to fund the industry to the time when hoped-for revenue is generated? Allow some scepticism.
If I am right, the industry will get into financing difficulties. Lenders may be unwilling to continue to provide their share of the funds required for investment; some may want to pull out. It is unlikely that there are unlimited funds for the equity investors. At some stage one of the big AI firms may have to say they are putting off planned investments.
The market will hiccough and confidence among equity investors will fall. That will impact on the confidence of lenders. Sure, some will have the tangible assets if anything goes wrong, but when there is a downturn, they may not be able to realise their assets’ value. The AI boom may be over, but AI will still be impacting on consumer behaviour and firm productivity.
This column is not very different from what informed commentators are saying, except they stop a bit earlier, leaving the readers with the thought of. Stein’s law: if it cannot go on forever, it won’t.
If AI does not generate sufficient revenue there will be a collapse in physical investment with a resulting downturn from general economic activity in the US – including rising unemployment – which will impact on the global economy. (I leave others to write about the impact of the downturn on US politics.)
However, we have only an imprecise understanding of the financing of the boom. When there is a bust, lenders and investors in AI firms will experience a reduction in their wealth and will cut back their consumption and investment activity, reinforcing the economic downturn from the reduction in investment. We do not know the degree there is over-leveraging (borrowing too much). It is over-leveraging with its secondary impacts on financial institutions and innocent parties which makes financial crises – exceptional disorder in financial institutions – so painful. A concurrent financial crisis compounding the economic downturn would be nasty.