One of the key scholars of the link between technology and finance — a major preoccupation for private asset investors today — is Carlota Perez, a British-based Venezuelan scholar whose work focuses on the ways in which technological innovations, finance and economics work together to produce great waves or periods of investment and, periodically, prosperity.

Perez’ work (much of which is on her website) is best explained by the title of her well known book ‘Technological Revolutions and Financial Capital: the Dynamics of Bubbles and Golden Ages’, where she describes the phases that accompany technological innovations as they acquire investment and begin to ripple through economies, and change the structure of those economies. In common with scholars of developmental waves she maps how major technological revolutions (such as industrial revolution, steam engines, railways, manufacturing and internet) develop in phases.

Financial vs production capital

The principal phases she outlines begin with the Installation phase, where capital rushes towards the new technology, fuelling the build of new infrastructure and new sub-industries and economic activity, but also driving asset bubbles which inevitably crash or hit a ‘turning point’ as she politely puts it. The next phase is ‘Deployment’ where the new technology is widely used, and public policy, regulation and society accommodate it. Investment-wise, the economic applications of the technology enter into the mainstream. Readers will quickly spot the relevance to the AI capital expenditure boom.

An interesting distinction Perez makes, which I think is highly relevant, is that between ‘financial capital’ by which we can understand fast moving speculative money, and ‘production capital’ which is devoted to building of specific technologies, skills and infrastructure and is by nature more ‘committed’. I am tempted to say that Europeans might hold that ‘financial capital’ is American and ‘production capital’ is European, but Europe has a deficit of both.

For private asset investors this distinction is critical because it delineates the often frothy, speculative behaviour of public markets (witness the collapse of the fund Situational Awareness¹, and the extreme volatility of the South Korean KOSPI index), from the economic contribution of ‘production’ focused capital, much of which comes from private investors.

AI race

Reading Perez’s work today, it seems to me that the difference between AI and the historic phases of technological innovation mentioned is speed. The speed with which models are improving, the speed with which capital is being deployed, the take-up by users, and the speed with which the winners and losers are emerging at the corporate level.

A couple of months ago, we wrote about the mixture of wonder and panic created by Anthropic’s Mythos model. Then, it was assumed that Chinese developers were some seven months behind the Americans, but last week, Moonshot, the Chinese AI developer released test results of its Kimi K3 model (conveniently, they are in the middle of a fundraise), which apparently match some of the capabilities of Mythos. There has not been much of a market reaction to this in the US but, the red-blooded capitalists of Silicon Valley and the AI industry are crying alarm, running for Washington and calling for the US government to ban Chinese models. Here, there is a strong undercurrent of US-Chinese competition, a phenomenon that runs through the private capital world.

Their dismay, consistent with the Perez framework, portends a speedy evolution in the structure of the international market for AI tools. China’s models are much cheaper than those in the US, partly because they piggyback on work done by the Americans, partly because of cheaper energy in China, and partly because of fierce competition amongst Chinese developers (and the winners get a boost from the Chinese government). In the near future, those who want cheap, cheerful and clever AI tools may well gravitate to the Chinese AI products (as Germans are now doing with Chinese cars).

The American AI models will still be useful for a number of things. The first is that they will continue to lead frontier AI development, in analytical power, and increasingly on specialised datasets and applications. Secondly, they will be favoured by Western corporates, especially in closed systems, and thirdly, they will continue to have a strategic role – either in applications across nations (from security, warfare or healthcare for example) or as strategic exports (here I am thinking of G42, the UAE’s AI firm that has an exclusive relationship with US AI models).

To that end, capital will flow more to cutting edge AI, sovereign AI and secure corporate AI in the US, and very likely in Europe (Mistral, Europe’s last hope is currently raising capital at a measly €20 billion valuation²). In my view, we are still some way away from a ‘reckoning’ in AI stocks, and proof of the hubris in the market is that the large investment banks involved in the SpaceX IPO continue to issue research notes, with embarrassing justifications for their lofty price targets (Morgan Stanley thinks SpaceX, which trades below $110 is worth $300³).

Still, markets are beginning to sniff the weaker players – Oracle, whose balance sheet is far weaker than rival hyperscalers, has seen its value halve in the past three months.⁴ Writer Sebastian Mallaby, amongst others, has stated that there is a high risk that OpenAI might run out of money.⁵

For the time being, the show goes on.