“Between AI and the Internet bubble, the parallel does not hold” – L’Express

Between AI and the Internet bubble the parallel does not

A seat at the AI ​​table is expensive. Nine-zero lifts have become the new industry standard. In two weeks, Elon Musk’s company, xAI, completed a new funding round of $6 billion and Anthropic, one of $4 billion. A craze that makes some economists fear that an AI bubble is forming. The American BlackRock, the world’s largest asset manager, has studied the issue closely. “The first generation of AI applications risks being overrated,” agrees Jean Boivin, former deputy governor of the Central Bank of Canada who now heads the BlackRock Investment Institute (BII). However, it identifies great investment opportunities in the sector, particularly in generative AI infrastructures. And looks with optimism at the capacity of artificial intelligence “to help us innovate faster”. Interview.

L’Express: Colossal investments are being made in generative AI. Are they excessive?

Jean Boivin: The situation is indeed unprecedented. The transformation linked to AI is on a scale comparable to that of the first industrial revolution. But the speed at which this transformation is occurring is unprecedented. To benefit from AI, we must build adequate infrastructures, “intelligence factories”. The aggregate investment in AI does not seem excessive to us in relation to what needs to be built. This does not mean that individually, all companies in the sector make informed and justified decisions. In AI, we are already observing a “winner takes all” phenomenon: the leading company wins most of the market. Anyone who thinks they can win the race is therefore encouraged to invest massively. But not all of them will reap sufficient profits to justify their investments.

READ ALSO: A button to stop a runaway AI? “Kill switches” fuel fears and fantasies

Do you think the industry is getting ready to go through its “Internet bubble”?

Some are asking this question because the valuations of companies and American stock indices in general have reached record highs. They are today at levels similar to those reached just before the bursting of the bubble in 2000. High valuations are also concentrated in one sector, as in 2000. But there are major differences. Profits increase in line with valuations. That of Nvidia, for example, has risen spectacularly but so have its profits. If we analyze all the indicators – balance sheets, income, etc. – we see that the health of businesses is much more solid than in 2000. The parallel with the Internet bubble does not hold.

A recent study that you conducted on AI shows that great uncertainty reigns around the impact of this technology: which areas will benefit from it, through what uses, over what time frame… Why is it so difficult to evaluate ?

This uncertainty is linked to the speed with which the AI ​​revolution is occurring. When search engines emerged, there was also a period of uncertainty around their revenue model. We wondered if they would move towards subscription or advertising. The answer is now clear. But this was not obvious from the start.

READ ALSO: Stuart Russell (Berkeley): “The capabilities of generative AI have been overestimated”

In artificial intelligence, we do not yet know who, of those who build it or find a specific application for it, will benefit the most from it. The fate of agriculture after the Industrial Revolution offers a useful lesson in the age of AI. It is one of the sectors that has improved its productivity the most thanks to mechanization. But its weight in the economy fell from more than 50% in 1850 to less than 2% in the 1980s. This reminds us of a very important point: the fact that a sector benefits a lot from a technology does not guarantee that it is he will get the most income from it in the future. In portfolios, we believe that the speed of transformation linked to AI and the uncertainty over the profits it can generate will give a premium to active management.

How can we navigate intelligently in this changing landscape?

Our study identifies three phases of AI: that of construction, then adoption which will perhaps be followed by a phase of transformation. We are convinced that there are great investment opportunities in the construction phase of AI infrastructures, which gives an important role to private markets. We do not yet have enough information to have a strong conviction on all potential future applications. The first generation of AI applications risks being overrated. We tend to analyze AI with poor reading frameworks, trying to identify which tasks will be accelerated by AI. This is making the mistake of “looking for the key under the lamppost”. The main capacity that will undoubtedly be increased by AI is not a specific task but the ability to innovate. AI can help us innovate faster. The 2024 Nobel Prizes suggest so. If over the next few years, the number of patents filed increases significantly, this will give weight to this thesis.

What impact could it have the election of Donald Trump on American AI heavyweights and where does China stand compared to them?

AI is the subject of a global race, it has become a geopolitical issue. No one in the United States has any interest in hindering progress in this area. Neither the outgoing administration nor the next one. That of Donald Trump will perhaps be even more careful not to put barriers in this sector. As for China, given the speed of progress in the field, we can consider that there is no significant gap between it and the United States in the short term. These two powers are in a position to dominate this technology, there are no other credible players in AI today.

READ ALSO: How hackers are infiltrating the financial sector everywhere

What penalizes Europe the most in this area?

One of the main obstacles is the nature of its capital market. Capital is able to mobilize very quickly in the United States to invest in the construction of the vast infrastructures necessary for the development of AI. In Europe it takes longer.

The cost of AI is currently high. Can it drop significantly?

The current construction phase will likely be inflationary, as it puts pressure on the resources needed to build AI, such as energy. Once the infrastructure is in place, economies of scale should be found and the marginal cost of AI could turn out to be very low. Data centers currently represent between 4% and 5% of total electricity demand in the United States and between 1% and 2% of global demand – a proportion that could double, even triple or quadruple in the coming years according to some estimates. It remains to be seen to what extent innovation in the energy sector will reduce the cost of producing renewable energy. The most optimistic believe that AI itself will unlock a lot of innovation in this area, but it is still too early to tell.

.

lep-life-health-03