How many investments will there be for artificial intelligence
All the economic objectives of Microsoft, OpenAI and beyond on artificial intelligence. Analysis by Mark Hawtin, Investment Director of GAM Investments
During the nineteenth century massive speculation in railway shares in the UK, which peaked in 1846/1847, led to a bubble in the industry. The belief that railways would revolutionize transport and trade, together with the availability of cheap credit, favored the rapid development of the railway network, which extended from 160 kilometers in 1830 to around 2500 in 1840, reaching 9600 kilometres. in 1847, the largest railway network in the world. The amount invested, i.e. 25% of GDP, corresponds to the exorbitant sum of 4 trillion dollars today, according to Andrew Odlyzko who talks about it in his book The Railway Mania of the 1860s and Financial Innovation.
PREVIOUS
At the height of development, hundreds of new railway companies were formed, many based on unrealistic plans for lines that were never built. Investors invested heavily in rail stocks, driving prices to unsustainable levels. The bubble began to burst in late 1845 when a combination of factors cooled investor enthusiasm. These include a series of high-profile rail accidents, concerns about the financial viability of many new rail companies and rising interest rates. The bursting of the railway bubble caused a financial crisis in the United Kingdom, and many investors suffered huge losses. However, the bubble also had a series of positive long-term effects, including the development of a national railway network which fostered economic growth.
It is a historical phase that was repeated with the Internet boom and then the crisis of 1999-2002. The investment cycle at the time focused on fiber optic cables, with the aim of increasing capacity in view of the demand for internet services. Again the capital expenditure was of epic proportions. In 1996, fiber optic cables in the United States spanned 1.6 million miles. By 2000, according to the Federal Communications Commission, the network covered 10 million miles, and new companies like WorldCom and GlobalCrossing went into debt to finance its construction. When WorldCom went bankrupt in 2002 it had $100 billion in debt, Global Crossing had $25 billion in debt. The network usage rate at the time, according to TeleGeography, was only 20%, and in 2010 it only reached 30%. Again, irrational enthusiasm generated huge investments and a mountain of debt to build the networks of the future. As with the railroads, although the irrational approach caused heavy losses among investors, the infrastructure that was developed allowed the Internet to take hold. Short-term losses often allow disruptive companies to make long-term gains.
HOW INVESTMENTS IN ARTIFICIAL INTELLIGENCE WILL GO
Today we therefore ask ourselves whether the significant increase in investments in infrastructures for artificial intelligence will end up like the investment cycles we have mentioned. We believe not. The impact will be less profound, especially on stock prices, however there are reasons to be cautious in the short term. The introduction of easy-to-use interfaces, such as Bard and Chat-GPT, make artificial intelligence applications available to everyone. This catalyst has led to increased investments in infrastructure, starting with the need for GPU graphics processors supplied by Nvidia. This year's Q1 earnings report shows earnings in dollars never before achieved by a company. Thus began the “arms race” in the field of artificial intelligence, while demand rose. Nvidia is planning a large increase in production capacity and, according to CEO Jensen Huang's forecasts, the company will invest one trillion dollars over the next four years to enhance data centers for artificial intelligence. This trend is confirmed by the Dell'Oro Group which expects fixed investments in data centers of 500 billion dollars in 2027. By comparison, investments in the car and truck sector amount to $33.4 billion per year (Source: Wikipedia).
These are huge numbers and, in reality, no one really knows whether this capability will be used or not, or how quickly, but there are fears that the importance of artificial intelligence is such that not investing would mean missing out on a huge opportunity in a technology with revolutionary effects that could be more important than the Internet 10-15 years ago. However, these investments must generate a reasonable return. Open AI aims to achieve revenues of $1 billion, while Microsoft hopes to achieve revenues of $10 billion with its Copilot product. These are small figures compared to the investments that have been made. In a recent article, Sequoia Capital wrote that current GPU sales of $50 billion per year would need to produce at least $200 billion in application revenue to justify the investment. We are undoubtedly far from this result.
In recent research, Bernstein tries to frame the size and scope of investments in AI infrastructure, as indicated in the figure. The projected composite annual growth rate (CAGR) of 75% from 2022 to 2025 is absolutely unprecedented in IT infrastructure cycles and would correspond to growth equal to the overall data center server market that exists today in 2025. That seems like a pretty ambitious goal. It's worth considering that, according to Bernstein, the average composite growth rate of the server market has been 3% in just the last 25 years.
CAPITAL WILL STIMULATE THE GROWTH OF ARTIFICIAL INTELLIGENCE
We believe that abundant capital, often from mega-cap, cash-rich, technology-dependent companies, will spur growth well ahead of demand, regardless of potential shortages in early-stage applications. This is not an entirely negative factor, considering that artificial intelligence will allow companies to significantly improve productivity, but it will likely open up a difficult phase for infrastructure providers. We have already seen this with data center providers in 2022/2023 when there was often talk of “data center optimization” to mask the lack of new capacity investment. We may see a sharp decline in demand for GPU chipsets for a quarter or two as existing capacity is absorbed. There could therefore be a difficult period ahead for the shares of infrastructure providers such as Nvidia. As a result, we believe that investment objectives should focus more on the users of AI infrastructures, rather than their builders. This affects several sectors, including healthcare, transportation, retail, financial services and industrials.
This is a machine translation from Italian language of a post published on Start Magazine at the URL https://www.startmag.it/innovazione/investimenti-intelligenza-artificiale/ on Tue, 26 Dec 2023 06:08:29 +0000.