Since the beginning of 2023, NVIDIA’s stock has risen by almost 900 percent at its peak. Since the release of ChatGPT in November 2022, AI technology has been hyped to an unparalleled degree. “AI will save the world,” wrote the well-known venture capitalist Marc Andreessen ((in his essay: “Why AI Will Save the World”, June 2023)
The GARTNER Hype Cycle (cf. TeaserPic for this blog post) describes the typical curve of just such technology hypes. The question now on the minds of investors and innovators is: Have we reached the “peak of inflated expectations”? Is there an AI bubble? If so, is it close to bursting?
These are the exact questions I will delve into in this blog. To start, I am relatively confident that we will only see a moderate correction overall. Additionally, I don’t anticipate such a development happening before early 2025.
Enormously high investments in AI infrastructure
The investments in AI are indeed enormous: Especially the big tech players, but also other companies, will spend over 1 trillion dollars (!!) on AI investments in the coming years: Google wants to invest 50 billion dollars in the necessary infrastructure by the end of the current year. Amazon just announced that the initially estimated 60 billion dollars will probably not be enough. The technology companies are investing primarily in AI chips, and many are striving to build a “supercomputer”. Microsoft for example, is developing a supercomputer with an estimated cost of over 100 billion dollars, consisting of millions of chips, in cooperation with OpenAI.
The costs for training AI algorithms are also skyrocketing: A study by Epoch AI postulates that the costs double every nine months. This exponential increase in costs means that the development of new models could soon exceed the sum of $1 billion, if you factor in electricity, hardware and employee salaries. Whether this trend will continue in the next decade is uncertain.
The economic newspaper Handelsblatt features a clear graph that vividly illustrates the skyrocketing costs of training LLM models:
Low-yield business models
These costs are contrasted by relatively weak business models. The WOW effect users experience when generating texts, images, and music doesn’t translate equally into economic benefits for many providers.
So many companies do not yet have a robust business model. According to a survey by the consultancy Deloitte, a quarter of companies find it difficult to identify use cases for the technology at all. “Until recently, this hardly concerned anyone—in fact, quite the opposite. The consensus was that profitable AI applications would naturally emerge, just as Amazon, Google, and Facebook followed the rise of the Internet.” notes economic newspaper Handelsblatt in its analysis The concern over an AI bubble bursting
Specifically, for the company that sparked the AI hype: According to a report from the tech magazine The Information, OpenAI is projecting a loss of five billion dollars this year, while its annual recurring revenue stands at just 3.4 billion dollars.
As a result, investments and medium-term earning prospects are diverging. At the end of July, Microsoft CFO Amy Hood stated that their multi-billion-dollar investment in AI infrastructure would pay off over the next 15 years “or beyond”—an exceptionally long and unusual amortization period.
The concern over an AI bubble
In the recently published post AI’s $600B Question (06/20/2024) the author David Cahn of Sequoia Capital calculated that AI companies would need to generate $600 billion in revenue, to justify the current level of investment in data centers in order to put the current gap between investments and revenues into perspective. Other big names such as Goldman Sachs are also adding fuel to the fire.
The market is already getting nervous. At the beginning of August, the NVIDIA share price was down 20 percent before recovering again. For the sake of completeness, reactions to developments in the Ukraine war, the Gaza conflict, the U.S. election, and the shift in interest rates are also currently influencing stock price movements. Not all fluctuations in share prices can be attributed to the AI investment hype or concerns about an AI bubble.
Here is the share price of two companies that have clearly benefited from the AI hype. (Retrieval of price data from 02 September 2024)
NVIDIA share price over the past 3 years:
Alphabet share price over the past 3 years:
Where the benefits of AI are already translating into profitable business models
Investors are primarily focusing on major tech players, like OpenAI, who are making massive investments in LLM algorithms. However, they are far from commercially naive. For instance, Aleph Alpha has adjusted its strategy to align with these commercial realities to achieve profitability in the medium to long term. There’s no need to accuse tech leaders of being overly fixated on the future without a pragmatic view.
In addition, the benefits (and commercial potential) of AI can already be seen in numerous areas. Below are some examples, and the recommendation to listen to the HANDELSBLATT podcast with KLARNA founder Sebastian Siemiatkowski. The claim: “With AI, we can reduce the number of employees from 3400 to 2000”. Click here for the podcast: Pailot has developed a powerful algorithm specifically for optimizing production processes (see also THIS BLOG HERE ). The added value is compelling: for one customer (a plastics manufacturer), output increased by 17 percent, a machine manufacturer saw a 28 percent reduction in production delays, and a sensor manufacturer boosted plant capacity utilization by 10 percent.
And despite all the persistence of the challenges with the “hallucination” of LLM models and the lack of robustness of statements, the AI scene is making major and minor progress throughout: Video production has gained momentum in 2023. The company DeepMind predicted novel material structures at the end of 2023: Google DeepMind researchers have used AI to accurately predict the structures of more than 2 million new materials, which has significant implications for renewable energy and computing. And with AlphaGeometry a breakthrough in the mathematical capabilities of AI has been achieved: In a groundbreaking paper published in magazine Nature, AlphaGeometry demonstrated its ability to solve complex geometry problems at a level comparable to that of a human Olympic gold medalist. AlphaGeometry solved 25 out of 30 Olympic-level geometry problems within standard time, a feat comparable to top human candidates, and solved a wide range of math and computer science questions more effectively than human mathematicians working alone.
My conclusion
In the analysis by Goldman Sachs (see note above), Jim Covello, Head of Global Equity Research at Goldman Sachs, stated that the bursting of an AI bubble is not unrealistic; however, the AI hype could still continue, “either because its promises are fulfilled or because it takes a long time for a bubble to burst.”
And at the beginning of August, META CEO Zuckerberg declared: “At this point in time, I would rather risk building up more AI computing capacity than necessary than be late”. And with that, both get to the heart of an important characteristic of this AI wave:
It is the reality that new technologies are inherently unpredictable by their nature. No amount of forecasting or numerical analysis in the 1990s could have drawn a direct line from the early days of the Internet to applications such as Uber and Instagram. It is therefore not irrational to maintain high expectations of AI technology. And the following applies to this technology: It is a technology with a true game-changer character; both for the economic world and for geopolitics (see also my blog on the book “Superintelligence” by Nick Bostrom). No country of relevance can and does not want to afford to drop out of this technology race (or: marathon) at the moment: not China, not the USA, and not even Europe (with promising players such as Mistral and Aleph Alpha in the LLM sector).
I have a bet with a colleague that the AI hype will last at least until Christmas this year (I’m standing by this bet), and I’m quite confident that I’ll win.
