Why doesn’t the AI of Google, Meta, Alphabet, Microsoft and Amazon excite the stock exchanges?
Has AI reached a commercial tipping point? The analysis by Julien Gaertner, Capital Group equity investment analyst
Artificial intelligence (AI) remains firmly in the spotlight after last November's launch of ChatGPT, a bot capable of producing text that appears to be written by a human. Hitting one million users in just five days and 100 million before the end of January, it's the latest example of “Generative AI,” a set of models that can create content including audio, lines of code, images, texts and videos. ChatGPT has already found many uses, from writing stories, scientific articles, jokes and applications to job offers up to composing music. Could the system prove to be the first “killer app” of the age of AI?
AI is very likely on the verge of causing a series of momentous changes that will affect various companies and industries, but it is essential for us to dig into the details to separate the reality from the hype. Although research by McKinsey shows that global AI adoption has more than doubled in the past five years, with nearly 50% of survey respondents saying they use it in at least one area of business, the initial enthusiasm seems to have met with a temporary setback probably due to the awareness of the extent of the changes required within organizations to adopt these technologies. But that hasn't stopped a flood of AI investment from businesses globally.
The fundamental function of AI is to make predictions and make decisions based on the data it has been 'trained' with. What has changed in recent years is that technical advances allow AI models to be trained on increasingly large amounts of data and thus reach new levels of functionality. With generative AIs like ChatGPT the focus, importantly, has shifted towards so-called “large language models,” which enable much better chat and text production capabilities. Early 2000s AI systems used machine learning primarily to improve analytical models; the most commercially relevant examples in this sense are the targeting of advertisements by Google and Facebook. Generative AI, on the other hand, can create new and unique content thanks to what is known as a “transformer architecture. This allows the AI to understand the relationships within a data set, such as text or images, creating the context sensitivity necessary to perform the most creative tasks.
Today we are in a situation where Generative AI systems achieve impressive results in many areas but fail miserably in others. The most optimistic forecast we can make is that the increase in the size of the models and the growth of the data with which they are trained will continue to favor exponential progress for a long time to come, and it must be said that so far the empirical evidence points in this direction. A pessimistic hypothesis, on the other hand, is that as the size of the models increases, the improvements run out, or you hit a wall represented by the amount of data available to train them.
From an investment point of view, if we start from the point of view that AI will generate large profits, at the moment this forecast seems to have little reflection in the share prices of the most important companies. For companies like Alphabet, for example, the opposite seems to be true after recent declines. OpenAI, an unlisted company founded as a non-profit in 2015, is reportedly worth $29 billion following Microsoft's recent investment and, despite its ambitious goals, has reported that this year it expects to generate revenues for only Approximately $200 million. Assuming that this prediction is correct, the valuation of the company discounts great enthusiasm.
Looking at other companies with the potential to offer similar products to ChatGPT, it's hard to see similar enthusiasm in their evaluations. Google is the pioneer of transformer-based models, and right now its stock doesn't seem to discount this $29 billion in AI assets. So-called “hyperscalers,” a small group of tech giants that includes Meta, Alphabet, Microsoft, and Amazon, have already spent billions on the hardware needed to build AI platforms and, therefore, may serve as AI architecture in the future. of other developers.
From an even more long-term perspective, in our opinion, the spread of AI can also have an important impact in areas such as the demand for semiconductors. If you ask ChatGPT a question or ask the Stable Diffusion text-to-image model to build an image, they can take a long time to respond. The reason is the extremely high computational work required for so-called inference, which involves sifting through billions of data points to produce the desired content. This tells us a lot about the very high semiconductor content within these systems. While it may be difficult for investors to identify the stocks of the companies that will eventually make the most of AI, very few make the semiconductors on which those systems are based.
We may currently be in a phase similar to that experienced by cloud computing in 2013, so that companies that can leverage AI to differentiate their offering or their sources of productivity growth could be greatly advantaged in the years to come. Beyond the obvious fields of technology and 'knowledge', potential application areas for AI include supply chain management, healthcare (in the fields of drug development and scan analysis), insurance, petroleum and gas (think of the data provided by satellites), public utility services (for network and load management) and autonomous agricultural machinery: strategies in the field of AI can therefore become an increasingly important component of the analysis of companies.
This is a machine translation from Italian language of a post published on Start Magazine at the URL https://www.startmag.it/economia/perche-lai-di-google-meta-alphabet-microsoft-e-amazon-non-eccita-le-borse/ on Sun, 28 May 2023 05:51:09 +0000.