Generative AI

Why such AI buzz now?

Artificial Intelligence (AI) has been around since the 1950s, but the recent burst of interest in the topic is driven by the explosion in computational power and data. This has led to an acceleration of AI based applications like large language models (LLMs) which have seen large-scale adoption from individuals. For example, the adoption rate for ChatGPT is 10 times faster than what was seen for Instagram. We would expect Generative AI as a megatrend to persist considering the ever-increasing amount of data in the universe as a result of the “internet of things” as well as generational factors, with younger populations having a positive attitude towards the topic. This mega force has started creating specific investment opportunities across geographies and sectors and will drive structural changes over the long-term. Everyone is scrambling to understand these dynamics.

AI most notably will act as a catalyst for productivity and growth. This is especially relevant to counter the impact on growth and inflation of other megatrends like aging populations. Three key macro themes related to the topic to keep in mind are: 1) We should not assume that the economies of countries that generate the most technological innovation will necessarily see the largest productivity gains, because those will come from adoption, not invention; 2) There will be a broad impact on the economy across sectors as a result of AI. For example, it is estimated around 80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of large language models¹; and 3) The idea that Generative AI is part of a broader technological innovation that is likely to deeply transform our systems of production. That said, we need to understand the interplay of AI with other technological innovations, in particular, 3D printing and blockchain.

Adoption of AI into investment processes

Across our Fundamental Equities platform, a large part of our existing investment process consists of getting information from text documents, reading company filings, earning transcripts, and news. AI has the potential to be transformational, in particular, using it to more efficiently ingest and summarize this type of information. Though this makes for better research capabilities, we do not see Generative AI as being able to substitute the work of an analyst or having the potential to generate an independent investment thesis. Similar to how spreadsheet software has changed the way we have incorporated financial modeling into decision making or the way search engines have optimized the way we access information, AI will act as a catalyst to enhance human decisions rather than replacing them. We are cautious about the limits that the technology can have. Experimenting early is the best way to determine the ways that Generative AI can be both a benefit and detractor to our processes.

Our Systematic Active Equities platform also views exploration and experimentation as key to understanding the capabilities of AI and has been focusing on this through three areas: 1) Newer waves of large language models are better equipped with data, including knowledge in economics and finance. We have been using these language models as an assistant to help researchers and portfolio managers drive actionable portfolio views; 2) LLMs now have non-trivial capabilities across languages, and many can generate texts in dozens of languages. Historically, we needed to invest in specialized machine learning pipelines for specific languages. Now, we have the ability with a single model to parse multi-lingual sources to standardize data; and 3) We have better insights into interpretability. We now can ask a model for a short rationale to indicate why a reaction may be positive or negative rather than the traditional black-box model that takes data in and would just spit out a number or a prediction.

Every company in the world ideally wants their own generative AI model, but there will be limits to adoption, with the key concern being information security. Companies needs their own proprietary data enabled through systems that are set up in such a way that AI cannot communicate externally data which is considered internal to the company. This creates great opportunity for specialists and technology consultancies that will help set up these walled gardens, which will result in this barrier of adoption becoming more favorable over time. Beyond that, we see three key limits to adoption: 1) Regulation; 2) Reliability; and 3) The digital domain.

Regulation: Heavily regulated sectors will take longer to adopt AI technologies. This includes healthcare and transportation / autonomous vehicles, although we thought adoption for the latter would have been faster than it is today.

Reliability: A concern that relates to ensuring models only produce truthful, accurate outputs when we’ve seen the ability to produce reliable, coherent data is often limited. This may be acceptable in the creative field but can be quite disastrous for applications that require accurate historical information.

The digital domain: AI adoption is moving much faster in knowledge related areas and slower where there is a large physical component. Understanding how much of the work is truly digital rather than connecting physical and manual processes is important to thinking about adoption.
These themes underpin that AI is not about substituting humans with technology, but rather enhancing human capabilities with technology to focus on the more value-additive portions of tasks.

Investment opportunities and forward outlook

We see Generative AI as here to stay and driving demand for investment opportunity. As a result, we are tracking high-frequency data on the adoption of ChatGPT across the world, particularly in the U.S. Currently, there are approximately 30 million weekly users and more importantly, 1 million weekly users² who are willing to pay for ChatGPT - signifying a huge consumer demand and corresponding investment demand. In terms of sectors where we see significant possibility for investment opportunity from providers, we bucket them into the following categories:

Hardware companies / infrastructure providers: There is a large market of data centers which need to be completely refactored to support the computational needs of Generative AI, bringing opportunities to last the next decade and beyond. The demand for hardware, data, and server farm locations is greatly outstripping the supply, with startups competing for GPUs. This dynamic has been reflected in markets year-to-date with Nvidia being the star stock that has tripled in the past few months, but also other chip companies and other cloud providers following suit. We see incumbents taking share here in a market which is relatively monopolistic.

Foundational model builders: There are few scale players at the moment, but more open-source solutions are being offered through smaller players with networks who specialize in specific type of tasks. We believe there is a market for both types of solutions, though competition will exist amongst the two. This is the market where we see the most uncertainty in terms of picking winners, the overarching question being how players remain proprietary in their data.

Application developers: There are many players here across start-ups and incumbents, and our view is some will manage to monetize on applications developed using Generative AI very quickly and there will be the largest number of investment opportunities driven from this space.

Adopters of AI into business models to enhance productivity: The companies and sectors that are best placed are those that have a higher cost of labor relative to revenues, and a high amount or high fraction of tasks that can be automated. We see sectors such as software development, advertising, media, and video games benefitting from this dynamic, though generative AI is a double-edged sword as while it can enhance productivity, it can also lower the barrier of entry. Using it correctly will be key to monetizing the tools to enhance earnings.
In terms of security selection, we look at four main characteristics when developing views on names within the Generative AI space: 1) The need for proprietary data; 2) A sticky product mix embedded into company systems to prevent entry from new players into the market; 3) A strong data infrastructure which is easy to ingest; and 4) The ability to go to market faster as the development cycle across Generative AI becomes shorter. Following these key themes can help investors understand some of the mispriced opportunities currently out there instead of names that they pick up through the news or headlines.