New technologies have the potential to bring about revolutionary changes, such as the emergence of the internet and smartphones, which created a group of companies with market capitalizations exceeding billions or even trillions of dollars. In recent months, the most talked-about topic has undoubtedly been “Generative Artificial Intelligence” (GAI), including OpenAI’s ChatGPT, Google’s Bard, and Baidu’s Ernie, among others. Readers who have tried GAI can attest to its capabilities, which have surpassed expectations, and its impact on various industries is becoming apparent.
In early May this year, the stock price of the online learning platform Chegg, targeted at university students, plummeted almost overnight. The management attributed this decline to the significant increase in students’ interest in using ChatGPT since March, raising concerns about its potential impact on the company’s future growth prospects.
The GAI industry is still in its early stages, making it challenging for investors to accurately grasp investment opportunities. However, two crucial questions come to mind: where does the competitive threshold lie in the GAI field, and which companies currently stand to benefit from GAI?
The GAI industry can be divided into three main parts: the middle, upstream, and downstream. The middle segment consists of GAI neural network model owners, who are responsible for building and training models, such as the three GAI companies mentioned above. The upstream segment consists of infrastructure providers, including hardware graphics processing unit (GPU) suppliers and cloud platform companies. Cloud platform companies purchase GPUs and use them to create systems for neural network model owners. Lastly, the downstream segment includes companies that cater to end customers. They obtain data from model owners and use it to create software and apps that serve users, such as GitHub CoPilot, which assists programmers in writing code.
While many people believe that intellectual property rights may be the most critical competitive threshold, there are currently numerous open-source “large language models” available for use. Additionally, many AI experts request permission from employers to publish research reports in professional journals, fostering a strong atmosphere of mutual learning. In fact, even the core structure of ChatGPT is based on Google’s Transformer model, first introduced in 2017. The “T” in ChatGPT stands for Transformer.
Industry insiders point out that the most crucial competitive thresholds are data quality and computing power, both of which essentially come down to financial resources. Although there is an abundance of free data on the internet, it requires significant investment to clean and prepare the data for use. If GAI is trained using biased data, it may generate a GAI imbued with discrimination and hatred.
As for computing thresholds, Google has stated that training Bard using only one GPU would take approximately 355 years. The price of a project-grade GPU is around 10,000 USD, which implies that the hardware investment alone would be astronomical if one wants to shorten the training time to a reasonable level. Industry insiders estimate that training a large-scale language model can cost up to hundreds of millions of USD. Moreover, training cannot be accomplished in a single step; it requires constant experimentation and failure. How many companies can afford such financial resources to successfully train GAI?
Not only are hardware investments and training costs expensive, but the operational costs of running GAI are also much higher compared to running search engines. Investment banks have calculated that the cost of each GAI response is ten times higher than that of a regular search. GAI model owners are still in the exploratory stage of how to create profitable business models using GAI. In the long run, if companies holding GAI models can eventually transform GAI into a platform similar to Microsoft’s Windows or Apple’s iOS, allowing different software to run on it, generating economies of scale and locking in customers, it would create a difficult-to-exit business model and greatly expand development opportunities.
As for client-side software, since the industry is still young, it seems that no business models with competitive advantages have emerged yet. Apps that use AI to help students with tutoring or assist programmers in coding, for example, currently lack the ability to retain customers in the long term.
In summary, the GAI industry is still a money-burning industry, and large companies have a competitive advantage due to their financial resources. In the short term, upstream companies seem to have more investment opportunities: a significant portion of the expenses for training GAI will be spent on purchasing services from cloud companies, and since GAI requires GPUs for operation, cloud service companies and GPU manufacturers will benefit from this.
Disclaimer
This document is based on management forecasts and reflects prevailing conditions and our views as of this date, all of which are accordingly subject to change. In preparing this document, we have relied upon and assumed without independent verification, the accuracy and completeness of all information available from public sources. All opinions or estimates contained in this document are entirely Zeal Asset Management Limited’s judgment as of the date of this document and are subject to change without notice.
Investments involve risks. Past performance is not indicative of future performance. You may lose part or all of your investment. You should not make an investment decision solely based on this information. Each Fund may have different underlying investments and be exposed to a number of different risk, prior to investing, please read the offering documents of the respective funds for details, including risk factors. If you have any queries, please contact your financial advisor and seek professional advice. This material is issued by Zeal Asset Management Limited and has not been reviewed by the Securities and Futures Commission in Hong Kong.
There can be no assurance that any estimates of future performance of any industry, security or security class discussed in this presentation can be achieved. The portfolio may or may not have current investments in the industry, security or security class discussed. Any reference or inference to a specific industry or company listed herein does not constitute a recommendation to buy, sell, or hold securities of such industry or company. Please be advised that any estimates of future performance of any industry, security or security class discussed are subject to change at any time and are current as of the date of this presentation only. Targets are objectives only and should not be construed as providing any assurance or guarantee as to the results that may be realized in the future from investments in any industry, asset or asset class described herein.
In respect of any discrepancy between the English and Chinese version, the English version shall prevail.