Artificial intelligence (AI) is rapidly evolving into a fundamental technology for all industries. As with the evolution of mobile phones a decade ago, the continual development and deployment of AI presents a number of opportunities for long-term investors. According to the IDC, global annual revenues for cognitive and artificial intelligence systems are estimated to rise from $8 billion in 2016 to $46 billion in 2020, a compounded annual growth rate of 55.1%.
The industries investing the most in AI related services include banking, retail, healthcare and manufacturing, which generate more than 50% of AI industry revenue. But, this figure ignores the potential economic value available to other enterprises that embed AI technology.
Longer-term, AI will have a profound impact on all industries, introducing efficiencies and innovative catalysts. In the shorter term however, the spread of AI is ramping up the demand for hardware, semiconductors, memory, big data, and software.
The development of AI means that machines are becoming smarter - able to understand the physical world with more accuracy and depth - enabling machines to process and act on information automatically. This, in turn, should drive more demand for hardware and sensor technology. Humans obtain 83% of their information from the sense of sight.
The information embedded within visual data far outweighs other senses, which means that cameras should be the most important source of input data for AI Key beneficiaries of this trend include Sony (6758), which is the indisputable leader in image sensors with 48%, and growing, market share by revenue.
Artificial intelligence is a seismic opportunity for the semiconductor industry. At the NVIDIA investor day in 2017, the company's management estimated that the total market value for semiconductor accelerators used for deep-learning training and inference applications could reach $26 billion by 2020.
Considering the processing capability of graphics processor units, we believe that this will remain the dominant accelerator for training workloads. Once a neural network is trained for a specific type of task, it will need to be deployed across varying technology platforms, particularly at the network-edge, smartphones, cars and other connected devices.
We believe that field programmable gate arrays and application specific integrated circuits will gain from this increase in AI application, benefitting leaders within this space such as Microsoft (MSFT), Intel and Alphabet (GOOGL), as well as semiconductor foundry companies such as Taiwan Semiconductor Manufacturing Company (2330).
According to researchers at DeepMind, "the ability to learn tasks in succession without forgetting is a core component of biological and artificial intelligence". The human brain can transfer old skills to new problems in order to reach a desired outcome. This requires a competent thought process and most importantly, memory.
Until recently, this had been a shortcoming of machines. During training, artificial neural networks would learn to master a single task and then call it quits. Now, artificial neural networks are being combined with a random-access memory in order to achieve multi-step reasoning.
To think and learn faster, systems need to be able to access larger amounts of data, at a much faster rate. Memory technology needs to achieve a balance of high performance, power efficiency and reasonable unit cost. High Bandwidth Memory using dynamic random-access memory is a great option - Samsung Electronics (005930) currently dominates both these markets.
The Internet and Big Data
Over the last decade, global technology conglomerates like Alphabet have developed the largest pools of data on the planet. As data and computer efficiency are the two primary ingredients to further neural network-backed technology advancement, we believe that the largest internet companies have a major advantage.
This combination of technological superiority and unrivalled monetary commitment to investment and research will help the large digital platforms sustain their leadership in AI The data sets required to train these models are growing exponentially, as are the number of nodes within these networks and their technological advantages are enabling these firms to conduct huge statistical data exercises at costs that appear increasingly inexpensive.
We believe Alphabet to be the clearest example of an AI driven business model today, and already leads the industry in terms of research, investment and technological development. The company is already leveraging its scale - seven platforms each with more than one billion users, expertise in AI and the technology it acquired from DeepMind to enhance the performance of its core products and services such as Search, YouTube and Android as well as supporting the development of new opportunities including autonomous driving and biotechnology.
The software building blocks for AI are largely established and we are now starting to see AI functionality being monetized across the industry. AI represents a large up-sell opportunity for software vendors - the first way these companies can monetize AI is by bolting an add-on AI service or product to an existing core product. We have already seen a few examples of this from companies such as Salesforce (CRM) and Coupa.
Software vendors can also provide the building blocks to construct a machine learning model, such as a platform or 'toolkit'. For example, a customer may be looking to embed some AI functionality into their payroll software, but may not have the time or money to do so internally. An AI platform solves this by providing all of the 'tools' necessary to build a machine learning model and then deploy it. Microsoft Azure is one of the major vendors offering this.
We believe that Microsoft is well-positioned to become one of the biggest beneficiaries of AI within the global software industry. The company's cloud business, Azure, is the world's largest cloud computing environment, with double the capacity of its closest rival Amazon (AMZN) Web Services.