Our perspective on the AI industry. TEN and Enerzai
Last week, I was invited by my former employer, SoftBank Ventures, to attend a Fireside Chat event with Sam Altman and Greg Brockman of OpenAI. There were many interesting conversations, and many people in the industry have already shared good insights through blogs and social media.
It is still too early to say exactly how generative AI will change the world in the future, but there is little disagreement about the fact that the shape and structure of industries and societies will change dramatically. AI technologies and business models will also continue to change, and many things will become clearer over time, but here are the broad directions that we are assuming about AI from an investment perspective currently.
General purpose platforms based on large language models (LLM): This sector is likely to be reorganized in the form of an oligopoly, like the past cloud market. Two giants (Microsoft, Google) and one startup (OpenAI) have already started the first round, but Facebook, which has accumulated massive and unique personal data and has invested heavily in the AI/ML field, is not likely to withdraw from the competition. Microsoft is already one step ahead by providing the ChatGPT API through Azure Cloud, but Google who has launched Bard and others are also likely to continue to compete. As always, there is no guarantee that the first one to strike will be the final winner. However, unfortunately, it seems that there will be limited opportunity for small startups in the field of LLM platform.
Services using LLM platforms (based on GPT): In the past, when there have been dramatic changes in the way humans and machines interact with each other, the related industries have also undergone drastic and disruptive changes. When people used personal computers, software such as Windows and Office became massive players. When people accessed the network through the internet, the web service industry such as e-commerce businesses became household names. When people began to access the network independently of location through mobile devices, the mobile platform industry, such as social media, became a global phenomenon. LLM-based AI platforms are perceived as a cultural touchstone in the way humans and machines interact with each other because they transfer many of the active, intelligent, and generative processing that the human brain's frontal lobe is responsible for to digital intelligence. Therefore, it is reasonable to assume that companies that find a successful formula in various industries during this transformation are likely to receive similar rewards as Amazon did in the internet age, and as Facebook, Instagram, and Netflix did in the mobile age. Startups will have many opportunities, but as always, it will take some time to understand the needs of customers, resolve friction with existing industries, and find concrete solutions for growth. On the other hand, as general purpose platforms have significantly lowered the barriers to entry for AI technology and services to a level that even general developers can implement, as the difficulty of implementing AI technology decreases, numerous companies will enter the field, and it will become increasingly difficult to become the winner of a specific service. This is because the total amount of time that humans can access AI and other services throughout the day is also finite.
AI platforms and services using different technologies than LLM or requiring strict privacy: Convolutional neural network-based medical AI services, healthcare and medical AI fields with sensitive issues of personal information protection, robot fields using reinforcement learning, edge computing and smart factories through time series data analysis, law and engineering fields that require professional data, etc. In technical fields where LLMs are not seen as having a superior element, there is still a need for infrastructure, services and platforms. As a result, even in these fields, there are many startups that are developing and specializing in AI services optimized for specific industries. These startups have the potential to develop independently of LLM platforms or in a complementary way with them. There are already many startups in this field, and they will continue to appear in the future.
Infrastructure, technology, and services that support the efficient and easy development of AI applications: There are many infrastructure, technology, and services that can emerge to address the bottlenecks in the value chain of various AI fields and to support the smooth development and operation of AI applications. For example,
Nvidia almost monopolizes GPUs, which are essential resources for AI. As a result, new semiconductor companies in the NPU field, which are optimized for specific AI services or functions, or with other brand-new approaches such as integration of GPU and memory can take some of the market.
MLOps, which efficiently support the functions of various areas such as allocation of AI resources, design, learning, deployment, inference, and monitoring of AI services, will grow by providing third-party functions or complementary functions to general-purpose platforms.
In some inference services of mobile, edge computing, and IoT devices, companies that provide ultra-lightweight AI models and technologies will appear, because there are limits to inference through high-level computation.
Our portfolio includes companies that are developing and specializing in AI services optimized for specific industries, as in #3 category above. For example, Medipixel has developed an AI engine that is applied to cardiovascular areas such as angiography and coronary intervention, and has received FDA medical device approval. Gentle Energy uses AI to analyze data from sensors on various factory equipment to maximize factory operation and energy usage efficiency.
On top of that, we recently made investments in two startups in the area of #4 as well.
Ten: An MLOps platform that optimizes the utilization of resources including GPUs based on Kubernetes, and provides functions such as allocation, learning, deployment, inference, and monitoring of these resources.
Enerzai: Provides edge-based AI model technology through data pre-processing, AI model and hardware optimization suitable for edge computing environments such as cars, drones, smart factories, and mobile devices.
The interesting part is that the two companies are targeting opposite markets. Ten focuses on efficiently processing large-scale and complex upscale-based AI services, while Enerzai focuses on optimizing to be able to use relatively high-level AI functions even on lightweight mobile devices.
Recently, the value of companies in the generative AI field has increased significantly, with massive investment flowing to mostly those in the United States. It is up to each investor and founder to decide which AI company to invest in and at what valuation. However, we would like to remind ourselves that history has proven that in the wake of big trends and the success cases of the Amazons, Googles, and Facebooks, there are countless startups that formed throughout the trend that disappear. As investors we objectively know that past success does not guaranteed future results, and thus we continue to pose the question to ourselves: Will we be able have the foresight to choose and invest in successful companies that will succeed in the face of daunting odds? I don’t know but history repeats itself.
