AI Stocks: From Matrix Multiplication to Market Valuation

This blog discusses the AI-related stock market, which has delivered remarkable performance over the past several years and whose long-term outlook still appears favorable. Recent volatility in AI-related stocks is driven by extremely high market expectations colliding with uncertainty over the profitability of AI, while high valuations, interest-rate sensitivity, geopolitical risks, and export controls further amplify short-term price swings. The purpose of this blog is to better understand the outlook of the AI-related stock market. 

Before beginning, I would like to clarify that I am not a financial expert and have only been actively trading stocks for about one year. However, I do have expertise in mathematics, AI, engineering, medical imaging, and related fields. This background, combined with feedback from many friends and the assistance of ChatGPT, has helped me organize and articulate my thoughts more clearly.

The AI market is actually a vast industrial ecosystem composed of multiple layers, including infrastructure, platforms, and applications. Over the last several years, the market has operated under a very simple assumption: more AI means more computation. Because modern AI fundamentally relies on repeated large-scale matrix operations, such as  $Y = XW$,  investors naturally concentrated enormous amounts of capital into companies supplying the computational backbone of AI.  In large language models based on the Transformer architecture, these matrix multiplications are performed repeatedly at an enormous scale during both training and inference. If \(X \in \mathbb{R}^{B \times d}\) represents a batch of token embeddings and \(W \in \mathbb{R}^{d \times m}\) denotes a trainable weight matrix, then the output feature representation is computed through matrix multiplication over billions or even trillions of parameters. In practice, modern Transformer models require not only a single matrix multiplication but a sequence of extremely large tensor operations associated with self-attention, feed-forward layers, and backpropagation updates. For example, the self-attention mechanism itself involves repeated computations of the form $Q = XW_Q,~ K = XW_K, ~ V = XW_V,$ followed by $\mathrm{Attention}(Q,K,V)=\mathrm{softmax}(\frac{QK^T}{\sqrt{d_k}})V.$. Since these operations must be executed across massive datasets and very deep neural architectures, the total computational complexity becomes enormous. During training, the optimization process repeatedly updates parameters through stochastic gradient-based iterations $\theta_{t+1}=\theta_t-\eta \nabla_{\theta}L(\theta_t),$ where \(L(\theta)\) denotes the loss function and \(\eta\) represents the learning rate. Consequently, modern AI training requires extraordinary computational throughput, memory bandwidth, inter-chip communication, electrical power, and cooling infrastructure. For a more detailed mathematical discussion, see my previous blog article on large language models.

This mathematical and computational structure explains why investors concentrated heavily not only on companies supplying GPUs, high-bandwidth memory, advanced semiconductor packaging, networking infrastructure, and hyperscale datacenter systems, but also on firms supporting the rapidly expanding electrical and industrial infrastructure required for large-scale AI computation. Companies such as NVIDIA, AMD, Broadcom, Micron Technology, TSMC, SK hynix, Samsung Electronics, and Samsung SDS became major beneficiaries because they provide critical computational, semiconductor, and enterprise infrastructure underlying modern AI systems. At the same time, the enormous electrical power demands of AI datacenters increasingly benefited energy- and power-related industries, including power equipment manufacturers, transmission infrastructure providers, electrical grid operators, and power plant companies. Firms involved in transformers, substations, transmission lines, industrial automation, smart-grid systems, and power management technologies also attracted growing investor attention because large-scale AI computation ultimately depends not only on semiconductors and algorithms, but also on stable, scalable, and energy-efficient electrical infrastructure. In this sense, the AI investment boom is evolving beyond a purely semiconductor-driven story and increasingly becoming an energy and industrial infrastructure story as well. 

As infrastructure development becomes more visible, investors are increasingly asking who will ultimately capture the durable economic value generated by AI. In other words, the AI-related stock market appears to be evolving through multiple stages toward a business model centered on productivity improvement and long-term monetization. The first stage was the infrastructure expansion phase, during which investors aggressively bought companies associated with GPUs, semiconductors, datacenters, and related infrastructure. However, once expectations become excessively elevated, even relatively modest concerns can trigger significant market corrections. 

The second stage focuses less on building AI itself and more on monetizing AI. In the coming years, the market may gradually transition from an “infrastructure enthusiasm” phase into a “monetization realism” phase. Investors are increasingly distinguishing between companies that merely develop AI models and those capable of generating stable and recurring revenue through AI-driven services, enterprise integration, and operational efficiency.

As a result, investors are increasingly asking who owns enterprise distribution channels, who controls customer relationships, which firms can successfully integrate AI into real operational workflows, and which companies can generate durable subscription-based or platform-based revenue streams. These questions naturally favor large platform and cloud companies such as Microsoft, Alphabet, Amazon Web Services, Meta, and Oracle in the United States. In Korea, companies such as Samsung SDS, NAVER, Kakao, LG CNS, and KT Corporation are also attracting increasing attention because of their ability to combine AI with enterprise systems, cloud infrastructure, manufacturing, logistics, and digital services. At the same time, important niche markets are emerging in areas such as AI cybersecurity, medical AI, industrial AI, robotics, semiconductor design automation, and enterprise workflow optimization. Consequently, the long-term winners may not necessarily be limited to a small number of frontier AI model developers.

However, market conditions remain highly sensitive to macroeconomic and geopolitical factors. As seen recently, when rising Treasury yields coincide with geopolitical uncertainty, energy concerns, and valuation pressures, investors tend to reduce exposure first in sectors that are highly cyclical, economically sensitive, and volatile. Consequently, even companies strongly connected to AI can experience substantial short-term corrections despite favorable long-term structural trends.

The third stage is already beginning to emerge, as markets increasingly focus less on AI technology itself and more on whether AI can generate measurable improvements in economic productivity. Investors are paying growing attention to whether AI can reduce labor costs, improve industrial efficiency, automate complex workflows, and enhance decision-making processes across real-world industries. As this transition progresses, entirely different sectors may become major beneficiaries, including robotics, industrial automation, medical AI, logistics optimization, smart manufacturing, autonomous systems, and enterprise operations. In the United States, companies such as Tesla, Palantir Technologies, Intuitive Surgical, and major industrial automation firms are increasingly positioned to benefit from AI-driven productivity gains. In Korea, firms such as Hyundai Motor Company, Hyundai Robotics, Samsung Electronics, Samsung SDS, LS ELECTRIC, and smart-factory and industrial automation companies may also become important long-term beneficiaries as AI increasingly transforms manufacturing, energy management, logistics, and enterprise operations.

Perhaps by around 2035, the ultimate winners may not necessarily be the companies possessing the most advanced chatbot. Instead, the firms most likely to benefit in the long term may be those that can most efficiently transform massive computational capability into sustainable economic productivity, industrial efficiency, and stable operating profits.