Understand the next wave of tech capital expenditure

The investment boom in artificial intelligence appears to focus on large language models, chip companies, and cloud platforms. However, from a long-term valuation perspective, AI infrastructure is not a single track but a multi-layered capital structure composed of data, computing power, energy, cloud, models, software, and end-user applications. Investors who bet on only one aspect may be vulnerable to valuation fluctuations, technological substitution, and commercialization timing; conversely, understanding AI infrastructure from an overall portfolio perspective gets closer to the essence of this industrial transformation.

The first layer of AI capital is data. The core asset of past internet platforms was user traffic, whereas in the AI era, the key assets are high-quality, continuously updatable, and contextually deep data. Public data can train general-purpose models, but what truly determines the AI value of enterprises is often proprietary data in vertical fields such as healthcare, finance, industry, law, and retail. This type of data may not appear directly on financial statements, yet it could become a source of future bargaining power, product differentiation, and customer stickiness. Therefore, long-term valuation should not only consider current revenue but also whether a company possesses reusable, structurable, and intelligent service-convertible data assets.

The second layer is computing power. Chips, servers, network equipment, data centers, and power supply form the most visible hard infrastructure of AI. This layer is the most prone to peaks in capital expenditures and also most likely to trigger market concerns about bubbles. The reason is that investments in computing power often precede revenue realization, with depreciation, energy costs, and supply chain cycles amplifying profit volatility. However, computing power is not just a cost—it’s also a form of capacity. As demand for model training, inference, and enterprise AI deployment continues to rise, companies capable of providing low-cost, high-efficiency, and scalable computing power operations may enjoy long-term excess returns. Investors need to discern which companies are merely selling short-term popular equipment and which ones possess architecture, software ecosystems, customer lock-in, and energy configuration advantages.

The third layer consists of models and platforms. Foundational model companies bear high R&D and training costs, but their business models are still evolving. On one hand, improvements in model capabilities may lead to economies of scale; on the other hand, open-source models, model distillation, and enterprise-built solutions may also lower the prices of certain model services. This means that the valuation of the model layer cannot rely solely on technical leadership but must also assess its ability to build distribution networks, developer ecosystems, enterprise workflow integration, and stable revenue models. Simply being the 'best model' does not necessarily equate to 'highest commercial returns'; companies that turn models into platforms and platforms into ecosystems are more likely to achieve long-term valuation premiums.

The fourth layer is applications. AI will ultimately unlock economic value by improving efficiency, reducing costs, creating new products, and changing user behavior. Enterprise software, customer service, automated programming, content generation, drug discovery, financial risk control, and industrial design will be key areas where AI infrastructure monetizes. However, competition at the application layer is the fiercest due to lower entry barriers and higher risks of product homogenization. What deserves long-term attention is not just one-off functional innovations but whether a solution can embed itself into customers’ core processes, capture scenario-specific data, raise switching costs, and establish recurring subscription or outcome-sharing business models.

From a portfolio perspective, AI infrastructure can be divided into three types of assets: 'picks and shovels,' 'platforms and channels,' and 'applications and cash flow.' Chips, cloud computing, data centers, and electricity belong to the foundational infrastructure layer, benefiting from demand expansion but highly dependent on capital expenditure cycles for valuation. Cloud platforms, model platforms, and development tools form the intermediate layer, where the focus lies on ecosystem strength and pricing power. Vertical applications and enterprise services are closer to the cash flow level, with success determined by customer retention and actual return on investment. A more robust allocation should avoid concentrating solely on the most popular single leader in the market; instead, risk should be diversified across different layers.

(Chip and Computing Power Series, Part 56)

$NVIDIA (NVDA.US)$ $Taiwan Semiconductor (TSM.US)$ $Advanced Micro Devices (AMD.US)$ $Intel (INTC.US)$ $Alphabet-C (GOOG.US)$

313K Views