Update on AI Valuation Logic and Risk Assessment Model

For decades, the semiconductor industry has been considered a typical cyclical industry: highly volatile demand and capital expenditures, inventory cycles driving prices, and corporate valuations mostly following a contrarian logic of 'low P/E ratio at peaks, high P/E ratio at troughs.' However, with the rapid penetration of artificial intelligence (AI), the semiconductor industry is undergoing structural changes, and investment methods that rely solely on the traditional cyclical framework are gradually becoming ineffective. AI is not only altering the demand curve but also reshaping investors' fundamental understanding of risk, growth, and valuation.

Firstly, what AI brings is 'structural growth' in demand rather than just short-term expansion. Unlike past consumer electronics or PC markets, AI computing demand is highly sustainable and driven by cloud services, enterprise data centers, and edge computing. This means that semiconductor demand is no longer entirely dependent on end-product sales but is closely linked to computing power needs. For investors, this requires rethinking revenue sustainability—past predictions were based on shipment volumes to forecast business cycles, but now attention must be paid to computing power demand, model scale, and the speed of AI application implementation.

Secondly, valuation logic is shifting from 'cyclical discount' to 'platform premium.' In the AI era, some semiconductor companies are no longer just a link in the supply chain but have become core nodes of the entire ecosystem. For example, companies with advanced process capabilities, AI acceleration chip design capabilities, or key equipment technologies exhibit higher competitive moats and network effects. Such companies can not only enjoy longer-term growth momentum but also have greater ability to dominate industry pricing power. Therefore, investors need to adopt valuation methods similar to those used for technology platform companies, such as using long-term discounted cash flow (DCF) combined with market share and technological barriers for assessment, rather than relying solely on P/E ratios or price-to-book ratios.

However, the optimistic expectations driven by AI also amplify risks, especially in terms of capital allocation and supply-demand mismatches. The semiconductor industry remains inherently capital-intensive, and when companies significantly expand capacity based on AI demand, if future demand growth falls short of expectations, it could lead to a repeat of past overcapacity issues. Therefore, risk assessment models must incorporate sensitivity analyses for 'demand uncertainty' and 'investment payback periods.' Specifically, investors should build multi-scenario models, such as high growth, baseline, and low growth scenarios, and evaluate changes in return on invested capital (ROIC) under different scenarios, rather than relying solely on a single forecast.

Additionally, supply chain centralization is also an unavoidable source of risk. The reliance of AI chips on advanced processes and high-end packaging has concentrated the industrial chain among a few key companies and regions. While this concentration improves efficiency and accelerates technological breakthroughs, it also amplifies risks related to geopolitics and supply disruptions. When valuing companies, investors should factor in 'supply chain resilience' into their discount rate adjustments, such as increasing risk premiums for companies with highly centralized risks or giving valuation bonuses to companies with diversified supply capabilities.

From the perspective of the investment framework, semiconductor investments in the AI era can be divided into three levels. The first level consists of 'infrastructure companies,' including advanced process and equipment suppliers, which benefit from the long-term capital expenditure trend in AI and exhibit higher stability. The second level comprises 'core computing power companies,' such as AI chip design firms, which have the highest growth potential but also experience significant valuation volatility. The third level includes 'application-oriented companies,' which are driven by AI but remain partially influenced by end-user demand. Different levels require corresponding valuation methods and risk weightings, rather than a one-size-fits-all approach.

(Chip and Computing Power Series, Part 53)

$Alphabet-C (GOOG.US)$$Tesla (TSLA.US)$$NVIDIA (NVDA.US)$$Amazon (AMZN.US)$$BABA-W (09988.HK)$

322K Views