Way to data scientist

Date: 3th January 2026

Tags: #InvestmentThoughts #AIInsights #Semiconductors #MarketTrends

Introduction: A Counter-Intuitive Discovery

Recently, while conducting a historical review of market data, one specific finding left me staring at my screen for a long time: if you stretch the timeline to 50 years (1975-2025), the sector with the highest efficiency for wealth creation in the US market is Semiconductors.

For a long time, my cognition was dominated by the mantra “Software eats the world.” I tended to believe that Software (SaaS) was the perfect business model—zero marginal costs, high gross margins, and asset-light operations. In contrast, I viewed semiconductors as “hard grunt work”—asset-heavy and tormented by cyclical fluctuations.

But data doesn’t lie. This review forced me to rethink: With the AI wave sweeping in, is my bias against “Hard Tech” too deep?

Reflection 1: Moore’s Law is Not Just Technology, It’s an Economic Miracle

I began to realize that the semiconductor industry possesses a “physical superpower” that other industries lack: Deflationary Growth.

In the energy sector, making big money often relies on rising oil prices; in the pharmaceutical industry, we worry about patent cliffs. But in the chip sector, Moore’s Law guarantees that performance doubles every two years while prices continue to drop. I previously only saw the competition brought about by falling prices, ignoring the fact that this drastic deflation actually triggers infinite elasticity of demand.

From mainframes to PCs, from mobile phones to the current Internet of Everything, it is precisely because computing power is getting cheaper that it can penetrate every pore of our lives.

Especially in the AI era, I found the logic has changed. I used to think hardware was just a carrier for software; now I see that hardware constitutes the hard constraint of the physical world. Without Nvidia’s GPUs, even the best AI algorithms cannot run. This “monopoly at the physical layer” is much more solid than I imagined.

Reflection 2: Why Is the Market Becoming Increasingly Crowded?

Another phenomenon that made me think deeply is the extreme concentration of the S&P 500 index. The market cap of the top ten companies accounts for over 40%, a scene unseen in a century.

I used to worry this was a bubble or a market failure. But looking at it through the lens of AI characteristics, I’m starting to understand this might be an inevitability.

The barrier to entry in the AI era is simply too high. Training a GPT-5 level large model requires tens of thousands of H100 graphics cards, consumes the electricity of a city, and demands billions of dollars in cash. This is no longer an era where two geniuses can create miracles in a garage.

This leads to a brutal reality: Technology dividends are being “nationalized” into the hands of a few giants. Only the “Magnificent Seven” (like Microsoft, Google, Amazon) can afford to play this game.

As an individual investor, this trend makes me feel both helpless and excited. Helpless because of the potential loss of grassroots innovation vitality; excited because the investment logic has become incredibly clear—at this stage, following the giants seems to be the path of least resistance.

Reflection 3: “Shovels” and “Fuel” — My Double Bet on the Future

Based on these reflections, I have made a significant correction to my future focus. If the next decade is the explosion period of digital civilization, then I am bullish on the two pillars supporting this civilization:

1. The Evolved “Shovels”: Computing Infrastructure

Since AI is a gold rush, I will no longer gamble on which App will go viral (that’s a job for VCs). I am more concerned with who is selling the shovels. Whether it’s GPUs, High Bandwidth Memory (HBM), or advanced manufacturing processes, these are the infrastructures that no winner can bypass.

2. The Scarce “Fuel”: Energy

This is my biggest cognitive reversal recently. I used to think of utility stocks as boring “bond-proxy” assets. But now, I view them as Food for AI.

AI is essentially the conversion of energy into intelligence. Data centers are power-hungry beasts, and the construction speed of power grids and clean energy is far lagging behind the demand of AI. This supply-demand mismatch could lead to a value re-rating for companies that own stable power sources and grid assets.

Conclusion

I wrote this note not to predict stock prices, but to fix my own mental models.

Over the past fifty years, semiconductors have proven to be a sharp weapon against inflation and cycles. In the next decade opened up by AI, Computing Power and Energy may become the main melody.

In this era where the winner takes all, perhaps the most rational way to survive is to acknowledge the hard constraints of the physical world and stand on the shoulders of giants who control scarce resources (Compute and Energy).

(Note: This article is for personal reflection only and does not constitute any investment advice.)