The First Principles of the AI Age

In an age of explosive growth in artificial intelligence, obtaining information has become easier than ever before. Yet this very convenience brings with it a crisis of cognitive shallowness. Faced with an ocean of data and instant knowledge, how an investor stays clear-headed amid all this noise has become a required lesson. True information literacy means discarding fast-food, fragmentary information, re-embracing first principles, and correctly harnessing powerful AI tools, turning them into instruments for raising one’s financial intelligence.

Computing Power as the First Principle of Wealth

Under the tide of the AI revolution, the traditional mindset of “linear asset growth” has completely failed. By returning to the essence of the data through first principles, an investor discovers a startling “acceleration phenomenon.”

Take the NASDAQ-100 as an example. Its historical annualized return shows a clear exponential steepening:

• Trailing 20 years: an annualized return of about 14.7%
• Trailing 10 years: climbing to about 19%
• Trailing 5 years: surging to 20.4%

This ever-rising slope far exceeds the roughly 9.5% long-term annualized baseline of the S&P 500. (The figures above are drawn from specific backtest windows for the NASDAQ-100 and will differ depending on the index version, whether dividends are included, and the data date; the point is not a guarantee that future returns will keep accelerating, but that platform-type enterprises have in recent years displayed higher capital efficiency and a steeper profit slope.) This data reveals the core logic of the AI age: tech giants, through algorithms and automation, are replacing traditional labor at an accelerating pace. This change in slope means that if you hold none of the “ownership” of these enterprises, your income and wealth may be squeezed ever harder by AI platforms and capital owners. In the AI age, computing power, data, and cloud infrastructure become the new factors of production; holding QQQ or 00662 does not mean directly owning a GPU, but indirectly holding the group of enterprises with the greatest capacity, worldwide, to convert AI into revenue, cash flow, and monopoly advantage.

▲ Figure 23-1 The steepening of the NASDAQ-100's annualized return, where the closer the observation window sits to the present, the higher the annualized return

Faced with lofty AI valuations, retail investors most often ask: “Isn’t this just a rerun of the 2000 dot-com bubble?” Teacher James shatters this fear with the most straightforward micro-level financial data:

The price of AI computing power is rising. It may have gone from one dollar an hour to five or six dollars, and some short-term contracts have already risen to 14 dollars. This rise in computing-power contracts benefits the profits of these infrastructure companies, these AI-infrastructure companies, or their growth rate will exceed expectations. In other words, AI investment has no bubble; its demand genuinely exists. (Video 00688)

The essence of the 2000 dot-com bubble was that “the companies themselves had no real profits, and pure story alone propped up the share price.” The AI wave of 2025 to 2026 is different: computing-power rental prices, GPU utilization, and cloud orders keep rising (the actual range of computing-power rents varies enormously by GPU model, rental term, and supplier), reflecting the rapidly swelling revenues of these infrastructure companies such as Nvidia, Microsoft, Google, and Amazon. This is not a pure valuation story, but one backed by the numbers on the financial statements.

When retail investors hear the argument that “the PE is too high, AI is a bubble,” rather than debate it, it is better to look directly at the underlying data: computing-power rents, GPU utilization, cloud RPO, and Capex. As long as computing-power prices, RPO, cloud revenue, and capital expenditure still corroborate one another, AI demand is not the kind of pure-story bubble seen in 2000. But this does not mean “there is absolutely no valuation overheating”: that the long term is a genuine revolution, and that certain companies in certain years at certain valuations are somewhat expensive, are two things that can both be true at once. What truly needs to be tracked continuously is the speed at which RPO converts into revenue, gross margin, depreciation, and free cash flow, rather than any single price.

Asset Owners vs. Computing-Power Consumers

The AI age divides humanity into two entirely new classes:

• Computing-power consumers (the labor class): they sell their time to earn a wage, yet in daily life they pay subscription fees to AI companies and have their attention precisely harvested by algorithms. Wage growth can never keep pace with the speed at which technology replaces labor.
• Asset owners (the capital class): they see through to the essence that “computing power is wealth,” and by holding tech indexes, they let the world’s smartest AI CEOs (such as Jensen Huang and Satya Nadella) work for them.

To sum it up in one sentence: we must transform from “tool-people who sell their time” into “resource owners who harness AI.”

When you own assets, the advance of AI is a “wealth pressure-booster”; when you own only labor, the advance of AI is an “unemployment alarm.” Which side you choose to stand on determines an individual’s fate in the AI age.

To fully grasp the cruelty of this paradigm shift, you must first see clearly its fundamental difference from every technological revolution of the past:

Past paradigm shifts eliminated the product, but people still existed. The paradigm shift of artificial intelligence will replace the whole of humanity, so when you go looking for work you really cannot find any. Only 0.001%, like C’s friend who invests, still have a way to live. Otherwise, even if you have a job, it will disappear later. If you do not invest, you will be poor. So in the future there will be only two kinds of people: one is the extremely rich, and one is the extremely poor. There is no middle. (Video 00695)

When digital cameras replaced film, Kodak went bankrupt but photographers still had work; when the CD replaced vinyl, the record-company model was renewed but musicians were still creating; when smartphones replaced feature phones, Nokia fell but the communications industry still needed a large labor force. All these past paradigm shifts eliminated only the “product,” and “people” could still find a place on the new track.

But the AI paradigm shift is different: AI directly replaces “human labor” itself. Once customer service, accounting, clerical work, design, programming, entry-level law, and entry-level medicine are all replaced by AI, these discarded human laborers have nowhere to go, because the new job opportunities that emerge will themselves also be taken over by AI. This is precisely the sharp warning Teacher James raises about the AI age: AI will compress a vast amount of standardizable, process-able knowledge labor, magnifying the gap between “asset owners” and “pure laborers.” Different occupations, industries, and skills are affected to different degrees, and not necessarily every human job will disappear; but the direction is clear: in the AI age, relying on wages alone grows more and more fragile, while holding quality productive assets (QQQ or 00662) grows more and more important. “Holding no productive assets at all” is equivalent to exposing yourself to the side that AI and capital keep squeezing.

Rejecting Junk Information and Verifying in Depth

Although AI can generate lengthy discourses in a matter of seconds, this has also caused the “financial parrot” to flourish. Most investors rely on short videos or AI-generated summaries, yet have lost the ability to verify independently.

Reading books is actually the most important thing. It is the most basic and also the cheapest investment. (Video 00415)

The true first principle is “returning to the raw data.” When carrying out asset allocation, one should not blindly trust the “safe and steady” advice that AI hands out, but should personally operate backtesting tools and precisely calibrate against the data. This active verification of the data is the “sovereignty over decisions” that AI cannot perform on your behalf.

Take “whether AI is a bubble,” the hottest debate of recent years, as an example. The traditional way to observe it is to read media headlines, listen to analysts’ opinions, or watch the ups and downs of the share price. But returning to the raw data, there is a financial-statement metric that leaks the truth earlier than the share price: RPO (Remaining Performance Obligation). RPO represents the order amount that enterprise customers have actually signed and committed to, but which has not yet been recognized on the books. It is the most direct evidence of whether AI demand is genuinely materializing.

Take Oracle as an example. Its FY2026 second-quarter report, released in December 2025, disclosed an RPO scale as high as US$523 billion, a year-on-year increase of 438%. This figure is important evidence that AI demand is genuinely materializing: if AI were merely an imagined story, enterprise customers could not possibly sign US$523 billion in real purchase commitments. But to be clear: RPO is a signed, not-yet-recognized performance obligation, not recognized revenue, and still less guaranteed profit; it remains subject to the performance period, gross margin, Capex, financing costs, and customer concentration, and the speed at which it converts into revenue must be tracked.

This is the concrete practice of “distilling information”: when the market is in an uproar over “whether AI is overheating” or “whether QQQ is overvalued,” rather than listen to 100 analysts’ opinions, it is better to open directly the financial statements of the top tech enterprises and look at the RPO, the Capex (capital expenditure), and the unit economics of the cloud business. This raw data sits closer to underlying reality than media headlines do, but it still needs to be judged together with the accounting basis, contract terms, customer concentration, and cash-flow conversion capacity, rather than treated as a crystal ball that never errs. For holders of the CLEC system, this kind of underlying data provides far more “conviction to hold” than any short-term price swing: when you see US$523 billion of orders in hand, the market’s short-term panic and turbulence become far easier to treat as noise that can be ignored.

Harnessing AI to Distill Value

AI should not be used to replace thinking, but to “distill value.” The 00662 community uses NotebookLM to process millions of words of transcripts and lecture notes, its core aim being to build an “external brain.” This kind of tool is best suited to a “closed-source” database: feed it only CLEC lecture notes, transcripts, and official documents, so that the answers stay as close as possible to the designated sources, rather than letting AI answer wildly from its general internet memory.

An AI tool is like a mirror: the value of its output depends entirely on how much basic logic the questioner possesses. Learning to use high-level questioning and rigorous frameworks to direct AI in data retrieval, risk simulation, and tax computation is the modern capitalist form of “one person commanding a thousand armies.”

The true practice of first principles is to use multiple models for a merciless cross-examination. Take “insurance-policy optimization” as an example. Rather than listen only to the sales pitch of an insurance agent, it is better to feed the entire family’s policies simultaneously to the latest Gemini 3.0 Pro, Grok, and ChatGPT, let the models critique one another, and compile a list of each policy’s coverage gaps, its high-fee low-coverage items, and the questions you should press your insurance advisor on. But hold to one line: whether to surrender a policy, reduce it to a paid-up policy, or keep it must be judged according to family responsibilities, health status, debt, pre-existing medical history, coverage gaps, and policy terms. Even if the multiple models unanimously advise “cut some low-efficiency life insurance,” this can only serve as a starting point for examination, not a scientific conclusion to be executed as-is (mistakenly cutting needed coverage, especially for those with family responsibilities or whose health has deteriorated so that they can no longer be reinsured, has irreversible consequences). Only on the premise that coverage is assured should you then redirect the truly ineffective premiums into a market-cap-weighted broad-market index to enjoy compounding; this is the correct demonstration of using technology to take back the “stupidity tax.”

The most valuable ability in the AI age is not producing conclusions faster, but finding faster the “verifiable, falsifiable, and repeatable” scientific conclusion. In this race for computing power, only by holding to first principles and keeping core assets deeply linked to technological progress can you, in this age of exponential change, hold the line of wealth that spans generations.

But even if you can use AI and can run multi-model cross-verification, there remains one line that must not be crossed: “maintaining a final human review over AI”:

Let me tell you all: AI will talk utter nonsense with a straight face, so you absolutely must be careful about this. Right now, when AI has to be used for the real thing, for something truly professional and serious, it is not up to the job. So to put it plainly, if you treat it as something formal, something for dealing with the government, and for filing taxes, I am telling you, you absolutely must go back to the most original official government documents. That is a must. (Video 00536, 2025 interview)

A note on timeliness: the comments above were made in early 2025, when mainstream AI models commonly suffered from “hallucination” and had a higher error rate in professional fields such as tax law and regulation. Entering 2026, a new generation of models such as Gemini 3.0 Pro, ChatGPT GPT-5, and Grok has clearly improved at organizing data, comparing documents, generating question lists, and aiding comprehension, and hallucinations have also decreased from the early days; but hallucinations have not disappeared, and the higher-risk the field, such as tax, law, insurance, and cross-border assets, the less one can finalize a decision on AI output alone. The correct attitude is “verify carefully, cross-check, and ultimately defer to official documents and licensed professionals,” rather than reject everything outright or accept everything wholesale.

Tiered handling for real-world practice:

• Concept exploration, logic practice, producing first drafts: AI can already serve as the main tool
• Cross-model verification: use at least two or more models, such as Gemini plus ChatGPT plus Grok, to critique one another, which can greatly reduce the error of any single model
• The final-decision layer (tax filing, legal contracts, cross-border taxation): you must still consult the original official documents by hand (the IRS or national tax bureau website, Ministry of Finance regulations) or consult a licensed accountant or lawyer for final confirmation

Treating AI as “a powerful assistant plus a draft that must be given a final human review” is the most pragmatic collaborative model for the investor in the AI age: you enjoy the efficiency gains AI brings, while retaining human oversight over the final decision.

An Advanced Cyclical Observation of the Five Great Economic Cycles (Optional Reading)

For the advanced investor with a large asset base who wishes to study “cyclical observation” further, Teacher James once systematically laid out the observation indicators for the five great macroeconomic cycles:

The hard part lies in the Way, not in these consumption cycles, industry cycles, capital investment, credit cycles, real-estate cycles, and market sentiment. I could talk for three days and three nights, and when I finished you still would not understand. A cycle is never passively just a moving average; a cycle is not just watching moving averages. We also have to look at global finance. And the ones I just mentioned, the consumption cycle, the industry cycle, capital investment, the credit cycle, the real-estate cycle, and market sentiment, come first; technical analysis, the size of the head, comes last. You have to understand this completely, and you cannot understand only one; you have to understand all of it. (Video 00558)

The order and observation focus of the five great cycles:

• The consumption cycle: observe public consumer confidence, retail performance, and sales of autos and durable goods
• The industry (inventory) cycle: observe manufacturing PMI, corporate inventory levels, and the shipment-to-inventory ratio
• The capital-investment cycle: observe corporate Capex spending, the heat of the IPO market, and whether emerging-market stock markets are surging broadly past the United States (a lagging market suddenly taking the lead can serve as one signal of overheated risk appetite, but not a standalone judgment of a bubble’s final stage; it still needs to be paired with credit spreads, valuation, and leverage)
• The credit cycle: observe the yield curve, credit spreads, and corporate default rates (one company falling is equivalent to a cockroach effect)
• The real-estate cycle: observe the house-price-to-income ratio and central-bank rate movements (after credit ruptures, the central bank cuts rates to rescue the market, and the last wave of capital floods into real estate)

Mapping this colloquial set of cyclical concepts onto the classic cycles of macroeconomic academia, we can organize the following table. But first, a word of caution: the “approximate cycle” below is only a value from historical experience, not a timetable that reports on schedule, and the cycles further overlap and influence one another. Its sole use is to judge “whether at this moment I should bring leverage down a little and raise cash a little,” never to guess tops and bottoms or trade in and out.

To stress it once more: cycle lengths are only experiential values; they deviate and overlap, and globalization, technological change, and major geopolitical events can all throw them off. So this table cannot be used to “predict which year the crash will come”; it can only, when multiple cycles flash red at the same time, remind you to adjust your portfolio’s Beta and cash level toward the conservative end. This does not conflict with the CLEC mainline of “buy when you have money, never sell no matter what,” because what it adjusts is the strength of defense, not the timing of entry and exit.

Strict warning: observing the five great cycles is “only for” the “de-Beta defense” (raising the cash ratio to 40% to 50%) of those with a large asset base facing an extreme bubble; it must absolutely never be used for the “sell low, chase high” of swing trading. For 99% of investors, “never sell no matter what” remains the single principle with the highest win rate. Beginners, please ignore this framework entirely and simply keep executing the 433 allocation and annual rebalancing.