Li Lu on AI: The Risk of Treating a Machine as a God

What struck me about Li Lu’s recent discussion at Columbia was the framework behind it. His individual predictions about AI, the United States, China, or markets mattered less than the way he brought those subjects together: technology changes the world, capital follows technology, and politics reshapes its boundaries. Yet long-term investment returns still depend on value, price, costs, competitive advantage, and how rationally people behave. AI may be the largest variable in this cycle without overturning the basic rules of investing. The more dramatic the technology and the more excited the market, the more those unglamorous rules deserve our attention.

Li Lu covered value investing, AI, China’s economy, US–China relations, the global political order, and compounding over a lifetime. Taken together, these subjects raised one central question for me:

When the world changes profoundly, what changes with it—and what holds?

That is where I would begin reading his remarks.

1. The danger of pricing AI as a god

Over the past few years, a familiar story about AI has grown beyond the idea of an increasingly capable tool. It now imagines something approaching a god of our own making.

The case for a tool is straightforward: AI can help people perform more and more tasks.

The more ambitious story says AI will eventually solve problems we cannot solve ourselves, transforming economic growth, productivity, wealth creation, and the structure of society.

The distance between those two claims is much greater than it first appears.

Li Lu repeatedly returned to that distinction. He acknowledged AI’s extraordinary capabilities and its potential to become one of history’s most important technologies. But he was wary of drawing a straight line from today’s systems to artificial general intelligence, then superintelligence, and finally a world in which intelligence faces almost no limits. Development, in his account, involves more than making models increasingly capable. Costs, energy, capital, human involvement, and social acceptance impose constraints of their own.

One implication seems particularly important to investors: society’s willingness to tolerate a technology’s risks may limit how far it can go.

What a system can do and what people will allow it to do are separate questions.

Consider autonomous driving. A system might be statistically safer than human drivers and still face public resistance when it makes a mistake. The stakes are higher still in medicine, finance, the courts, and the military. In such settings, technical feasibility is only the beginning. Who takes responsibility when something goes wrong? Will the public accept the risk? Will regulators permit it? Are people willing to surrender the final decision?

AI’s practical limits therefore extend beyond the capabilities of its models.

Engineering determines what a technology can do. Economics, institutions, and human attitudes help determine whether it will actually be used.

That is why I find “Will AI reach AGI?” an inadequate investment question.

I would rather ask:

Even if AI acquires a particular capability, who will pay for it? Who will bear the risk? How much existing spending can it replace? How much additional value can it create?

Those questions bring the discussion back from technological awe to economics.

None of this rules out productivity gains far beyond what we see today. If AI substantially lowers the marginal cost of knowledge work, it could become an exceptionally powerful general-purpose technology.

Investors still need to keep one distinction in view:

A technology can create enormous value for society without delivering enormous returns to everyone who invests in it.

Electricity, cars, and the internet transformed the world. Capital committed to those industries early on did not automatically earn excess returns commensurate with their social contribution.

The investment question, then, is: how much of AI’s future is already in the price?

When valuations assume almost unlimited potential, the risk is that even a remarkably successful AI industry may fall short of the success investors have already paid for.

2. Real progress can still produce a boom and bust

One of Li Lu’s more consequential observations was that AI’s development could involve several rounds of boom and bust, rather than a single uninterrupted expansion.

To me, the point is less a forecast of the next bubble than a broader economic lesson:

A technological revolution that continually consumes capital, energy, infrastructure, and organizational resources will have a capital cycle as well as a development curve.

AI looks like software, but much of the industry increasingly resembles a capital-intensive business.

Training models requires computing resources; running them requires more. Data centers need land, electricity, and networks. Chip production entails enormous capital expenditure. Infrastructure investment continues throughout the supply chain.

This is a long way from a pure software business with marginal costs approaching zero.

Once markets become convinced that a technology has a vast future, money pours in. Investment buys more computing capacity, better models, and wider adoption. Those advances appear to vindicate the original optimism, drawing in further capital.

That feedback loop cannot run indefinitely.

Eventually, capital must answer a plain question:

How much actual cash flow will these investments generate?

If cash generation fails to keep pace with investment, the cycle turns. Returns on capital weaken, valuations contract, projects fail, excess capacity emerges, and markets reprice the industry.

This helps explain why technological revolutions can pass through successive booms, busts, and renewed growth.

An AI bubble, if there is one, would not by itself mean the technology is a fraud.

Indeed, some of the most dangerous bubbles form around technologies with real transformative potential.

A worthless technology has difficulty sustaining widespread conviction. A technology that genuinely changes the world gives speculation something far more durable to feed on.

The internet was real. So was the dot-com bubble.

Railways were real. So were railway bubbles.

AI may follow a similar course.

In discussing the AI market, Li Lu returned to valuation and a useful long-term anchor: stock-market returns cannot indefinitely outstrip the earning power of businesses and the growth capacity of the economy.

That is an easily neglected part of value investing.

Valuation ultimately comes down to arithmetic, however impressive the technology.

An outstanding company in a remarkable industry still has to answer three questions: how much future cash can it generate, how much will reach shareholders, and how much must an investor pay today?

Technology can increase the cash those businesses produce.

The price you pay determines how much of that success becomes your return.

I would therefore spend less time debating whether AI deserves the word “bubble” and more time asking:

Have I already paid for ten or twenty years of exceptional success?

If so, AI could change the world and still leave the investment disappointing.

3. Look beyond the future everyone is buying

This may have been one of the most practically useful points in Li Lu’s discussion.

When capital concentrates in AI, it need not make the entire market expensive. It can produce a more uneven landscape:

Some assets attract intense enthusiasm while others become cheap through lack of attention.

Li Lu described a deeply divided market and suggested that such disparities could improve the opportunity set for rational investors.

The idea fits naturally with value investing.

Markets do not reliably assign the right price to every asset.

Much of the time, they allocate attention instead.

Capital pursues fashionable industries, analysts cover fashionable companies, and the media follows fashionable stories. Entrepreneurs respond to what investors want to fund. The busiest areas accumulate more information, discussion, and higher valuations. Elsewhere, businesses may receive less research, attract less competition for their shares, and be mispriced.

That raises an obvious question:

If the opportunity is so clear, why does everyone overlook it?

Because investing rationally often demands more of our temperament than our intelligence.

A severalfold rise in an AI stock can make you feel you have missed your chance. Years of stagnant prices in an established business can make you conclude it has no future.

Markets have a way of feeding those impressions:

Rising prices generate reasons to be bullish; falling prices generate reasons to be bearish.

Investors then read a rally as evidence that the market has discovered value, and a decline as evidence that it has uncovered a problem.

A value investor has to return to the business:

What has actually changed in the company since its share price moved?

When the change in business value is much smaller than the change in price, the discrepancy may be an opportunity.

There is an essential qualification, however.

An overlooked asset is not necessarily a bargain.

Some companies deserve the market’s neglect. Declining industries, deteriorating business models, falling returns on capital, and poor management can all turn a low valuation into a value trap.

The task is more demanding than buying whatever happens to be out of fashion:

Find assets whose fundamentals are better than their prices imply, but which are being overlooked as attention concentrates elsewhere.

That is a promising place to direct research in the current market.

The more capital AI attracts, the more deliberately investors should widen their field of view.

There is no need to oppose AI on principle. Attention has an opportunity cost in every market.

When everyone looks in the same direction, better prospective returns may lie elsewhere.

4. China’s economy has two stories running at once

The most useful part of Li Lu’s discussion of China, in my reading, was a way of looking at the economy that went beyond declaring it either troubled or full of opportunity:

Two distinct layers of the Chinese economy are operating at the same time.

One is under pressure from the property adjustment, household balance sheets, and weak consumer confidence.

The other involves industrial upgrading, technological progress, extensive supply chains, growing self-reliance, and stronger global competitiveness.

It is a mistake to collapse them into a single story.

Looking only at the first can make China seem devoid of long-term opportunity. Looking only at the second can obscure the real costs of the property correction, household balance-sheet repair, and insufficient demand.

Li Lu discussed property, household savings, manufacturing, and technological self-reliance in terms of this two-layer economy.

I find that framework particularly useful for investors.

A country’s GDP growth has never been interchangeable with the returns on its stocks.

Shareholders depend on how much growth a business can convert into free cash flow, and the price they pay for a claim on that cash.

One of China’s major changes is the adjustment of a growth model built around property, land, and credit expansion, alongside the rising importance of manufacturing and technology.

Investors cannot simply carry the asset-allocation assumptions of the last decade into the next.

They also need to resist the opposite temptation:

Industrial strength does not make every manufacturer a good investment. Rapid technological progress does not guarantee high shareholder returns.

In industries with heavy capital requirements, fierce competition, and continually expanding capacity, consumers may capture the gains from innovation before shareholders do.

That makes the useful research question more specific than whether to be bullish or bearish on China:

Which businesses can turn China’s manufacturing scale, engineering expertise, and technological advances into sustained returns on capital?

These are very different questions.

If opportunities in Chinese assets exist, they may depend more on selecting particular businesses—alpha—than on a broad bet on the economy, or beta.

The economy is reallocating resources.

The market is reassessing what those resources are worth.

I would look for companies well placed in the emerging economy whose prospects, after careful research, appear insufficiently reflected in their prices.

5. Political division does not mean complete economic separation

In Li Lu’s discussion of US–China relations and the world order, one distinction stood out to me:

The political order and the integrated global market are not the same thing.

Relations can deteriorate, strategic competition can intensify, supply chains can be partly rebuilt, and access to technology can be restricted. None of this means the global economy can quickly separate into entirely isolated blocs.

Decades of globalization have created an intricate division of labor.

Links among capital, technology, goods, talent, energy, and consumer markets cannot all be severed by a handful of policies.

We may be seeing globalization enter a more complicated phase rather than simply end:

Greater political division can coexist with substantial economic interdependence.

Li Lu emphasized the interdependence of global markets and the extensive economic ties between China and the United States, while also stressing the importance of strategic stability.

That places a difficult demand on investors:

Take geopolitics seriously without allowing it to replace economic analysis.

Investors once too readily assumed globalization would continue on its existing course forever. Today, the error can run the other way: greater US–China rivalry is taken to imply complete economic decoupling.

Both assumptions are too simple.

The research has to become more specific:

Which industries will decouple? Which will build operations on both sides? Which will become more regional? Where will security concerns bring additional capital investment?

Some strategic industries may become increasingly domestic. Consumer goods, resources, capital markets, and many intermediate goods may retain considerable cross-border ties.

Investing could become more complicated as a result, but opportunities need not disappear.

Greater political risk may instead widen the differences between companies’ competitive positions, supply-chain roles, asset prices, and cash-flow quality.

For a long-term investor, predicting which country will ultimately prevail is less useful than understanding the business exposure:

Which political risks does this company face? Can it adapt? Will its competitive advantages survive different economic and political conditions?

A national story should not become a stock recommendation by default.

A country can prosper over the long term while many of its companies remain poor investments.

A country facing serious macroeconomic difficulties can still produce excellent businesses, including some the market undervalues.

Making those distinctions is much of the hard work of investment research.

6. Compounding begins with protecting what you have built

The first five themes concern the world around us. Li Lu’s closing discussion of compounding brought the focus back to the investor.

To me, this was an easily overlooked part of the conversation, and one worth returning to over many years.

When investors talk about compounding, they tend to think of three inputs:

Principal, rate of return, and time.

The things that accumulate over a lifetime extend far beyond money.

Knowledge builds on knowledge.

Skills improve through use.

Reputation grows through consistent conduct.

Trust deepens over time.

The benefits of caring for your health accumulate.

Judgment improves as you learn from experience.

Li Lu brought these forms of accumulation together and emphasized a condition for compounding: first, protect what you already have.

It sounds modest, but it may be one of the most consequential principles in investing.

Compounding’s power lies less in a spectacular year than in the enormous difference a small advantage can make when sustained for long enough.

The same logic works in reverse.

A large, permanent loss can erase years of progress.

Avoiding loss is therefore more than an expression of caution. It reflects an understanding of time.

Over a sufficiently long investment horizon, permanent impairment can be a greater enemy than temporary volatility.

The same is true outside investing.

You can spend a decade establishing your credibility and damage it with a single bad decision.

You can spend a decade building knowledge, then watch it become obsolete if you stop learning.

You can spend a decade improving your health and exhaust those reserves through sustained neglect.

With experience, an investor comes to appreciate that:

Finding opportunities and preserving what you have accumulated are inseparable parts of investing.

That helps explain Li Lu’s repeated emphasis on a circle of competence, a margin of safety, and avoiding commitments you do not understand.

These principles manage our vulnerability as much as our money.

The greatest danger may be an opportunity so enticing that it persuades us to accept a loss we cannot afford.

What survives a changing world

If I had to put my reading of this discussion into a single sentence, it would be this:

AI can change how we produce things, China can reshape the global economy, political arrangements can shift, and markets can create fresh bubbles—all while the basic logic of value investing holds.

A remarkable technology still requires discipline about price.

An ambitious story still needs cash flow to support it.

An exuberant market makes a margin of safety more valuable.

A complicated economy demands that we distinguish national growth from business value.

Geopolitical tension makes it necessary to distinguish political arrangements from economic ties.

A long life gives us more reason to protect our knowledge, capital, reputation, health, and judgment.

That is what I would take away from Li Lu’s remarks.

He offered no simple answer to the question of which stock to buy next.

I found a more useful reminder:

Investing requires a way of judging the future that can withstand uncertainty, rather than a prediction that depends on getting everything right.

AI may become more powerful than we expect and pass through more speculative cycles than we imagine.

China’s economy may prove more resilient than pessimists anticipate, while its adjustment takes longer than optimists hope.

US–China relations may worsen, or mutual dependence may give the two countries reasons to find another equilibrium.

No one can be certain of those outcomes.

The more dependable anchors are value, returns on capital, economic constraints, and responsibility for our own decisions. Prices eventually have to answer to the value beneath them; investment has to justify the capital it consumes; technology operates within economics; and investors bear the consequences of their judgment.

The skill worth cultivating over time is recognizing what continues to matter amid the noise.

I would end with three questions:

First, if AI progresses at half the pace I expect, would the assets I hold still be worth owning?

Second, if the market stops rewarding my preferred narrative for three years, could I still earn a reasonable return?

Third, are my investments and life choices building assets that can compound—or spending what I have already built on a short-lived thrill?

If you can answer all three satisfactorily, the market’s next move may matter rather less.

The author’s summary and reflections on Li Lu’s October 5, 2026 discussion at Columbia University. This essay is not investment advice.