The greatest investment direction for humanity

In the past two years, the capital frenzy triggered by large AI models has swept the globe with unprecedented intensity.

NVIDIA Corporation (NASDAQ: NVDA), from 2022 to the end of 2025, saw its stock price increase over 15 times, with a market value once exceeding $5 trillion;

Google A (NASDAQ: GOOGL), from 2023 to now, has seen its stock price grow 3.5 times, with a current market value of $4.7 trillion;

TSMC (NYSE: TSM), since its low point in 2022, has risen nearly 6 times, with a current market value exceeding $2 trillion;

Broadcom Inc. (NASDAQ: AVGO), during the same period, surged over 9 times, with a market value also surpassing $2 trillion;

Micron Technology (NASDAQ: MU), in the past year, skyrocketed 7.5 times, with a current market value over $720 billion;

Intel Corporation (NASDAQ: INTC), within one year, surged 4.5 times, with a current market value over $540 billion;

SanDisk (NASDAQ: SNDK), since its listing in February 2025, has soared over 40 times in just over a year;

There are also many popular AI industry chain companies, with stock price increases of several times or even dozens of times within just a year.

Many haven’t yet realized how exaggerated these numbers truly are.

Let’s put it this way: as of May 6, just 15 leading giants in the US stock AI industry chain (NVIDIA, Apple, Google, Microsoft, Amazon, TSMC, Broadcom, Meta, Tesla, Micron, AMD, Intel, ASML, Oracle, Palantir), their total market value combined exceeds $30 trillion.

The latest total market value of US stocks is about $92.5 trillion, and these 15 companies account for one-third of that;

The total market value of A-shares is approximately 118.27 trillion yuan, equivalent to $17.25 trillion, which is 1.75 times the combined market value of these 15 giants!

By 2025, the nominal GDP of the United States is projected to be $30.8 trillion, roughly equal to the total market value of these 15 companies!

This storm isn’t just happening in the US stock market,

Korean Samsung Electronics (KS: 005930), since the beginning of the year, has seen its stock price increase over 3.7 times, with a market value surpassing $1.04 trillion;

Korean SK Hynix (KS: 000660), during the same period, increased over 3.6 times, with a market value reaching $50k;

On the A-share side, Industrial Fulian’s stock price tripled in a year, with a market value once exceeding 1.6 trillion yuan, becoming the first trillion-yuan market cap giant benefiting from AI industry;

Zhongji Xuchuang, up over 10 times in a year, with its market value at its peak also surpassing one trillion yuan;

There are also many leading companies in chips, optical communications, and PCBs, whose market values have also soared several times in a short period, becoming billion-yuan giants.

It can be said that this is undoubtedly the most extreme global asset boom in human history.

Faced with such an epic wave of asset frenzy, some pull out old newspapers from the 2000 internet bubble, trying to prove that everything will eventually fade away.

But they forget one thing.

In human history, no true civilization-level revolution has ever stopped because it “looked too expensive.”

The same applies this time.

01

Looking at human civilization history, it is essentially a story of continuous technological breakthroughs overcoming physiological, environmental, and cognitive limitations.

One rule has never been broken: whoever masters the core technology of the era, whoever holds the key to wealth.

In the agricultural era, iron tools and water conservancy were core technologies; those who mastered advanced farming techniques became the biggest beneficiaries of their time;

In the industrial era, steam engines and electricity were core technologies; they gave rise to giants in steel, energy, and manufacturing, creating companies that spanned centuries;

In the information era, the internet is the core technology; tech giants like Apple, Google, Microsoft, Meta, Tencent, Alibaba created unprecedented mythologies of market value;

And today, the core technology of the intelligent era is computation power.

This time, the carrier of computation power is ultra-large-scale AI clusters.

Tens of thousands, hundreds of thousands of GPUs connected together, like a giant power station of the digital age. The electricity consumed to train a cutting-edge large model alone is enough to power an ordinary household for hundreds of years.

And such clusters are rising worldwide at a pace of several each week.

Who controls the largest computing clusters, defines the upper limit of intelligence.

OpenAI, Google, Meta, xAI… Their competition, on the surface, is a race of model parameters, but in essence, it’s a military buildup of computing power reserves.

Elon Musk plans to build a cluster of 100k H100s, while Jensen Huang says future data centers will be “AI factories”—these giants are all openly craving extreme computing power.

They are frantically buying computing resources, not out of madness, but out of clarity.

The greatest value of computing power lies in its demand being “perpetual and unlimited.”

In the future, human society will fully move toward intelligence, and the penetration of computing power will be everywhere, like water and air, becoming an essential for survival and development, with demand never exhausted and continuously expanding.

For example, in industry, behind high-end terms like industrial unmanned, intelligent manufacturing, digital twin, industrial simulation… all require massive computing power to realize decision-making intelligence, automated production, cost reduction, and efficiency improvement;

In daily life, the popularization of autonomous driving, humanoid robots, smart homes, AI assistants, metaverse scenes—all need real-time computing power to ensure experience, truly integrating technology into everyday life and changing human lifestyles;

Especially in frontier exploration, breakthroughs in life sciences (drug development, gene sequencing, anti-aging research), space exploration (interstellar simulation, spaceflight computing, space colonization), quantum technology, controlled nuclear fusion—all require top-tier computing power to help humans break cognitive boundaries.

More importantly, the demand for computing power will not decrease due to economic cycles or policy adjustments; instead, it will continue to grow with the advancement of the intelligent era.

This “limitless, irreversible” rigid demand is an advantage that no traditional industry can match.

02

In the intelligent era, three fundamental elements are—computing power, electricity, and data.

But why is computing power the most core?

Let me give an analogy.

Data is the “crude oil” of the new era. It is embedded in every click, every conversation, every image. By 2025, the total data generated globally each year will exceed 175ZB—stacked together, it would require hundreds of millions of 12TB hard drives to store.

But without computing power to refine it, this crude oil is just mud buried underground.

Electricity is the “blood” of the new era. It drives everything, but blood itself doesn’t think or decide.

The electricity consumption of AI clusters is indeed expanding at an astonishing rate, to the point that some data centers have to be built next to nuclear power plants. Power grid companies are even beginning to require AI computing clients to submit electricity applications three years in advance—something never seen before in history.

But without computing power, everything resets.

Without computing power, data is just a pile of silent bits, and electricity can only light up bulbs, not smart devices.

This is the true meaning of “cornerstone”—it’s the stone beneath the foundation; remove it, and the entire building collapses instantly.

Imagine the history of electricity.

Once humans used electricity, going back to nights without it became unthinkable.

Light bulbs, telephones, TVs, computers… every invention deepened our dependence on electricity until it became as essential as air.

Computing power is undergoing the same process.

Today, you are completely inseparable from your phone, computer, and electronic devices; every app you open, every word you type consumes massive amounts of computing power behind the scenes.

You might not even realize that when you take a photo with your phone, dozens of AI models are running simultaneously—face detection, scene recognition, HDR fusion, noise reduction… each requiring computing power.

By 2030, a “day without computing power” will seem as absurd as a “day without electricity” today.

Once computing power invades a field, it never recedes.

03

This is not even the whole story.

The most fascinating part of computing power is—human greed for it has no limit.

You can buy fewer cars, fewer phones, or drink fewer coffees. Economists call this “demand elasticity”—when prices go up, people naturally buy less.

But you wouldn’t want AI to “think a little less.”

And if you want AI to be a bit smarter, the required computing power must multiply many times over.

Deloitte’s report predicts that by 2026, reasoning will account for two-thirds of AI computational load; by 2030, this proportion could reach 75%.

This means that the demand for computing power is not a straight line upward but an accelerating steep curve.

Because if one hundred million people worldwide ask AI ten questions each day, the reasoning computation used that day is equivalent to retraining GPT-4.

On the other hand, the supply side of computing power is like running in a swamp.

Currently, chip process technology is approaching physical limits. 3nm, 2nm, 1nm… each step is as difficult as climbing a mountain.

TSMC’s 2nm factory investment exceeds $30 billion, and it takes four or five years from construction to mass production. Even with technological breakthroughs, ramping up capacity isn’t instant.

Globally, only one company can produce the most advanced extreme ultraviolet lithography machines—ASML from the Netherlands.

A single machine costs $350 million, with only dozens produced annually. Each weighs 180 tons, requiring three Boeing 747s to transport.

Want to increase capacity? Sorry, Zeiss’s optical lenses can’t be ground out, and the laser equipment from Trumpf can’t be made.

And energy.

The power consumption of a large AI cluster is comparable to that of a medium-sized city.

In the US, many newly built “data center alleys” have had to suspend operations because the local power grid can’t handle the load, with new projects waiting years to connect.

Computing power companies are starting to build their own power plants and buy nuclear reactors—what was a joke a year ago is now a common practice.

In other words, demand for computing power is skyrocketing exponentially, but supply is shackled.

This “super gap” will be the structural theme of the next decade.

04

Looking ahead to the next ten years, the path of the computing power industry chain is extremely clear:

NVIDIA, Broadcom, AMD, and those behind the scenes but indispensable HBM manufacturers (SK Hynix, Micron), advanced packaging (TSMC’s CoWoS, ASE), semiconductor equipment providers (ASML, Applied Materials, KLA).

No matter who wins or loses in future AI models, these “shovel sellers” will profit.

In the gold rush, the most profitable and safest are the shovel sellers.

This principle has never failed in hundreds of years.

In the 1849 California Gold Rush, the real winners weren’t most prospectors, but Levi’s, which sold shovels and jeans.

Today, these AI “shovel sellers” are selling the most expensive, rarest, and most indispensable “shovels” in human history.

There are also optical modules, high-speed interconnects, switches… Every time the computing cluster doubles in size, the demand for interconnection bandwidth multiplies. From 800G to 1.6T to 3.2T, the upgrade speed of optical modules far exceeds the traditional “generation every four years.”

Why can optical communication stocks in US and A-shares rise several times or even dozens of times in a year? Because they are the “neural networks” of the computing power era. Without them, even the strongest chips can only fight alone, unable to form a legion.

There are even future-looking technologies like quantum computing, photonic computing, in-memory computing, and edge AI computing… all of which could produce new computing power giants in the next decade.

05

Final Words

The essence of investment is the realization of cognition; and the height of cognition determines the upper limit of returns.

We should step out of short-term stock market speculation and think more from the perspective of human civilization evolution, national technological development, and industry technological iteration to understand the value of computing power and broaden our horizons.

As the core foundation of the intelligent era, computing power’s fundamental logic of “irresistible demand, unassailable position, insurmountable barriers, and unlimited growth” makes it undoubtedly one of the greatest investment directions today.

Investing in computing power doesn’t even require you to believe in any particular company, CEO, or technological route.

All you need to believe is one thing:

Humanity’s desire for intelligence is an instinct engraved in our genes. And computing power is the physical embodiment of all intelligence.

Perhaps twenty years from now, people will look back at 2025–2030 and say:

That was the 1860s of the railway age, the 1900s of the electricity era, the 1995 internet era.

But this time, the wave is more certain, more surging, and more unstoppable than ever.

Faced with such an exciting new era and new world,

Will you embrace it? Or ignore it?

The latest AI-selected stocks for May are now released—does NVIDIA catch your eye?

InvestingPro members, see the list of AI-selected stocks here.

Not subscribed yet? Subscribe here for up to 50% off!