Not Allocating AI Assets Is Itself a Risk

When the Wealth Logic Changes: From Steel and Concrete to Trillion-Scale Computing Power

If your investment portfolio has zero AI exposure, that itself might be a risk. This isn't alarmist rhetoric—it's the reality following a fundamental shift in where technology capital is flowing.

A landmark signal: Tencent's capital expenditure exceeded 50 billion yuan in Q2, doubling year-over-year, with the company explicitly stating it plans to continue expanding. Keep in mind that Tencent's core business growth isn't particularly high—so why spend so aggressively? Simply put, even a giant like Tencent is afraid—afraid that failing to invest in AI means being completely left behind in the next round of competition.

This figure is remarkable even among global tech giants. For comparison, Meta's capital expenditure in the same period was approximately $8 billion (about 58 billion yuan), primarily directed toward AI infrastructure. This phenomenon is known in the industry as the "AI arms race"—major tech companies racing to build GPU clusters, train large models, and construct AI ecosystems. Tencent's anxiety comes from multiple directions: ByteDance's Doubao large model rising rapidly, Alibaba's Tongyi Qianwen open-source strategy eroding market share, and Baidu's ERNIE Bot seizing first-mover advantage. In this competitive landscape, hesitation in capital spending could mean permanent disadvantage in large model capabilities and cloud computing market share.

When a company flush with cash and deeply versed in business fundamentals chooses to "go all in" rather than play it safe, that itself sends a signal: the underlying wealth logic is shifting.

The Money-Making Logic of the Past 30 Years: Borrow, Invest, Collect Rent

To understand today's choices, we first need to see clearly the money-making logic at the national level over the past decades. It's actually no different from personal finance principles—just three steps: borrow, invest, collect rent.

The First Pot of Gold: Using Real Estate to Lock In Future Income

Industrialization requires massive capital—where does the money come from? The answer is getting ordinary people to willingly hand over their money, and housing was the most effective tool. Apart from buying a home, there's virtually nothing else that can lock in a family's income for the next 30 years in one shot.

Ordinary people take on mortgages, and the government uses land sale revenues to build high-speed rail and factories—that's the source of the first pot of gold. In essence, real estate was used to shift future income into today's spending.

Over more than two decades of urbanization in China, a unique "land finance" system took shape. Local governments earned revenue by transferring land use rights—at the 2021 peak, national land transfer fees reached 8.7 trillion yuan, accounting for over 40% of local fiscal revenue. Simultaneously, commercial banks used residential mortgages to discount 20-30 years of future household income into present purchasing power. As of 2023, China's household leverage ratio reached 63%, approaching developed-country levels. The essence of this mechanism is intertemporal resource allocation—trading future consumption capacity for present infrastructure funding. But when price-to-income ratios exceed 30x in first-tier cities and household savings rates decline, this model's marginal utility drops sharply.

But this logic no longer works. The "six wallets" have been emptied, mortgage payments consume half of income, and young people are refusing to marry or have children. The old engine is stalling, but debts still need to be repaid.

The Second Pot of Gold: Moving Debt Onto the World's Balance Sheet

What do you do with domestic overcapacity? Move it overseas. Sell EVs to Europe, ships to the Middle East, and cover the globe with solar panels.

The essence of this approach: when you buy my car, it's your citizens taking on auto loans; when you import my goods, it's your country bearing the trade deficit. This is called moving debt onto the world's balance sheet.

Take new energy vehicles as an example—in 2023, China's auto exports reached 4.91 million units, surpassing Japan for the first time to become the global leader. BYD, SAIC, and other brands are building factories or establishing sales networks in Southeast Asia, the Middle East, and Europe. When overseas consumers purchase Chinese-brand vehicles through installment loans, they're essentially using local credit systems to absorb China's overcapacity. Similarly, China's shipbuilding industry captured over 60% of global new orders in 2023. The risk in this model lies in geopolitics—the EU imposing tariffs on Chinese EVs and America's "small yard, high fence" strategy are both attempting to block this debt transfer channel.

The Ultimate Form: Collecting Rent While Lying Down

What if even cars and ships become unsellable in the future? The ultimate form is rent collection.

Just look at the United States. Its manufacturing has been hollowed out, yet it remains the global leader. Because the operating systems are theirs, the chip architectures are theirs, and global settlements run through their systems. iPhones assembled in China earn hard labor money—the bulk of patent fees, licensing fees, and financial fees ultimately flow back to California. This is the ultimate form of debt transfer—collecting rent while lying down.

America's "rent economy" is built on several key infrastructure layers. The first layer is operating systems and chip architectures: Windows and macOS monopolize PCs, iOS and Android monopolize mobile, and ARM and x86 architectures cover virtually all computing devices. The second layer is financial infrastructure: the SWIFT system processes global cross-border payments, while Visa and Mastercard cover global payment networks. The third layer is the intellectual property system: Qualcomm earns over $6 billion annually from patent licensing alone, and Apple's App Store takes a 30% commission. The essence of this model is controlling "standards" and "protocols"—once something becomes the global standard, users must continuously pay. This is also why China's breakthroughs in AI are seen as a critical opportunity to break this rent-collecting structure.

The Fundamental Difference Between AI Infrastructure and Traditional Infrastructure

With the "borrow—invest—collect rent" framework understood, today's situation becomes clear. The "15th Five-Year Plan" mentions 26 trillion yuan across six networks, three of which—the computing power network, the power grid, and the communications network—are directly related to AI. This is essentially spending money to purchase future rent-collecting rights.

The six networks include: computing power, electricity, communications, water conservancy, transportation, and logistics. The first three AI-related networks are expected to account for a significant portion of total investment. The core of the computing power network is deploying intelligent computing centers nationwide, aiming to elevate China's AI computing power from the current ~300 EFLOPS to higher levels. The power grid must solve the enormous energy consumption of AI data centers—a 10,000-GPU cluster's annual electricity consumption equals that of a medium-sized city. The communications network requires 5G-A and 6G low-latency networks to support real-time AI inference applications. While the 26 trillion yuan investment is enormous, compared to the 2008 four-trillion stimulus (equivalent to about 10 trillion today after adjusting for inflation), the annual investment intensity spread over five years is roughly comparable.

But there's one fundamental difference from the 2008 "four trillion."

The Broken "Employment" Link

Back then, building high-speed rail and highways didn't generate short-term profits, but it drove massive employment. When steel and concrete were in motion, migrant workers earned money and turned around to buy homes and appliances—consumption picked up. The infrastructure → employment → consumption chain was fully connected.

This round of AI infrastructure is different. Banks are generous with credit lines, and money is being deployed, but it barely drives employment. How many people does it take to build a computing center? Large AI models don't just fail to create jobs—they actively eliminate them. Thus, infrastructure drives capital returns, but the "employment" link in the middle is broken, so consumption naturally can't keep up.

The difference in employment generation between AI infrastructure and traditional infrastructure is structural. In the 2008 four-trillion stimulus, every 100 million yuan invested in high-speed rail construction directly created about 2,000 jobs, with even more through indirect multiplier effects in the supply chain. But a modern AI data center might require investments of billions of yuan while needing only dozens to hundreds of people for daily operations. The deeper contradiction: once AI models are trained, their applications actually replace white-collar jobs—customer service, translation, basic programming, and data analysis are being rapidly eroded by large models. McKinsey's 2023 report predicts that by 2030, up to 300 million jobs globally could be affected by generative AI. This creates a paradox: AI infrastructure may generate high capital returns, but these returns flow primarily to capital owners rather than workers.

This also explains why current goods consumption growth is barely above 1%, and why the stock market stalls the moment momentum fades.

How Ordinary People Can Participate in AI Dividend Distribution

The key question is how ordinary people can participate in the dividend distribution.

Previously, with traditional infrastructure, ordinary people could directly participate by selling their labor, and indirectly benefit through property appreciation. But in this round of AI dividends, ordinary people can neither participate through labor nor share through real estate.

In the traditional economic model, ordinary people had three main channels to participate in economic growth dividends: labor income (employment), asset appreciation (property), and capital gains (stocks). When the first two channels narrow simultaneously, the stock market becomes the only viable means of participation. Looking at global experience, in the past decade of US wealth growth, the S&P 500 rose over 200%, with tech stocks contributing the lion's share. The market cap growth of the "Magnificent Seven" (Apple, Microsoft, Google, Amazon, NVIDIA, Meta, Tesla) alone exceeded the GDP of many countries. While China's market follows a different path, the logic is similar—when AI industry chain profits concentrate in a few leading companies, holding equity in these companies is the most direct way to participate in dividend distribution.

So what's left? Only stocks as a channel.

This also answers the opening question from an investor's perspective: Tencent fears being left behind—how are ordinary investors any different? When the nation's wealth logic has already shifted from steel and concrete to trillion-scale computing power, if your portfolio still only holds soy sauce and baijiu stocks, that's probably not "conservative"—it's a bet. A bet that this "intelligent debt migration" won't work, a bet that the old logic can still survive.

How to Rationally Allocate AI Assets: Balance, Not All-In

Here's a telling detail—even Buffett is voting with his feet on this logic. He's reducing traditional asset positions while adding to Google. The principle is the same: the true "rent-collecting stocks" of the future will most likely be held by those who own the foundational AI technologies.

Buffett's 2024 investment moves attracted widespread attention: Berkshire Hathaway significantly reduced its Apple position (from the top holding to roughly one-third), while initiating a position in Google's parent company Alphabet in late 2023. For traditional value investors, this is an important signal. Buffett has long been cautious about tech stocks, once stating he "doesn't invest in things he doesn't understand." But Google's investment logic fits his "moat" theory—over 90% search market share, YouTube's advertising monopoly, Google Cloud's deep AI positioning, and over $80 billion in annual free cash flow. This is essentially an "AI-era rent-collecting stock"—a company that controls underlying technology standards and user entry points.

But this absolutely doesn't mean blindly going heavy on tech. Tech stocks are volatile with high valuations—completely different from consumer stocks. The key is achieving allocation balance: you need to give AI a seat at the table while controlling the volatility risk it brings.

It all comes down to one sentence: when reading the news, don't just watch the spectacle—think about whose pocket this money is moving from and into. Once you understand this wealth transfer pathway, whether to allocate to AI and how much becomes largely self-evident.