Can the AI Boom Continue to Run on "Debt"?—The True Deadline for Recouping Massive Investments

The frenzy surrounding generative AI has already transcended being a mere buzzword in the tech industry. Services like chatbots, image generation, code generation, and AI agents are becoming part of everyday use, prompting companies to hasten AI adoption to maintain competitiveness. In the stock market, semiconductor, cloud, data center, and power infrastructure stocks are garnering attention, and AI has become a dominant investment theme.

However, the question now being asked is not the technical one of "Is AI really amazing?" The more pressing question is the economic one: "When and how will these massive investments be recouped?"

An article on aktiencheck.de, introducing the analysis by AllianceBernstein economist Eric Winograd and others, tackles this issue head-on. AI-related investments are already boosting the U.S. economy, but for their effects to continue, merely increasing capital expenditures won't suffice. After the phase of investing in data centers and semiconductors, AI must actually generate productivity improvements. In other words, what the market expects is not the "investment boom" itself, but the "productivity boom" that follows.

AI Has Already Become Part of the Macroeconomy

The scale of AI investment has already outgrown the framework of individual companies' growth strategies. According to the article, investments in computers, information processing equipment, software, and other AI-related areas have significantly expanded in the U.S. over the past few years. Notably, the investment expansion by major cloud providers like Amazon, Google, Meta, Microsoft, and Oracle, known as hyperscalers, stands out.

According to AB's analysis, the total capital expenditure of these major companies was less than $100 billion in 2021, but it is estimated to increase to $768 billion by 2026 and nearly $1.6 trillion by 2030. Looking at these numbers alone, it is clear that AI is not a "future story" but already a massive source of demand driving current economic activity.

Building data centers requires land, building materials, power equipment, cooling systems, communication networks, servers, GPUs, memory, and specialized personnel. As AI companies increase capital expenditures, not only semiconductor manufacturers but also construction, power, air conditioning, copper wire, fiber optics, and financial institutions benefit. The broadening of AI-related stocks in the stock market is due to such ripple effects.

In that sense, the AI boom is not just a "pipe dream." Factories are actually operating, data centers are being built, jobs are being created, and corporate profits are rising. This is why the market has evaluated AI as a growth driver for the U.S. economy.

But the problem starts here.

It's Not the "Investment Amount" but the "Rate of Increase in Investment" that Drives Growth

When considering the impact of an investment boom on economic growth, it's not just the size of the investment that matters. What matters is how much it has increased from the previous year, in other words, the growth rate of the investment.

For example, if a company invests 1 trillion yen this year, but also invested 1 trillion yen last year, the additional growth-boosting effect is limited. On the other hand, if it was 500 billion yen last year and 1 trillion yen this year, that increase will strongly impact economic growth.

According to AB's analysis, the growth rate of AI-related capital expenditures is expected to peak at about 85% in 2024, then decline to about 76% in 2026, and further slow down towards 2030. Even if the absolute amount of investment remains high, if the growth rate falls, the direct contribution to GDP growth will weaken.

Here lies the essential turning point of the AI investment boom. In the initial stage, the economy is boosted simply by companies pouring massive funds into data centers and GPUs. However, in the next stage, the slowdown in investment growth must be compensated by productivity improvements brought about by AI.

AB suggests that even if AI-related capital expenditures boost U.S. growth by about 1.5 points in 2026, that contribution could roughly halve by 2030. In other words, for AI to remain the main driver of economic growth, a handover from capital investment to productivity is necessary.

If this handover succeeds, the AI boom will become a story of sustainable growth. If it fails, it will turn into a story of overinvestment and premature expectations.

The Bottleneck Is Not Just GPUs

In the early stages of the AI boom, the focus was on "not enough GPUs." Securing NVIDIA's high-performance GPUs was thought to determine the outcome of the AI development race. Of course, semiconductor supply constraints are still important. However, as the scale of investment swells, the bottleneck is not limited to GPUs.

Data centers require vast amounts of land. A power grid capable of supplying large amounts of electricity stably is also necessary. Water and cooling equipment to cool servers are indispensable. Personnel involved in construction, electrical work, plumbing, security, and communication infrastructure are also essential. Furthermore, rising prices of GPUs, CPUs, and memory squeeze the profitability of AI investments.

AI is often spoken of as "magic in the cloud," but behind it lies an extremely physical and capital-intensive infrastructure. Data centers are built on real land, consume real electricity, and incur real cooling costs. The more digital the AI industry becomes, the more substantial infrastructure investment it requires.

This point is frequently discussed on social media. In Reddit's investment communities, questions like "Who is actually investing big money in AI?" and "Are there companies really making significant revenue from AI other than semiconductor manufacturers?" are being posted. In one post, concerns were raised about hyperscalers investing hundreds of billions of dollars in data centers while clear AI revenue disclosures are limited.

On the other hand, there are opposing views. Some argue that giant companies like Microsoft, Google, Amazon, and Meta have very strong cash flows from their core businesses, and even if some AI investments fall short of expectations, the entire company won't collapse. Especially companies with existing businesses in advertising, cloud, e-commerce, and enterprise software already have a profit-generating foundation, unlike the dot-com companies of the past.

This conflict is important. The AI boom cannot simply be dismissed as a "bubble," nor can it be assumed to "absolutely succeed." What is happening now is that companies with huge profits are making upfront investments to capture an even larger future market. The outcome may vary significantly from company to company.

The Next Focus Is Shifting from "Internal Funds" to "External Funds"

Another important point emphasized in the aktiencheck.de article is the funding structure of the AI boom.

So far, the AI investments of giant tech companies have been mainly funded by internal cash flows. In other words, profits earned from existing businesses are reinvested in AI infrastructure. This makes the risk relatively visible to investors. As long as growth investments are made within the range of profits, excessive credit risk is unlikely to arise.

However, as the scale of investment further expands, internal funds alone may not suffice. AB points out that from 2027 onwards, many companies may become more reliant on bond and equity markets. In other words, the AI boom is expanding from a "tech stock story" to a "credit market story."

This is a critically important change. If companies expand AI investments through borrowing or bond issuance, the success or failure of AI becomes an issue not only for shareholders but also for creditors. Furthermore, if not only hyperscalers but also suppliers and data center operators increase their funding, AI-related leverage will spread across the entire value chain.

On social media, interest in this "AI debt" is growing. On Reddit, discussions are seen with themes like "I'm more scared of the increase in debt than the AI capital investment itself." Especially for companies like Oracle, which are advancing large-scale investments against the backdrop of AI infrastructure demand, there are concerns about the burden on their balance sheets. On the other hand, companies like Microsoft and Amazon are argued to be issuing bonds not because they are struggling with cash flow, but because they can secure funds at low cost.

This discussion indicates the maturation of the AI boom. The early market focused on "which companies are growing their sales with AI." But now, the focus has shifted to "who is bearing the cost of that investment" and "can they generate returns that exceed the cost of capital."

In the world of investment, a growth story alone is not enough. No matter how attractive the future may seem, if the capital cost required to realize that future is too high, shareholder value will be eroded. The next test of the AI boom lies precisely here.

On Social Media, Views Are Becoming More Complex Than Just "Bubble or Boom"

Reactions on social media regarding AI are not simply divided into optimism and pessimism. Rather, what stands out is a more nuanced view: "AI is real, but not all investments will pay off."

On LinkedIn, posts comparing the AI investment boom to the dot-com bubble are gaining attention. One poster acknowledges the potential of AI to change society but questions whether current valuations are significantly anticipating future success. In the comments, opinions like "It might be a bubble in the short term, but it will change the world in the long term" and "Companies that use AI practically to create new value will grow, but those that are just stories will collapse" are seen.

This reaction is very similar to past technology booms. Railroads, electricity, the internet, smartphones, and the cloud all changed society. However, not all companies invested in during each boom period succeeded. The internet was real, but many dot-com companies disappeared. Communication infrastructure was necessary, but excessive fiber optic investment led to losses for many investors.

The same can be said for AI. The reality of AI technology and the justification of all current valuations are separate issues.

On Reddit, similar discussions are taking place. In a post with the theme "AI is real, but the AI bubble is also real," the question of what proves the value of AI is raised. Is it sales, profits, productivity improvements, cost reductions, or full-scale adoption by companies? The poster expresses caution against simply thinking, "AI seems important, so eventually, the economics will catch up."

This aligns precisely with AB's analysis. The long-term growth contribution of AI must be proven not by the size of capital investment but by productivity improvements.

What Does It Mean for AI to Increase Productivity?

The term "productivity improvement through AI" is often used, but its content is surprisingly vague.

At the individual level, the effect of AI in shortening work time by assisting in email creation, document preparation, code generation, research, translation, and summarization is already being felt. Examples of programmers using AI coding assistance, marketers generating ad copy, and customer support implementing chatbots are increasing.

However, to boost productivity at the macroeconomic level, individual work efficiency alone is not enough. The entire business process of a company must change to produce more value with the same number of people or achieve the same sales at a lower cost. Furthermore, this must spread not only to some tech companies but also to a wide range of fields such as manufacturing, logistics, finance, healthcare, education, retail, and administration.

AB points out that while U.S. productivity has improved in recent years, referencing research from the San Francisco Fed, this improvement may still be largely due to labor market-related factors rather than a broad technology-driven productivity increase. A massive productivity wave that lifts the entire economy, like the IT revolution of the 1990s, has not yet been clearly observed.

There is a wide range of views among researchers. Some believe AI will hardly boost labor productivity growth rates over the next decade, while others expect a significant boost of over 3% annually. The average expectation is about a 1% contribution, but the prediction range is too wide to draw definitive conclusions.

In other words, the market is still placing a high value on "pre-proof" expectations.

Can AI Agents Become "Colleagues You Don't Want to Fire"?

An interesting perspective emerging on social media is whether AI agents can be seen not just as tools but as entities that continuously generate value within organizations.

A LinkedIn post suggested that to prove AI investment is not a bubble, companies need to treat AI agents as "positions on the organizational chart that you don't want to fire," creating enough value to justify their existence. Simply paying a monthly fee to make document creation a bit easier is too weak to justify trillion-dollar infrastructure investments. AI needs to handle tasks that companies truly want to continue paying for, tasks that won't be cut even in bad economic times, and tasks that humans can't handle.

This is a very important perspective. Much of the current use of generative AI is still close to being a "convenient auxiliary tool." However, to recoup massive capital investments, AI needs to penetrate core business operations and become a source of ongoing expenditure.

For example, if AI autonomously handles parts of software development, processes customer interactions 24/7, automatically generates sales materials, undertakes initial stages of financial analysis and legal reviews, and optimizes factory operations, companies will treat AI spending not as mere costs but as business infrastructure.

If this stage is reached, AI investments are likely to pay off. However, if AI use is limited to improving efficiency in some tasks and does not lead to clear revenue increases or labor cost reductions, the returns on investment will be questioned.

The Stock Market Has Already Priced in Much

Another issue is that the market has already priced in a lot of AI's success.

AI-related stocks, particularly in semiconductors, memory, cloud, data centers, and some power-related companies, have been heavily bought. Another AB commentary suggests that the narrative of AI capital investment has become very dominant in the stock market, with the rise in stock prices of AI-related companies overshadowing the overall market movement.

It is natural for the market to price in expectations. Investors form stock prices by discounting future profits to present value. However, if expectations are too high, merely "good" results will not suffice. Companies must continue to deliver "very good" or "better than expected" results, or stock prices may easily fall.

This point is repeatedly highlighted on social media. Even if AI changes society, it doesn't necessarily mean investors are buying its benefits at an appropriate price. The view is that technological success and investment success are separate matters.

Particularly at risk are small-cap stocks valued solely on the word "AI" and companies with unestablished revenue models. AB also points out that if the growth rate of AI capital investment declines, valuations of smaller, more speculative stocks could be pressured. While funds are abundant, they are bought for "future potential," but once investors enter a phase of scrutinizing capital efficiency, selection will proceed rapidly.

In the next phase of the AI boom, being merely "AI-related" will not suffice. It will be questioned which companies have real demand, which companies have pricing power, and which companies generate cash flow exceeding investment costs.

Another Warning from the Bond Market

AB's article also focuses on long-term U.S. Treasury yields. There are