1. AI Ecosystem Cash Flow
The reason why semiconductor companies' stock prices have risen more than hyperscalers over the past year is relatively simple.
Hyperscalers reinvested most of the money they earned from AI into data center construction and AI infrastructure investments.
In contrast, that investment ultimately translated into sales and profits for semiconductor companies such as Nvidia.
The entity that continues to bring in external capital into the AI ecosystem is essentially hyperscalers.
When general companies pay for cloud, advertising, software, etc., that money goes to hyperscalers, who then use it to build AI infrastructure. And a significant portion of that money flows to semiconductor companies in the form of GPUs and memory.
Simplifying the flow of cash:
External company → Hyperscaler → Semiconductor company
Conversely, the flow of services and products is the opposite.
Semiconductor → Data center → AI service → External company
In other words, semiconductors are at the end of the ecosystem where they ultimately recover value.
The problem is that hyperscalers have made massive AI investments but haven't yet reaped returns commensurate with those investments.
Ultimately, their cash flowed mostly to semiconductor companies within the AI ecosystem, and the cash flow of hyperscalers could inevitably weaken.
2. So why did hyperscalers decline less?
Over the past year, the market hasn't seriously questioned the sustainability of the AI ecosystem.
It believed that AI investment would continue to expand and assigned high value to semiconductor companies, which are realizing profits most quickly.
However, about two months ago, the atmosphere changed.
For several reasons, doubts began to emerge in the market as to whether the AI economy could indeed sustain its current scale.
But here's an interesting phenomenon.
If the AI boom collapses, hyperscalers, who have invested the most money in AI, should also suffer a major blow.
Nevertheless, in the actual market, hyperscaler stock prices often declined less than semiconductor companies or even rose.
This phenomenon is not easily understood by simply saying that "AI investment is burdensome."
Originally, shareholders often dislike large investments. This is because investments involve uncertainty and risk.
Therefore, it is common for stock prices to rise when a reduction in investment is announced and to fall when a significant increase in investment is announced.
One of the reasons why Apple has recently shown relatively strong stock price movement is here.
Unlike other big tech companies, it hasn't poured astronomical capital into AI. Therefore, there is an expectation that if the AI boom slows down, the damage will be relatively small.
However, this quarter, Google announced that it would further increase its AI-related CAPEX.
Microsoft and Amazon are also expected to significantly expand their investments.
In other words, the market is worried about large-scale investment cuts, but instead, they have announced increases.
Despite this, Google, Microsoft, and Amazon's stock prices remained strong, while semiconductor-related stocks plummeted.
How can we interpret this difference?
3. Solution
The answer to this question can be found in the two structures explained above.
First, the market has begun to view AI and semiconductors as separate sectors,
Second, it is important to understand how cash flows within the AI ecosystem.
Nvidia's core business is essentially selling AI chips and memory (associative computing power) to hyperscalers.
Hyperscalers, on the other hand, already have massive existing businesses.
Advertising, cloud, e-commerce, enterprise software, etc. They use cash generated from these existing businesses to invest in a new business called AI.
In other words,
Hyperscalers are not companies that survive solely on AI.
On the other hand, the current boom in the semiconductor industry is heavily dependent on the expansion of AI investment.
As explained earlier, the nutrients of the AI ecosystem come from the cash that hyperscalers bring in from outside.
To put it figuratively, hyperscalers are like bats that fly out of the cave to get food and return to supply nutrients inside the cave.
The ecosystem within the cave is sustained thanks to those nutrients.
Semiconductor companies can be seen as creatures at the end of that ecosystem, receiving nutrients and growing.
Therefore, if the AI boom subsides, hyperscalers can survive to some extent based on their existing businesses.
On the other hand, the semiconductor industry, where AI investment expansion is at the core of growth, is likely to be directly hit.
Ultimately, the market at this point seems to be reflecting this difference by valuing the survival potential of hyperscalers relatively higher and applying a higher risk premium to the semiconductor industry.
4. Interim Summary
In the AI ecosystem, semiconductors are positioned to capture value at the very last stage.
As a result, they have been able to record the highest growth rates during the period of increased AI investment.
Semiconductor companies realized revenue and profits immediately upon selling their chips.
Hyperscalers, on the other hand, face the challenge of generating actual revenue from AI services after making massive investments.
In other words, semiconductors made money right away, while hyperscalers had to prove the results of their investments in the future.
Therefore, the market has so far assigned a higher premium to semiconductor companies.
However, as doubts about the sustainability of the AI economy grow, the market's perspective is also changing.
During the AI boom, the industry that benefited the most was semiconductors.
Ironically, if the AI boom were to subside, the semiconductor industry would likely be among the first to be affected.
Just as bats can continue to find food outside their caves, but if the bats stop providing nutrients, the cave ecosystem that depends on them will be hit first.
From this perspective, we can understand why the market has recently begun to evaluate semiconductors and hyperscalers differently.
The AI Ecosystem and Market Perspective (6~10)
6. There are two dimensions to a collapse
There is a saying in the stock market:
"A bubble will burst someday, but as long as I'm not the one who gets hurt."
For investors, it is more important when a collapse occurs than whether it occurs.
This is because if the collapse is far in the future, they can still benefit from the bubble until then.
Another important factor is the scale of the collapse.
The larger the related industry, the larger the investment scale, and the more complex the interindustry connections, the greater the impact of the collapse.
What we fear is not simply the failure of a single company, but a chain reaction that shakes the entire industry.
So, what will determine the collapse of the AI boom?
I think it can be summarized into two points.
First, when the collapse will occur.
This ultimately depends on capital and profits.
Second, how large the collapse will be.
This is determined by the ratio of investment scale to actual profit, and the size of leverage.
7. When will the collapse occur?
How long the AI ecosystem can sustain itself ultimately depends on how much capital continues to flow in from the outside.
There are two types of capital involved here.
The first is investment capital.
Investment capital accelerates the growth rate of the ecosystem and postpones the point of collapse.
However, the more it accumulates, the greater the impact when the collapse occurs.
The second is revenue.
Revenue is cash earned from actual business operations, not debt that needs to be repaid.
Therefore, it is sustainable and does not increase the scale of the collapse.
The most ideal structure is one where initial investment capital is gradually replaced by revenue.
In other words,
initially, investment capital grows the ecosystem,
and over time, the AI business itself generates revenue that covers the investments.
If revenue grows quickly enough to cover existing investment scales, the market will recognize its sustainability.
Conversely, if investment continues to increase while revenue lags behind, concerns about a collapse will inevitably grow.
Here, revenue can also be considered in two ways.
A: Total revenue flowing into the entire ecosystem
B: Legacy revenue generated from existing businesses
C: Revenue generated by new businesses created through AI investment
In other words,
A = B + C
Ultimately, C needs to grow sufficiently for the justification of investments.
And finally,
revenue from AI businesses (C) must be able to cover AI investment (CAPEX).
This is why the market is worried about a collapse.
Currently, the rate of increase in investment is much faster than the rate of revenue growth for publicly available AI services.
Google's AI revenue growth was positive, but the increase in CAPEX was much larger.
With Nvidia announcing further large investments, the market began to recalculate risk.
8. How Big Will the Collapse Be?
The magnitude of the collapse depends on several factors.
First,
the larger the amount of investment capital flowing into the ecosystem, the greater the impact.
This is because that capital ultimately needs to be recovered.
Second,
the larger the future investment assets accumulated by companies, the greater the impact.
A prime example is data centers.
If AI revenue does not materialize sufficiently, massive data centers built at great expense will lose their expected economic value.
Third,
the more complex the investment structure and governance structure, the faster the impact will spread.
The harder it is to determine who is connected to whom and to what extent, the more difficult it becomes to respond.
Fourth,
the greater the difference between success and failure scenarios, the greater the volatility.
This is because while success can generate enormous profits, failure can result in most of the investment capital being lost.
Currently, hyperscalers have converted most of the cash they earned from existing businesses into CAPEX.
Moreover, they are leveraging credit with Nvidia to attract external funding into the ecosystem.
The problem is that the vast amount of capital flowing in is largely being invested in data centers and AI infrastructure, but returns commensurate with this investment scale have yet to materialize.
I believe this is the core concern of the market right now.
9. Is the "Circular Finance" Frame Appropriate?
I think the "circular finance" frame currently being used in the market is somewhat inaccurate.
In reality, when looking at the investment structure of big tech,
it's closer to large companies providing funding to AI startups that have technology but lack assets or credit.
Because banks found it difficult to bear the risk of technology-focused startups, other financial structures such as private equity were utilized.
On the surface, it may appear as if money is circulating back and forth, which can intuitively feel unsettling.
Many people also recall the subprime mortgage crisis.
However, I believe these two cases are fundamentally different.
Difference from Circular Trading
Circular trading involves inflating sales figures on paper without actual production increases.
There is a discrepancy between real and book value.
In contrast, the current AI investment structure involves the actual production of GPUs, the construction of real data centers, and the increase in actual computing resources.
In other words, there are tangible assets.
The problem is that part of the purchase funds were provided by the seller.
However, from an investment perspective, this can also be interpreted as an investment in growing customers.
If customers do not grow, it will not be sustainable,
but if customers start generating sufficient revenue through AI services, they could eventually transition to normal purchases.
Difference from Subprime Mortgages
The 2008 financial crisis was based on the assumption that house prices would continue to rise.
Loans were given to people who lacked repayment capacity, relying solely on the belief in rising house prices,
and when prices fell, the entire structure collapsed.
In contrast, the current AI investments are targeted at companies.
Companies fundamentally exist to generate profits,
and they have the potential to create cash flow through actual business operations after investment.
Of course, there is a risk of failure.
However, it is important to distinguish this structure from that of the time when reliance was solely on external variables such as rising house prices.
10. So What Should We Look For?
Ultimately, what matters is not the circular structure itself.
The key is scale and time.
The market should ask the following questions:
How much capital has been invested so far?
How sustainably can hyperscalers' existing businesses generate cash flow in the future?
How quickly is the profitability of AI businesses improving?
How much more external funding can be attracted in the future?
I believe these four factors are the key variables that determine the sustainability of the AI ecosystem.
투자 규모가 클수록 앞으로 필요한 시간도 길어질 수 있습니다.
반대로 하이퍼스케일러가 기존 사업에서 꾸준히 현금을 창출한다면 그만큼 AI 생태계는 더 오랫동안 유지될 수 있습니다.
외부 투자 역시 같은 역할을 합니다.
추가 자금은 생태계의 수명을 연장합니다.
물론 실패한다면 그만큼 충격도 커질 것입니다.
그러나 성공한다면 그 충격은 현실이 되지 않습니다.
결국 가장 중요한 질문은 하나입니다.
AI 사업이 충분한 수익을 창출하여 지금까지의 투자를 정당화할 수 있는가.
저는 이 질문이 현재 AI 산업을 바라볼 때 가장 중요한 기준이라고 생각합니다.
순환금융 자체가 핵심이 아니라,
AI가 흑자를 낼 수 있는 시점까지 충분한 시간과 자금을 확보할 수 있는지가 본질이라는 것입니다.
AI 생태계와 시장의 시선 (11~15)
11. 앞으로 주목해야 할 변수들
AI 생태계가 흑자 구조에 도달하기 위해서는 앞으로도 일정 기간 외부 자금이 계속 공급되어야 합니다.
그 과정에서 몇 가지 중요한 변수들이 있습니다.
대표적인 예가 오픈AI나 엔트로픽 같은 대형 LLM 기업들의 상장입니다.
이들 기업이 상장을 통해 대규모 현금을 확보한다면, 그 자금은 AI 컴퓨팅 인프라 사용으로 이어질 가능성이 높습니다.
컴퓨팅 수요가 증가하면 자연스럽게 데이터센터 투자와 반도체 구매도 늘어나고, 이는 AI 생태계 전체의 현금흐름을 연장하는 역할을 합니다.
물론 연기금이나 대형 금융기관, 공공기관 등 장기 자본의 참여 역시 같은 효과를 낼 수 있습니다.
이러한 자금은 단기적으로는 AI 생태계의 생존 기간을 연장합니다.
반대로 실패한다면 향후 붕괴의 규모를 키울 수도 있습니다.
하지만 AI가 충분한 수익성을 확보하기 전에 생태계가 무너지는 것과, 충분한 시간을 확보해 흑자 구조에 도달하는 것은 전혀 다른 결과를 만들어냅니다.
따라서 중요한 것은 투자 자체가 아니라, 투자가 AI의 자생력을 확보할 때까지 시간을 벌어줄 수 있는가입니다.
그리고 여기서 제가 중요하게 보는 변수가 하나 더 있습니다.
바로 SK의 하이퍼스케일러화입니다.
12. SK의 하이퍼스케일러화가 갖는 의미
이전 글에서도 언급했듯이, 저는 엔비디아와 SK의 협력 및 투자 움직임을 단순한 거래가 아니라 SK가 하이퍼스케일러 역할을 수행하려는 시도로 보고 있습니다.
이는 AI 생태계 전체에서 상당히 의미 있는 변화가 될 수 있습니다.
새로운 하이퍼스케일러가 등장한다는 것은, AI 데이터센터를 구축하고 컴퓨팅 자원을 외부에 제공하는 새로운 사업자가 생긴다는 뜻입니다.
왜 이것이 중요할까요?
앞서 사용한 비유를 다시 적용하면, 동굴 안으로 영양분을 가져오는 박쥐가 하나 더 늘어나는 것과 같습니다.
기존에는 소수의 글로벌 하이퍼스케일러가 AI 생태계에 외부 자금을 공급하는 역할을 대부분 담당했습니다.
그런데 새로운 플레이어가 같은 역할을 수행하게 되면, AI 생태계로 유입되는 자금의 통로가 하나 더 생기는 셈입니다.
그동안 반도체 기업들은 하이퍼스케일러가 벌어온 현금을 공급받는 위치에 있었습니다.
하지만 SK처럼 AI 서비스를 직접 제공하거나 데이터센터 사업을 확대하는 기업이 등장한다면 이야기가 달라집니다.
반도체 기업으로 흘러간 자금이 다시 AI 서비스를 통해 외부 시장으로 연결될 가능성이 생기기 때문입니다.
즉, 기존에는 영양분을 소비하기만 하던 인프라 계층이 이제는 직접 외부에서 수익을 창출하는 역할에도 참여하기 시작하는 것입니다.
저는 이것이 AI 생태계의 자생력을 높이는 중요한 변화가 될 수 있다고 생각합니다.
13. 결국 문제는 두 가지다
AI 산업의 미래는 결국 두 가지 질문으로 압축됩니다.
첫째, 기술은 얼마나 빠르게 발전하는가
AI의 성능이 높아지고 비용이 낮아질수록 더 많은 기업이 AI를 도입하게 됩니다.
AI는 결국 인간 노동과 경쟁하는 서비스입니다.
따라서 가격이 충분히 낮아지고 성능이 높아질수록 시장은 자연스럽게 확대됩니다.
그래서 저는 AI의 효율화 기술이 무엇보다 중요하다고 생각합니다.
모델 경량화, 추론 비용 절감, 토큰당 원가 하락 같은 기술은 단순한 기술 발전이 아니라 AI 사업의 수익성을 결정하는 요소입니다.
AI 서비스는 무한정 가격을 올릴 수 없습니다.
궁극적으로는 사람의 노동과 경쟁해야 하기 때문입니다.
따라서 비용을 낮추는 기술이 AI 산업의 지속 가능성을 결정하는 핵심이라고 생각합니다.
Secondly, can it sustain funding until then
Ultimately, capital is needed before technology is sufficiently developed.
No matter how much AI may be able to make in the future, if funding runs out before then, the market will not survive.
The dot-com bubble also collapsed because funding dried up before profits could be generated, rather than because the technology itself was flawed.
Therefore, the following factors will become increasingly important in the future:
Successful listing of LLM companies
Stable cash generation from existing hyperscaler businesses
Emergence of new investors like SK
Continued participation of long-term capital
Ultimately, technology and capital move on the same time axis.
I believe that technology is a fundamental solution, and capital plays a role in buying time until the technology matures.
14. Market Asymmetry
I think there is one asymmetry in the current market.
The market is very sensitive to news related to capital.
News such as CAPEX increases, borrowing expansions, and changes in investment structures immediately reflect risks.
On the other hand, it shows a relatively insensitive attitude towards the development of AI technology itself.
However, it is technology that ultimately determines the future of the AI industry.
As technology advances, AI will become cheaper and secure more demand, and the breakeven point will be moved forward.
Capital plays a role in buying time during this process.
Therefore, I think that the market tends to overestimate capital risk and underestimate the potential of technological advancement.
Of course, creditors and banks have different perspectives.
Creditors have limited returns when successful.
On the other hand, losses can be very large if there is a default.
Therefore, they are structurally more sensitive to risks.
Since these funds have significant influence in the market, stock prices often reflect risks first.
However, this does not mean that all future success possibilities are denied.
I think it is necessary to distinguish this point.
15. Conclusion
To summarize the above:
First,
The market is relatively valuing companies that can maintain stable cash flow even after a potential AI collapse, while considering the possibility of an AI collapse.
Second,
The market sees the capital circulation and investment structure within the AI ecosystem as important risk factors.
Third,
However, I believe that the key variable is not the cyclical structure itself but scale, time, and technology.
How much capital has been invested?
How long can it sustain based on that capital?
And whether AI can secure sufficient profitability within that time frame are essential.
Fourth,
I think the current market is reflecting risks related to capital quickly, while relatively undervaluing the long-term value that technological advancement will bring.
If this asymmetry exists, there is a possibility that the market may form prices lower than the actual value.
Of course, this does not necessarily mean that now is definitely the bottom.
I am not someone who can analyze charts or predict short-term prices.
However, I have analyzed what the market reflects more and what it reflects less.
If technological advancements begin to actually prove the profitability of AI in the future, there is a possibility that factors that have not been sufficiently valued until now will be newly reflected in stock prices.
This article does not intend to recommend buying a specific stock or predict its price.
There are always various possibilities in the market.
What I want to convey is that instead of being swept away by vague fear, it is important to calmly understand the structure of the market and the flow of funds, and objectively recheck the situation from your own perspective.
I believe that the most important thing in investing is ultimately the ability to make good judgments.
