● AI Bubble Fears
AI Investment Outlook Cut Shock: The Real Meaning Is Not “An AI Bubble Burst,” but a “Repricing of the Cost of Capital”
The most important point in this issue is not simply that “SemiAnalysis lowered its AI outlook.”
The real core point is that AI investment, semiconductor stocks, data center infrastructure, interest rate outlooks, and the global economic outlook became intertwined all at once, and the market has started to pay more sensitive attention to “capital accessibility” than to “growth potential.”
In particular, this adjustment is less a sign that AI demand has disappeared and more a valuation reset caused by overly aggressive forecasts coming down to a more realistic level.
At first glance, the headline makes it seem like “the last AI bull has also turned bearish,” but if you dig into the numbers, SemiAnalysis still maintains one of the highest AI infrastructure investment outlooks among major Wall Street institutions.
In other words, this event should be seen not as the end of the AI industry, but as a signal that the AI investment cycle is moving from an “expand at all costs” phase to a phase that scrutinizes profitability, interest rates, power, and supply chains.
1. What SemiAnalysis lowered was not the direction of AI investment, but its “growth pace”
SemiAnalysis sharply lowered its outlook for global AI accelerator investment.
The previous 2028 outlook for AI accelerator investment was $1.05 trillion.
After the revision, the outlook fell to $760 billion.
The reduction is about $300 billion.
In Korean won, that is close to 500 trillion won.
Just looking at this number, it is understandable that the market was shocked.
SemiAnalysis has in particular been one of the most optimistic research firms on AI investment.
In March, it strongly argued that there was an AI semiconductor supply shortage.
In June, it said Nvidia’s data center revenue could be 20% higher than Wall Street expected.
Then, after lowering its forecast this time, the market interpreted it as “the last bastion of AI bulls has fallen.”
But the important thing is that even though the forecast was lowered, the market size itself continues to grow.
-
2026 AI accelerator investment outlook: $392 billion
-
2027 AI accelerator investment outlook: about $630 billion
-
2028 AI accelerator investment outlook: $760 billion
-
2029 AI accelerator investment outlook: $800 billion
-
2030 AI accelerator investment outlook: $920 billion
From 2026 to 2030, this still represents roughly 2.4x growth.
So saying “AI investment is over” is an exaggeration.
More accurately, “the growth rate of AI investment has slowed” is the right description.
2. The reason the market was surprised lies more in “who lowered it” than in “how much it was lowered”
The reason this downward revision was taken more seriously is because of SemiAnalysis’s position.
This firm had previously issued AI investment forecasts that were much more aggressive than Wall Street’s.
That becomes especially clear when comparing global AI infrastructure investment outlooks for 2028.
-
SemiAnalysis revised outlook: $1.9 trillion
-
Morgan Stanley outlook: $1.76 trillion
-
UBS outlook: $1.62 trillion
-
Goldman Sachs outlook: $1.4 trillion
-
Market consensus: about $1.3 trillion
Here is the key point.
Even after lowering its forecast, SemiAnalysis is still projecting the highest level among major institutions.
The issue was that the previous outlook, at around $2.6 trillion, was simply too high; the revised outlook has not turned bearish.
In other words, this adjustment is closer to “a bullish forecaster coming back down to Wall Street’s upper range” than to “a bull turning bearish.”
The market did not distinguish that difference and sold semiconductor stocks first.
3. AI accelerators and AI infrastructure are not the same thing
To understand this issue, you need to distinguish between AI accelerators and AI infrastructure.
AI accelerators are chips that handle AI computation.
The representative example is Nvidia GPUs.
Google TPU, Amazon Trainium, and AMD AI chips are also included here.
AI infrastructure, on the other hand, is a much broader concept.
It includes not only AI chips, but also data centers, servers, networking equipment, power facilities, cooling systems, storage, memory, and operating costs.
That is why the AI accelerator forecast and the AI infrastructure forecast come out at different scales.
This time, the AI accelerator investment outlook for 2028 was lowered from $1.05 trillion to $760 billion.
By contrast, the revised 2028 AI infrastructure investment outlook is still around $1.9 trillion.
If you treat the two as the same number, the market interpretation can get distorted.
4. The first reason SemiAnalysis lowered its forecast: slower revenue growth at AI companies
The first basis for SemiAnalysis’s lowered forecast is the revenue growth pace of AI companies.
In particular, it viewed third-quarter revenue growth at frontier AI labs such as OpenAI and Anthropic as slowing compared with before.
Why does this matter?
The money used to buy AI chips ultimately comes from revenue generated by AI services.
AI companies cannot keep buying far more chips than the money they earn can support.
Accordingly, SemiAnalysis appears to have recalculated its model based on “chip purchases that AI companies can realistically afford.”
Based on the original figures, OpenAI’s revenue growth is still relatively strong.
Meanwhile, Anthropic was noted as having slowed somewhat in some periods.
Anthropic’s annualized revenue is interpreted as having increased from about $6.5 billion at the end of July to about $7.1 billion at the end of September.
That means an increase of about $600 million over two months.
Of course, that is still strong growth.
Still, the slowdown versus the previous quarter increased market concern.
However, with annualized revenue later being cited at about $8.1 billion as of the end of October, there was also evidence of re-acceleration.
In other words, the slowing revenue growth at AI companies is not yet a confirmed trend; it is still a highly volatile phase.
5. The more important second reason: data centers are built with debt
Personally, the most important reason behind this outlook cut is the second one.
That is the funding environment.
AI data centers are capital-intensive infrastructure businesses.
Most are financed through debt, bonds, project finance, and equity investment.
But when interest rates rise, the cost of building the same data center becomes much higher.
In today’s market environment, where rates are hovering above 5%, investors are being more stringent.
If hyperscalers like Google, Microsoft, and Amazon do not promise long-term usage contracts, bond investors and equity investors are not willing to put in money easily.
In other words, even if AI demand exists, infrastructure investment can slow down if money becomes expensive.
This point is directly connected to the global economic outlook.
If inflation rises again or energy prices swing because of war risks, the interest rate outlook becomes even more uncertain.
The macro environment can shake stock prices more than the growth prospects of the AI industry itself.
6. The third reason: some U.S. data center schedules are delayed
SemiAnalysis’s third reason is delayed data center capacity.
It said about 5.8 gigawatts of data center capacity in the U.S. had been pushed from the end of 2027 to early 2028.
5.8 gigawatts is roughly equivalent to about six large nuclear reactors.
That is by no means a small scale.
Still, SemiAnalysis considered this factor less important than the first two.
The core issue is not the delay itself, but the fact that power, land, permitting, transmission networks, and cooling infrastructure are becoming bottlenecks to AI data center expansion.
7. Nvidia was identified as the biggest victim of the outlook cut
In this model revision, about 66% of the reduced wafer area was said to belong to Nvidia.
The remaining 34% is divided among Google TPU, Amazon Trainium, AMD, and others.
Broadcom was relatively less affected.
Broadcom is a key company related to Google TPU design.
Amazon’s in-house Trainium chip was also said to be barely affected.
The reason this interpretation matters is clear.
It means that when budgets are cut, Nvidia GPUs may be among the first to be adjusted.
When AI infrastructure investment is reduced, not every company is hit equally.
The difference between hyperscalers with in-house chips and companies that rely heavily on general-purpose GPUs can become much larger.
Some in the market also mention a leadership vacuum around Jensen Huang or fatigue with Nvidia dependence.
But the essence is not a leadership issue so much as the fact that clients have begun to more aggressively review in-house chips and alternative supply chains to optimize costs.
8. In reality, companies are spending more on AI
Here is an important twist.
Stock prices fell, but actual corporate AI spending did not.
According to Ramp, which handles corporate payment data, the top 1% of AI spenders are spending $1,199 per employee per month on AI.
That is more than fivefold higher than roughly $190 in January of last year, in less than two years.
$1,199 per employee per month is a substantial cost.
Some companies are effectively putting amounts into AI tools and AI infrastructure that are close to employee salaries.
The top 10% of AI spenders are similar.
AI spending per employee rose from $130 to $801.
That is about a 6x increase.
The reason this data matters is that it reflects actual payments, not forecasts.
Markets worry about an AI investment slowdown, but on the ground, AI adoption costs keep rising.
9. The semiconductor supply chain is still tight
When AI semiconductor stocks plunge, people immediately think, “Isn’t supply gluts coming?”
But component supply chain data points in the opposite direction.
According to data released by TrendForce on November 16, lead times for major components are still getting longer.
Lead time is the amount of time it takes to receive a product after ordering it.
-
Hard disk lead time: expanded from 16 weeks to about 50 weeks
-
ABF substrate lead time: expanded from 12 weeks to 48 to 56 weeks
-
MLCC lead time: expanded from 12 weeks to about 40 weeks
-
DRAM lead time: expanded from 8 weeks to about 18 weeks
Six of the seven items are still described as being in shortage.
Just because stock prices fell does not mean the real bottlenecks disappeared immediately.
If anything, demand is still pressuring supply in AI servers, memory, substrates, power components, and storage.
10. Are AI stocks expensive like the dot-com bubble?
In the market, people keep saying, “AI stocks are like the dot-com bubble.”
But in terms of valuation alone, there is a big difference between then and now.
Looking at March 2000, the peak of the dot-com bubble, the S&P technology sector’s forward 12-month price-to-earnings ratio was about 55x.
Cisco was about 130x.
By contrast, Coca-Cola was about 24x, and the consumer staples sector was about 17x.
At that time, tech stocks were overwhelmingly expensive, and defensive stocks were relatively cheap.
Based on current figures in the original text, the situation is different.
The Nasdaq 100 is cited at about 28x.
Nvidia is cited at about 11x, and Micron at about 4x.
By contrast, Walmart is around 40x and Costco around 42x.
Just from these numbers, one could even argue that it is closer to a “retail defensive stock bubble” than an “AI bubble.”
Of course, since each company has different accounting standards and earnings cycles, simple comparisons should be made carefully.
Still, it is at least an overreach to view the entire semiconductor stock group today as equivalent to the dot-com bubble peak.
11. Hyperscaler valuations are also in a historically low range
The relative valuations of hyperscalers like Google, Microsoft, and Amazon are also important.
According to the original text, the ratio of hyperscaler P/E to the overall market P/E rose to about 1.9x in 2018.
The 14-year median is about 1.45x.
Right now, it is around 1.2x.
More recently, it is mentioned that it fell to as low as 1.07x, the lowest level since 2013.
This means that even though hyperscalers are making much more money than before, their premium versus the market is at a more than 10-year low.
You could say that AI investment burdens have actually pushed the valuations of mega-cap tech stocks down.
12. The essence of this decline is supply and sentiment, more than earnings
It is difficult to explain the recent decline in tech stocks solely by weakening earnings.
Supply and investor sentiment are likely having a larger impact.
Hedge funds significantly increased their momentum stock exposure in the first half of the year.
They held large amounts of recently surging AI-related stocks, and some even used leverage.
That exposure rose to as much as about 60% of fund assets.
On a five-year chart, that was a very high level.
But now that exposure has fallen to about 7%.
That means it has dropped by 53 percentage points in just a few months.
Given that the 2024 low was about 5%, much of the hedge fund selling pressure may already be priced in.
CTA funds are similar.
CTA refers to trend-following funds.
They buy when prices rise and sell when prices fall, using rule-based strategies.
CTA positioning in the Nasdaq 100 is currently cited at about negative $3.87 billion.
They have already sold a substantial portion of their holdings and have shifted to the short side.
The 10-year average is about positive $4.9 billion.
The 10-year low is about negative $6 billion.
The current positioning is cited as similar to the lows in late 2018, 2022, and spring 2025.
Historically, those periods overlapped with times when the Nasdaq formed a bottom.
Of course, that does not mean the same thing will definitely happen again this time.
Still, it is worth checking whether short-term selling pressure has been substantially exhausted.
13. The real core point that gets less attention in other news and on YouTube
The most important point in this event is not the “AI outlook cut” itself.
The real core point is that the AI investment cycle is now determined by three conditions.
-
First, how much money AI companies can actually make.
-
Second, how cheaply they can raise the money to build data centers.
-
Third, how quickly the power, component, substrate, memory, and supply chains can loosen.
Most market commentary focuses on Nvidia’s stock price and GPU demand.
But the more important bottleneck going forward may not be a single GPU.
Power grids, transformers, cooling systems, ABF substrates, memory, data center land, and long-term power contracts are all likely to determine AI growth rates.
Another important point is the rise of in-house AI chips.
As the ecosystems around Google TPU, Amazon Trainium, AMD, and Broadcom grow, Nvidia’s monopoly premium may face some pressure.
AI demand is not dying; what is changing is which companies capture that demand.
Therefore, from an investment perspective, “AI or not AI” matters less than “who maintains pricing power within AI spending.”
14. What to watch in the market going forward
The first checkpoint is interest rates.
If rates stay high, the profitability of data center projects falls.
The pace of AI investment expansion may also slow.
The second checkpoint is AI company revenue.
You need to watch how quickly AI service revenue is actually growing at OpenAI, Anthropic, Google, and Microsoft.
If rising AI usage does not translate into revenue and profits, the market may become skeptical again.
The third checkpoint is component lead times.
You need to confirm whether supply of ABF substrates, HBM, DRAM, power components, and storage is loosening.
If shortages continue, that is a signal that AI infrastructure investment is still resilient.
The fourth checkpoint is hyperscalers’ long-term contracts.
It matters how aggressively Google, Microsoft, Amazon, and Meta are signing data center usage and power contracts.
Project finance is only possible if those contracts exist.
The fifth checkpoint is supply and demand positioning.
You need to watch whether hedge funds and CTAs add more short exposure or shift to short covering.
Supply and demand positioning can lead the rebound before earnings do.
15. Conclusion: AI investment is slowing, but AI is not collapsing
This SemiAnalysis forecast cut is clearly a burden for the market.
In particular, semiconductor stocks and AI infrastructure-related names can become more volatile in the short term.
But looking at the data, it is hard to say that AI demand has disappeared.
Corporate AI spending is actually increasing.
Lead times for major components are still getting longer.
Although the AI accelerator investment outlook was lowered, the structure still shows growth through 2030.
Even after lowering its forecast, SemiAnalysis still maintains one of the highest levels among major Wall Street institutions.
The current correction is more reasonably seen not as an AI bubble burst, but as a process in which overly optimistic expectations are being brought into line with the realities of interest rates and capital funding.
Still, it would be premature to conclude that a V-shaped rebound is coming right away.
Interest rates, war risks, inflation, and data center delays can continue to move stock prices.
For investors, rather than trying to make one big all-or-nothing bet, a more realistic strategy is to scale into AI infrastructure companies and semiconductor supply chain firms whose earnings remain intact.
The AI industry is not over; it is now entering the phase where real costs and profitability are being tested.
< Summary >
SemiAnalysis lowered its 2028 AI accelerator investment outlook from $1.05 trillion to $760 billion.
Even after the revision, however, AI investment is still set to keep growing through 2030.
SemiAnalysis still maintains one of the highest AI infrastructure investment outlooks among major Wall Street institutions.
The core reasons for the forecast cut are not an AI demand collapse, but slower AI company revenue growth, high interest rates, and financing burdens for data centers.
Actual corporate AI spending is rising rapidly, and lead times for major semiconductor components are still lengthening.
The recent pullback in tech stocks appears to be driven more by hedge fund and CTA selling, weakening investor sentiment, and interest rate pressure than by deteriorating earnings.
Rather than an AI bubble burst, it is more appropriate to view this as a phase in which the AI investment cycle is being recalibrated to a more realistic pace.
[Related Articles…]
- AI Infrastructure Investment Cycle and Big Tech Profitability Outlook
- Semiconductor Supply Chain Shifts and Data Center Demand Analysis
*Source: [ 월텍남 – 월스트리트 테크남 ]
– 마지막 ‘강세론자’ 마저 변심했다 ㄷㄷ..


