● AI Infrastructure Demand Surge
Why BlackRock CIO said “AI tech stocks are cheap”: what matters now is not stock prices but AI infrastructure demand
What to focus on in this article is not simply, “Nvidia has fallen a lot, so should we buy?”
We will organize with numbers why Rick Rieder, BlackRock’s global fixed-income CIO, saw tech stocks as “cheap” even right after the Nasdaq 100 plunged, whether demand for AI semiconductors and data centers has really weakened, and whether the memory semiconductor cycle is different from the past.
In particular, there is another point that most news coverage misses.
Rather than saying AI growth has broken down, the market right now is more about too many investors crowding into the same stocks and then being forced to reduce positions.
So the essence of this correction is whether it is an “AI bubble burst” or a “global asset-allocation rebalancing.”
1. Why did BlackRock CIO Rick Rieder say “tech stocks are cheap” the day after the plunge?
Rick Rieder is the CIO overseeing global fixed-income investments at BlackRock, the world’s largest asset manager.
He appeared on Bloomberg TV the day after the Nasdaq 100 fell 6.4% in a single day and Nvidia dropped 9.3%, saying tech stocks were cheap.
The Philadelphia Semiconductor Index had also posted its worst decline since March 2020.
For ordinary investors, it is only natural to think, “Are you telling us to catch a falling knife?”
But Rieder’s logic was not simply a buy-the-dip argument.
What he focused on was not the stock decline, but the gap between valuations and earnings growth.
2. Key numbers: tech stocks trade at 17x P/E, versus a 5-year average of 25x
The key figure in the original piece is that tech stocks trade at roughly 17 times earnings.
That is well below the recent 5-year average of about 25 times.
In particular, excluding the 2020 COVID-bottom period, current tech valuations are said to be cheaper than during the 2022 major bear market or the 2018 U.S.-China trade war.
The important point here is not simply “the P/E is low.”
Usually, a low P/E means a stock is cheap because growth is weaker.
But AI infrastructure companies are instead expected to deliver strong earnings growth.
3. AI infrastructure companies’ earnings growth is projected to exceed 80% through 2027
AI infrastructure-related companies are expected to see EPS growth of more than 80% through 2027.
By contrast, non-AI-related companies are shown at around 10% growth.
This gap is the core point.
AI semiconductor companies with much higher annual growth rates are being assigned low P/Es, while consumer companies with relatively lower growth are getting high P/Es.
The original piece points out the irony that companies like Nvidia, which are growing by nearly 100% per year, are being given lower valuations, while companies like Costco or Walmart, with growth in the 10% range, are being valued at nearly 40 times earnings.
From this perspective, the market is already pricing in a scenario in which “AI is highly likely to fail.”
4. How this differs from the dot-com bubble: more like “already-corrected growth stocks” than an “rising bubble”
Many investors compare the current AI market to the dot-com bubble.
But Rick Rieder’s view is different.
During the dot-com bubble, stocks of companies without earnings support rose dramatically.
By contrast, today’s AI infrastructure companies have concrete demand in the form of actual revenue, order backlogs, supply shortages, and data-center investment.
Also, recent price action is interpreted less as an endless upward bubble and more as growth stocks that have already undergone large corrections on the way up and are reverting toward the mean.
If you fail to distinguish this difference, you may mistake every AI semiconductor correction for a bubble burst.
5. The real reason stocks fell: position crowding rather than weakening AI demand
According to Rieder, the cause of the recent plunge is not a collapse in AI demand.
The key is large-scale liquidations by hedge funds and institutions, along with concentrated positioning.
In simple terms, too many investors were holding the same stocks.
Large-cap tech stocks had become a bigger part of the U.S. market, and funds had concentrated into semiconductor and AI infrastructure names.
The problem is that when everyone already owns a lot, there are fewer incremental buyers.
When market volatility rises in that state, some funds must sell for risk-management reasons.
At that point, even good companies get sold.
So the rebalancing Rieder referred to is not about abandoning stocks altogether, but about reducing some of the oversized equity exposure and restoring balance with bonds and other assets.
6. So is the conclusion an AI bubble burst or a short-term correction?
Rieder’s conclusion is clear.
He does not see the current situation as an AI bubble burst.
His interpretation is that this is not a market collapsing because valuations are too expensive, but rather a short-term overheating of positions that are now being unwound.
Ultimately, his core message is that stock prices follow earnings in the long run.
That said, there is also something investors should watch closely.
Cheap does not mean every tech stock is good.
Within AI infrastructure, you need to distinguish companies with actual orders, companies benefiting from supply shortages, and companies directly tied to data-center investment.
7. The most important AI infrastructure segment: data-center networks and optical communications
The area especially emphasized in the original piece is networking and optical communication components used in data centers.
AI data centers do not function by simply packing in more GPUs.
You need network infrastructure to connect GPU to GPU, server to server, and to move massive amounts of data quickly inside the data center.
The growth rate of this market is said to exceed 95%.
On a standalone market basis, optical communications is interpreted as one of the areas showing the highest growth.
Yet ironically, the valuations of related companies remain low.
It is argued that companies such as Marvell, Broadcom, Lumentum, Credo, and Coherent are trading at less than 20 times earnings.
If companies that can grow annual revenue by 2x to 3x are receiving these valuations, the market can be seen as pricing in a fairly conservative scenario.
8. Lumentum CEO’s comment: “There is a big gap between stock declines and reality”
An especially striking part of the original piece is the Lumentum CEO’s comment.
He said that even though the company’s stock had fallen 40% in a week, there was a large disconnect between market reality and the share price.
The key is the order backlog.
The company has a huge backlog and said that even if it expands production capacity as much as possible, it will only be able to meet about 30% of the required supply.
If that statement is true, the current stock decline may be driven more by market risk aversion and forced selling than by disappearing demand.
9. What Big Tech CEOs are all saying: “AI demand is real”
The original piece argues that AI demand remains strong based on comments from major CEOs during the third-quarter earnings season.
OpenAI CFO Sarah Friar said the company has to keep investing in data centers to meet customer demand.
CoreWeave CEO Michael Intrator also positively noted that completed permitted data centers continue to be approved.
Amazon CEO Andy Jassy said customers are quickly consuming the capacity being installed and that there is no sign of excess supply.
Microsoft’s CEO also said the company is building more data centers and showed very high conviction in the long-term return on investment.
Meta’s Mark Zuckerberg made similar remarks.
TSMC’s chairman also said demand is very strong and that he sees no sign of a slowdown.
In particular, TSMC conducted on-site checks to confirm whether data centers are actually being built and concluded that demand is real.
The fact that TSMC, at the top of the semiconductor supply chain, made that confirmation should not be taken lightly.
10. The entire semiconductor supply chain is saying the same thing
The story that AI demand is strong is not coming only from Big Tech.
Intel’s CEO described it as the most severe supply shortage in history.
AMD and Broadcom CEOs also said massive demand is required.
SK hynix said, in effect, that DRAM and HBM supply has effectively sold out for the next several years.
Micron is sending a similar message.
Optical communications companies Coherent and Lumentum also say orders keep pouring in and that customers are already reserving capacity into 2028.
At that point, it is fair to say this is not just one company’s optimism, but the same message repeating across the supply chain.
11. The key takeaway from Citi’s report: memory may no longer be a simple cyclical industry
One of the most important parts of the original piece is Citi’s analysis of memory semiconductors.
Memory prices are said to have surged more than tenfold recently.
What is surprising is that this surge is attributed to a supply-demand mismatch of just 3.8%.
Traditionally, the memory industry has been a classic cyclical sector.
When prices rose, companies added fabs, and when supply increased, prices collapsed.
But Citi believes this cycle may be different from the past.
Forecasts suggest that supply shortages could continue at around -3.6% through 2028 and still be negative at -0.3% in 2029, meaning supply may remain short relative to demand.
More importantly, the analysis suggests shortages could intensify again in 2030 and 2031.
In other words, memory semiconductors may be entering a phase of structural supply shortages beyond a simple recovery in the cycle.
12. Why LTA long-term supply contracts matter: they act as a shield against price declines
A key keyword in the memory market recently is LTA, or long-term supply agreements.
According to the analysis, for hyperscalers, meaning major cloud companies, to find it worthwhile to break an LTA and pay a penalty, spot prices would need to fall by more than 60% from current levels.
But for spot prices to fall by more than 60%, extreme oversupply and a full downturn would be required.
Given the current supply-demand structure, the analysis suggests that such a scenario is unlikely to materialize immediately.
That is why Wall Street analysts believe the LTAs already signed are unlikely to be easily broken.
This is not widely covered in general news, but it is a very important factor when judging the earnings stability of memory companies.
13. Then why haven’t SK hynix and Samsung Electronics risen much?
The original piece points out that valuations for Korean semiconductor stocks are extremely low.
SK hynix and Samsung Electronics are described as trading at around 3 times earnings, with P/B only slightly above 1 times.
By global standards, these are very low valuations for equities.
Yet the stocks have not risen as strongly as expected.
The reason is not easy to explain by company competitiveness alone.
The bigger issue is the structural supply-demand imbalance in the Korean market.
When foreign investors sell, there is not enough domestic long-term capital to absorb it, and as a result the stocks are repeatedly pressured.
This shows that the Korean equity discount is not simply a matter of company earnings, but of market structure and capital flows.
This trend could also affect investor sentiment toward global memory names such as Micron and SanDisk.
14. The core point other news often misses: right now the market should focus on “collateral and credit risk” more than “demand debates”
Most YouTube channels and news reports simplify the issue into “it’s an AI bubble” or “it’s a buy-the-dip opportunity.”
But there is another, more important point.
The real risk of AI investment is not simply Nvidia’s share price falling.
If AI demand weakens much more than expected, semiconductor chips could move into an oversupplied state.
In that case, GPU prices would fall, and financial structures that use GPUs almost like collateral could come under pressure.
That could spread into credit risk across both bond and equity markets.
In other words, the core of the AI bubble debate is not the stock chart, but the financing structure tied to AI infrastructure investment.
In an environment where the interest-rate outlook stays high for a long time, companies that expanded data centers with excessive debt could feel the strain.
So from here on, investors need to look not only at AI demand, but also at data-center companies’ debt ratios, long-term contract structures, cash flow, and bond spreads.
That is the most important checkpoint individual investors often miss.
15. Positive scenario: Agentic AI and Physical AI are still in the early stages
The original piece leans toward the view that AI demand will not easily fade.
The reason is the perspective that the AI we use today represents only a tiny fraction of its full potential.
At present, generative AI is mainly used in areas such as search, document creation, coding, and image generation.
But the next stage is Agentic AI.
Agentic AI is AI that receives a user’s instructions and plans and executes multiple steps on its own.
For example, it could go beyond simply writing a report and extend into research, comparison analysis, schedule management, purchase requests, and workflow automation.
The stage after that is Physical AI.
Physical AI is AI that operates in the real world, such as robotics, autonomous driving, smart factories, and logistics automation.
When this area begins to scale in earnest, the required computing power and data-center infrastructure could become far larger than they are today.
From this perspective, AI semiconductor demand may be overheated in the short term, but the long-term growth potential remains large.
16. Negative scenario: if AI usage is lower than expected, all assumptions break down
Of course, there are risks.
If actual AI service usage turns out to be lower than expected, the current data-center investment plans could become excessive capital expenditure.
If that happens, GPUs, HBM, networking equipment, and optical communication components could all face oversupply.
When prices collapse, company earnings deteriorate quickly.
In particular, companies that expanded capacity using high debt could face both interest-rate pressure and slowing revenue at the same time.
In that case, the impact would not stop at the stock market correction; it could also shake the credit market.
Therefore, investors should keep checking not whether “AI always goes up,” but whether AI demand is actually being converted into real revenue and cash flow.
17. What I would do next: selective rebalancing centered on AI infrastructure rather than blind buying
If it were me, I would not buy all tech stocks indiscriminately right now; instead, I would selectively rebalance around AI infrastructure.
First, I would focus on the areas where actual supply shortages have been confirmed within AI semiconductors.
HBM, high-performance DRAM, packaging, data-center networks, and optical communication components are the key examples.
Second, I would distinguish between companies whose valuations have already become excessively low and those left only with simple expectations.
A low P/E does not automatically make something cheap; what matters is whether earnings estimates can be sustained.
Third, from a global asset-allocation perspective, I would not look only at equities.
As Rieder said, if equity exposure has become too large, some of it should be balanced with bonds or cash-like assets.
Fourth, Korean semiconductor stocks need to be viewed not only through company fundamentals but also through foreign inflows, the exchange rate, and the Korean market discount.
Fifth, I would build a long-term portfolio around companies whose AI infrastructure investments actually translate into revenue.
18. Key indicators investors should watch
1) Data-center investment plans
You need to check whether capital expenditure plans from Amazon, Microsoft, Meta, Google, OpenAI, and CoreWeave keep being revised upward.
2) HBM and DRAM supply contracts
Watch whether long-term supply agreements and pricing terms for SK hynix, Samsung Electronics, and Micron remain intact.
3) Order backlog at optical communication companies
The order backlog and customer reservation periods for Lumentum, Coherent, Credo, Broadcom, and Marvell are important.
4) AI service usage
Check whether AI is actually translating into growth in work, search, coding, advertising, and cloud revenue.
5) Credit spreads in the bond market
Whether financing costs for data-center investment companies are surging is a very important risk signal.
19. Conclusion: now is the time to focus on “AI infrastructure supply and demand,” not just the phrase “AI bubble”
On the surface, this correction looks like a collapse in AI tech stocks.
But the numbers tell a slightly different story.
Tech stock P/Es are below historical averages, and AI infrastructure companies’ earnings growth is still projected to be strong.
Comments from CEOs of data-center, memory, HBM, optical communications, and networking equipment companies are mostly pointing in the same direction.
Demand is still strong, and supply is still short.
Of course, if AI demand slows much more than expected, all assumptions could be shaken.
But based on company statements and supply-demand data so far, it is entirely reasonable to argue that the market is pricing in the AI failure scenario too quickly.
In the end, what is needed now is neither panic nor blind optimism.
You need a strategy that looks at whether AI infrastructure demand is actually being sustained, whether valuations are reasonable relative to earnings growth, and where interest-rate expectations and credit risk are heading.
< Summary >
BlackRock CIO Rick Rieder said tech stocks were cheap even right after the Nasdaq 100 and Nvidia plunged.
The main reason is that tech stock P/Es are around 17 times, below the 5-year average of 25 times, while AI infrastructure companies’ EPS growth is projected to exceed 80% through 2027.
The recent decline is interpreted more as rebalancing due to hedge-fund liquidations and crowded positioning than as an AI demand collapse.
Data-center, HBM, DRAM, optical communication, and networking equipment companies are still talking about strong demand and supply shortages.
Citi analyzes memory semiconductors as potentially entering a long-term supply-shortage phase rather than remaining a simple cyclical industry as in the past.
However, if AI usage falls short of expectations, there is a need for caution because it could lead to semiconductor oversupply, lower GPU prices, and credit risk.
If it were me, rather than buying tech stocks indiscriminately, I would selectively rebalance around AI infrastructure, memory, optical communications, and the data-center supply chain.
[Related Articles…]
AI Infrastructure Investment Cycle and Semiconductor Market Outlook
Semiconductor Supercycle and Global Stock Market Trends Analysis
*Source: [ 월텍남 – 월스트리트 테크남 ]
– 저라면 앞으로 “이것” 하겠습니다.
