The direction of the capital market is becoming increasingly difficult to predict.
SK Hynix has just posted its most profitable quarterly report ever: revenue of 7.932 trillion Korean won, a staggering 257% year-over-year increase; operating profit of 6.054 trillion Korean won, a massive 557% year-over-year increase, with an operating margin as high as 76%. This means the company earns $0.76 in profit for every dollar of revenue. Yet, on the same day, SK Hynix’s stock price plunged more than 9%.
Over the past four weeks, approximately $2 trillion in market value has vanished from global markets. The AI semiconductor sector has been hit hardest: SK Hynix dropped 47%, posting its worst monthly performance since the 2008 financial crisis; SanDisk plunged over 50%; Micron fell 33%; Nvidia and ASML were also dragged down. The Philadelphia Semiconductor Index has retreated more than 20% from its all-time high on June 22, and year-to-date in July, it has declined over 22%, marking the sixth-worst monthly drop in history.
Meanwhile, Apple has regained a $5 trillion market cap, and Microsoft has quietly rebounded... the "old guard" tech stocks have become a safe haven.
If you're not in this industry, you'd likely think the story of AI died sometime in July, and those AI skeptics would say, "See, we were right."
But the opposite is true. Understanding this seemingly contradictory market movement will determine the thickness of your wallet over the coming months.
Macroeconomic context: The Fed holds steady + "Dirty laundry" exposed
At 2:00 PM Pacific Time on July 29, the Federal Reserve announced the results of its July monetary policy meeting.
The FOMC voted 9 in favor and 3 against to keep interest rates unchanged at 3.50%-3.75%. This marks the fifth consecutive hold, but also the highest number of dissenting votes since September 2016.
The three officials who voted against—Cleveland Fed President Beth Hammack, Minneapolis Fed President Neel Kashkari, and Dallas Fed President Lorie Logan—all advocated for a 25-basis-point rate hike. Notably, Kashkari, previously viewed as a dove, surprised markets with his shift to a hawkish stance. Fed Chair Kevin Warsh stated frankly at the post-meeting press conference: “I asked for a good family debate, and I certainly got one.”
Before the meeting, the CME FedWatch tool showed that traders assigned a 70.6% probability to holding rates steady and a 29.4% probability to a 25-basis-point rate hike. Citibank explicitly called this “the most divergent moment since September 2024.”
Higher and longer rates—that’s the real signal from the Fed.
This is fatal for high-valuation, long-cycle growth stocks. In a DCF model, when the discount rate rises, the present value of distant cash flows drops sharply. Chip stocks, often trading at forward P/E ratios of 50 to 100 times, appear especially vulnerable in this interest rate environment.
Memory Chip Crash: Triple Fear
First layer: China's open-source models are challenging the "compute power faith".
On July 16, Moonshot AI released the open-source large model Kimi K3 ahead of the 2026 World Artificial Intelligence Conference. With a total parameter scale of 2.8 trillion, it is the world’s first open-source model in the 3-trillion parameter class. The model employs a Mixture of Experts (MoE) architecture, features native visual understanding capabilities, and supports a context window of up to 1 million tokens. Benchmark tests show its performance is on par with the most powerful proprietary systems from Anthropic and OpenAI, and it has ranked first globally in evaluations such as frontend programming.
The core fear on Wall Street is: Will efficient models make GPUs less valuable? For the past three years, the market has believed that “stronger models → larger parameters → more GPUs → higher Nvidia value.” K3 has cracked this logic chain—if open-source models can match top-tier proprietary models at one-tenth the cost, will data centers still need so many H100s?
Second layer: CXMT listing, the specter of the HBM price war.
On July 27, China's DRAM manufacturer CXMT listed on the STAR Market at an issue price of RMB 8.66, surging 471.59% on opening to RMB 49.5. On its first trading day, its market capitalization exceeded RMB 3.3 trillion, surpassing ICBC and Kweichow Moutai to become China's most valuable listed company. This IPO raised RMB 57.9 billion, the largest in China in 15 years.
What the market truly fears is not CXMT’s current market share, but its future. If China can scale up production of HBM (the most expensive component in GPUs) and flood the market with low-priced alternatives, how long can Samsung, SK Hynix, and Micron maintain their gross margins above 70%?
Third layer: Domestic DUV lithography machines, the "China story" of ASML is about to collapse.
On July 27, foreign media The Information reported that a Chinese state-backed enterprise has begun mass-producing immersion DUV lithography equipment. Following the news, ASML's stock plunged 7%-8% that day, with its cumulative decline exceeding 18% for July. As a result, Applied Materials fell 6.7% and Lam Research dropped 7.9%.
China's market accounts for approximately 20% of ASML's total revenue. Once domestic substitution becomes a reality, this "20%" could potentially drop to "0%". This is the true fear on Wall Street.
But are these fears well-founded?
Many industry analysts believe the threat is real, but the short-term impact has been significantly overestimated.
Regarding CXMT: Although its market capitalization on the first day of listing was impressive, the global DRAM market is still dominated by Samsung (approximately 39%), SK Hynix (approximately 29%), and Micron (approximately 22%), collectively holding about 90% of the market. CXMT holds nearly zero market share in the HBM sector and will likely require several more years to achieve meaningful price disruption in HBM.
Regarding domestic DUV: This year’s planned delivery is 5 units, and next year’s output is expected to be around 20 units, compared to ASML’s annual shipment of hundreds of units. ASML still holds a monopoly on EUV—this is an irreplaceable technology for manufacturing advanced chips below 3nm.
Regarding open-source models: cheaper models do reduce training costs, but inference costs will be the major expense in the future. Agentic AI is exploding. More importantly—after the release of Kimi K3, it rapidly attracted a massive number of users, pushing its compute cluster to its limits within just a few days. Cheaper models require more chips to run, not fewer. BlackRock also believes that the recent sell-off has been "overdone."
While chip stocks are plummeting, Apple's market capitalization has surpassed $5 trillion, rising about 15% since July; Meta has risen about 14.7%.
Why?
Apple did not participate in the data center arms race. Citigroup analysts noted that Apple has been rewarded by the market for its decision “not to participate in the data center arms race.” At a time when investors are questioning the return on AI capital expenditures, avoiding heavy spending has become a competitive advantage.
Citigroup’s Scott Chronert team bluntly stated: “The concept of the Magnificent Seven is outdated.” The stock price correlation among the Magnificent Seven has broken down—Microsoft and Meta have come under scrutiny for their massive capital expenditures, while Apple has risen due to its restraint.
Baird investment strategy analyst Ross Mayfield put it plainly: “The flow of capital away from tech stocks is partly driven by value considerations. GDP is solid, the labor market remains robust, and consumer spending is accelerating again.”
The market is shifting from “storytelling” to “focusing on reality.” Sectors with real cash flows and reasonable valuations—such as consumer goods, healthcare, and industrials—are becoming safe havens in a high-interest-rate environment.
What’s the next step in the layout?
First, don't rush to heavily buy the semiconductor sector.
The Philadelphia Semiconductor Index has fallen nearly 25% from its peak; short-term oversold conditions may trigger a technical rebound, but the expectation of a September rate hike hangs like a sword of Damocles. Prior to this, BTIG’s Chief Market Technologist Jonathan Krinsky noted that the semiconductor sell-off is not yet over.
Second, focus on the "adopters" of AI rather than the "enablers."
Citi's Chronert accurately predicted last December that AI trading would shift from "enablers" to "adopters." However, note that "adopters" fall into two categories: one consists of hyperscale cloud providers that spend heavily on data centers and are also under pressure; the other comprises software and consumer tech companies that use AI to improve efficiency without building their own data centers—the latter deserves more attention.
Third, be cautious of a September rate hike.
Three insiders at the Federal Reserve have already adopted a hawkish stance. CME FedWatch shows a 57.7% probability of an interest rate hike in September. Goldman Sachs’ Haigh previously analyzed that a September hike is at a delicate balance, and any further action may depend on the situation in the Middle East and upcoming inflation data. If a rate hike does occur in September, high-valuation tech stocks will face renewed pressure.
Fourth, prioritize cash flow and valuation.
This is no longer the era from 2010 to 2021 of “zero interest rates, buy anything.” Interest rates are now above 3.5%, and capital costs are real. Companies with aggressive CapEx and negative free cash flow will continue to face pressure under rising rate expectations. In contrast, companies with stable cash flows, reasonable valuations, and resilience even if their AI narratives prove false, are the true safe havens in this rotation.
In the semiconductor industry, the long-term demand thesis remains intact. SK Hynix’s record-breaking profits—both revenue and operating profit hitting all-time highs—point in the same direction: demand for AI infrastructure has only just begun. However, short-term market sentiment, crowded trades, and a shift in the macro interest rate environment are driving a harsh “valuation reset.”
Author: bootly
Twitter: https://twitter.com/BitpushNewsCN
BitPush Telegram community: https://t.me/BitPushCommunity
BitPush TG subscription: https://t.me/bitpush