After experiencing a tenfold surge, signs of a peak in the memory market are finally becoming apparent.

On July 7, Samsung Electronics reported its financial results, revealing a staggering 1810% year-over-year increase in operating profit, far surpassing market expectations. Its quarterly profit even eclipsed that of NVIDIA, making it the most profitable company worldwide. However, this outstanding performance triggered a classic 'buy the rumor, sell the news' sell-off, with Samsung's stock plummeting over 9% on the same day, leading to a trading halt in the South Korean index.

In fact, the sector had already witnessed nearly half a month of intense volatility prior to Samsung's sharp decline. From Samsung and SK Hynix in the Asia-Pacific market to Western Digital and Micron on Wall Street, extreme volatility characterized by 'up 10% today, down 10% tomorrow' became increasingly common.

The catalyst for the memory market's volatility was the weakening belief in 'infinite demand and insufficient supply.' For instance, Meta's decision to lease computing power triggered a sharp drop in the memory market, with the core trading logic centered around market concerns over an oversupply of AI computing power and a slowdown in memory procurement.

The true barometer of whether demand is faltering lies in the capital expenditures of major manufacturers.

The memory market's frenzy is predicated on the annual AI capital expenditures of downstream giants, amounting to hundreds of billions of dollars. Marginal changes in these expenditures exert the strongest gravitational pull on AI-related trading trends in the secondary market.

However, embarrassingly, the free cash flow—a leading indicator reflecting capital expenditure commitments—in the hands of major manufacturers has begun to raise red flags.

According to Wall Street projections, the combined free cash flow of Amazon, Alphabet, Microsoft, and Meta is expected to plummet to around $4 billion in the third quarter, compared to an average of $45 billion per quarter since 2022.

The depletion of free cash flow casts doubt on whether major manufacturers can sustain high-intensity capital expenditures, thereby determining the valuation watershed for memory chips.

After all, in this AI-driven rally, the market has had little time to digest valuations gradually. Funds are compelled to price assets within an extremely narrow timeframe, with stock price gains measured in multiples.

The extreme speed of this process has made the marginal demand increment the linchpin of market pricing for memory chips: when memory procurement growth halves, even if it remains positive, the market immediately downgrades long-term price and gross margin expectations, leading to corresponding stock price corrections.

/01/ The belief in 'infinite demand and insufficient supply' is crumbling

Who could have predicted that the South Korean stock market would also succumb to a 'speculative frenzy'?

Since late June, the South Korean index has experienced violent swings almost daily, triggering circuit breakers in both directions.

The core reason for the index's volatility lies in the memory duo, Samsung Electronics and SK Hynix, which account for over half of the index's weighting. Since late June, market divergence over these two companies has intensified, with their stock prices frequently experiencing 'up 10% today, down 10% tomorrow' roller coaster rides.

This extreme volatility is not confined to the Asia-Pacific market but has swiftly spread to Wall Street. Memory giants like Micron and Western Digital on the other side of the ocean have also witnessed frequent violent swings of around 10%.

The memory sector's transition from unilateral gains to severe volatility is closely tied to recent demand concern events.

First, concerns over demand in the consumer electronics sector surfaced. On June 25, Apple announced global price hikes for products like MacBooks and iPads, attributing the move to soaring memory chip prices, which the company could no longer absorb on its own.

Even Apple couldn't withstand cost pressures, leading to negative market interpretations: AI-driven memory price hikes may have already breached the affordability threshold for consumer electronics terminals. This implies an imminent turning point for memory demand and prices, directly dampening long-term profit expectations for memory chips. After Apple's price hike announcement, Samsung and SK Hynix plunged nearly 9% in a single day.

If Apple's price hike shook traditional consumer electronics demand, Meta's move directly shook the core belief in AI memory.

The underlying logic that previously fueled the memory sector's surge was 'infinite demand and insufficient supply.' However, Meta's announcement to lease idle AI computing power and provide model services raised market doubts: if even the tech giant with the highest AI computing power purchases and spending is now leasing out 'surplus capacity,' is the supply-demand relationship across the entire AI chain really as tight as previously portrayed?

After the belief weakened, Micron plunged over 10% in a single day, while Samsung and SK Hynix fell nearly 7%. The core trading logic is clear: the market began worrying about an oversupply of AI computing power, leading tech giants to slow their memory chip procurement growth.

Amid a series of demand weakness concerns, signals to take profits at highs emerged. Samsung Electronics' results, which fully exceeded expectations, instead triggered a 'buy the rumor, sell the news' sell-off.

Of course, this round of weakening demand expectations was also influenced by complex factors such as weakening non-farm payroll data leading to fading rate hike expectations, Wall Street banks voicing support, and South Korean retail investors leveraging heavily to buy the dip. As a result, the memory sector did not experience a unilateral downtrend but often saw violent rebounds after sharp declines.

What is certain, however, is that the memory sector has completely bid farewell to its previous phase of blind frenzy. Whether the rally has hit its valuation peak has become the core focus of investor debates.

/02/ Shovel buyers are running out of cash

This rally exhibits two peculiar characteristics: first, the AI chain remains resilient regardless of how crowded it becomes. Second, sectors outside the AI chain refuse to rebound no matter how much they decline.

The core reason for this extreme divergence is institutional funds frantically trading 'certainty.' In an environment where most industries lack clear growth logic, the AI chain, represented by memory chips, offers very certain growth expectations.

Supporting these growth expectations are the capital expenditures of tech giants. As the saying goes, 'to get rich, build roads first; to pursue AI, invest in infrastructure first.' To capture AI dividends, the combined capital expenditures of the world's four tech giants (Meta, Microsoft, Alphabet, Amazon) are expected to reach around $725 billion in 2026, up 77% from around $410 billion in 2025.

Ultimately, the memory frenzy is built on the annual AI capital expenditures of downstream giants, amounting to hundreds of billions of dollars. Marginal changes in these expenditures remain the biggest gravitational force for all AI-related anxieties and pride in the secondary market.

In the current rally, even without an AI bubble burst, a mere slowing trend in the capital expenditure growth of giants could potentially reverse the memory market trend.

The logic is that AI is reshaping global productivity and production relations at an unprecedented pace. This extreme 'speed' hardly gives the market time to digest valuations gradually. Funds are compelled to take sides and price assets within an extremely narrow timeframe, with stock price gains measured in multiples.

The extreme multiple gains have made the marginal demand increment the linchpin of market pricing for memory chips: when memory procurement growth halves, even if it remains positive, the market immediately downgrades long-term price and gross margin expectations, leading to corresponding stock price corrections. As mentioned earlier, Apple's price hike and Meta's computing power leasing both triggered sharp declines, essentially trading on 'growth slowing expectations.'

Can tech giants sustain high capital expenditure growth indefinitely? Cash on hand speaks louder than any statement.

Free cash flow can be seen as a leading indicator of giants' capital expenditures, as it measures the remaining discretionary cash after paying operating costs and capital expenditures.

After a round of aggressive, epic expansion (e.g., Google's capital expenditures this year are already eight times those of 2020), the cash flow situations of giants are not optimistic.

According to Wall Street projections, the combined free cash flow of Amazon, Alphabet, Microsoft, and Meta is expected to plummet to around $4 billion in the third quarter, compared to an average of $45 billion per quarter since 2022.

A combined free cash flow of $4 billion clearly cannot support the previous annual capital expenditures of hundreds of billions of dollars. Therefore, Meta, Google, and Amazon have successively started raising funds through various channels. For instance, Amazon already raised $50 billion through debt financing in the first quarter.

But if AI returns remain elusive, how long can giants sustain debt-funded investments? This question about cash flow commitments and return on investment is becoming the watershed determining the valuation trend of the entire AI chain.

/03/ Tense valuation risks

Firm believers in the memory market argue that from a forward P/E perspective, current valuations of Samsung, SK Hynix, and Micron are not expensive, even severely undervalued.

The logic is that AI-driven incremental demand is transforming memory chips from strongly cyclical commodities to AI infrastructure essentials, significantly boosting profit stability and shifting the valuation framework from price-to-book (PB) to price-to-earnings (PE).

In believers' projections, if calculated using forward P/E (Forward PE = current stock price ÷ consensus expected EPS for the next 12 months), Samsung and SK Hynix's forward PEs are only around 6x. Even valuing the memory sector at a traditional manufacturing PE of 10x, there is still nearly 50% upside potential.

However, this logical projection has two controversies. First, can memory chips permanently transition from cyclical stocks to growth stocks?

The cyclicality of memory chips stems from the fact that 'technological iterations require new production lines, and the release of new capacity often does not synchronize with downstream demand growth.' This long-cycle supply-demand mismatch is an inherent trait of the industry.

While AI has created massive incremental scenarios for memory chips and brought extremely high profit stability during the demand explosion phase due to supply shortages, questions remain. If Samsung or Micron achieves a major breakthrough in HBM yield, undermining the HBM scarcity narrative, could the industry slide back into an old cycle of overcapacity?

Compared to whether cyclicality has changed, the potential impact of marginal demand slowing on valuations is more thorny. Even if the AI industry does not experience a bubble, as long as memory chip demand growth marginally slows, preventing further price hikes, it would deal a heavy blow to current valuation models.

For example, Micron's 2026 annualized non-GAAP net profit is estimated at around $56 billion, corresponding to a forward PE of about 18x. This valuation does not seem excessive.

However, the denominator of this PE is based on super-cycle profits where DDR4 contract prices surged 10x in 15 months, and gross margins jumped from 36% to 75%. Multiplying peak-cycle profits by a seemingly 'reasonable' multiple to derive a seemingly 'inexpensive' valuation is often the classic valuation trap when cyclical stocks peak.

In 2000, Cisco's PE was also 'only' around 60x, based on revenue growth exceeding 50% for 15 consecutive quarters. When growth slowed from 50% to 20% and then to 0%, EPS didn't need to fall much for the stock price to plunge 80%, as both the multiple and earnings contracted simultaneously.

When the entire market shouts 'this time is different,' at least pause to ask: what happened last time everyone was so certain?