Author: Dan Nystedt

Compiled by: Shen Chao TechFlow

Shen Chao Introduction: Micron has dropped over 24% from its peak, SanDisk is down nearly 21%, and the market blames a research paper from Google.

But the author has an insider source: the real reason is that buyers of smartphone memory chips have stopped accepting higher bids, which is the first sell signal that seasoned memory cycle investors have been waiting for.

The article outlines the current position from an industry cycle perspective and whether the narrative of an AI supercycle still holds.

The full text is as follows:

Bulls in the memory chip stocks blame the recent drop on a Google research paper, but the reason for the price decline lies elsewhere.

The true reason is much simpler: prices for certain smartphone memory chips have stopped rising. Buyers have finally said "no" — this is the first signal experienced memory cycle investors look for before selling.

Some smartphone manufacturers are now planning to reduce (or cease production of) mid-to-low-end models by 2026. The high prices of DRAM and NAND have made these phones too expensive.

This news was relayed to me through a friend.

I was quite frustrated.

"My people called me two weeks ago and said that some buyers have recently refused to accept higher DDR4 bids," he said.

I: "Two weeks ago?! "

"Well, that's why I invited you for coffee."

I: "I was the one who invited you!"

Since then, memory chip stocks have started to pull back. Micron has fallen over 24% from a recent high of $471.34. SanDisk has dropped nearly 21% from $777.60.

Investors following the memory cycle playbook quickly exited. This industry (DRAM and NAND) is known for its cyclicality.

Over the last 50 years, memory chips have experienced about a dozen major boom/bust cycles, with at least three since 2010:

• 2012-2015: Smartphones (transition from 3G to 4G + explosion of social media). Cloud data centers provided support. Smartphones surpassed PCs to become the largest memory consumers.
• 2016-2019: Cloud/ultra-large data center expansion + smartphones (storage upgrades, 5G approaching).
• 2020-2023: COVID-driven surge in demand for home office PCs/servers. Cloud service providers became the largest memory buyers.
• 2024-202?: AI-driven upcycle - focusing on training (HBM) and inference (SRAM) server memory.

These cycles are so common that they are referred to as the "pig cycle," a name derived from the livestock industry.

High prices encourage farmers to raise more pigs, but breeding takes time - so new pork doesn’t hit the market until a year later. When it does hit, it floods the market—all at once—and prices plummet.

In the DRAM/NAND sector, the "breeding time" is the long wait to build new wafer fabs.

When prices are high, companies rush to expand production, but when everyone increases output at the same time, the new supply once again crushes prices.

When prices rebound, the cycle repeats — memory chips are like pigs.

Memory cycle investors follow their playbook. No matter how many people write, "this time it’s different" — that in itself is one of the classic signals that bullish sentiment has spiraled out of control.

"The market is never wrong — opinions are often wrong." — Jesse Livermore (Reminiscences of a Stock Operator, 1923)

Is this the top?

In the long run, there are reasons to believe that this memory chip cycle will last longer than ever before. The surge in AI demand is real, although there are doubts on Wall Street; the infrastructure continues to be built.

The start is always a chip price surge driving stock prices up. Chip buyers seem to have finally stopped panic buying.

The slight sell-off triggered by the Google TurboQuant paper, more than being related to research findings, is about what it signifies: everyone is looking for solutions to the high memory chip prices.

He is right. High DRAM and NAND prices make new ideas worth pursuing.

Memory investors may find their next good idea in companies that focus more on R&D. On-chip memory, for example, is currently very popular.

NVIDIA acquired Groq's inference technology and core team members for $20 billion. This resulted in the NVIDIA Groq 3 LPU chip (manufactured using Samsung's 4nm process) packaging about 500MB of on-chip SRAM in a traditional-sized chip—ranking among the highest—with a bandwidth of up to 150 TB/s, for faster inference (Grok, Gemini or faster responses from ChatGPT).

Research is also underway to utilize stacked wafers and other methods to achieve more on-chip SRAM in next-generation products.

This is not all.

New memory technologies such as ReRAM or MRAM may one day help by providing more on-chip storage capacity while maintaining speeds like SRAM. Because they do not lose data after power loss, and operate more efficiently, they can provide additional fast memory for chips without occupying as much valuable space or consuming as much power as today’s pure SRAM.

These are just a few examples of memory that may perform better in the AI era. AI companies are willing to pay a premium for performance and energy efficiency.

The broad moat that DRAM and NAND enjoyed for a long time — the lowest price per bit — is becoming less significant as prices surge and shortages loom.

So, is the memory boom over?

Not likely.

The memory chip boom still appears to be in its early stages and may run longer than normal cycles. The label "supercycle" is appropriate because of the massive scale of AI data center construction.

Memory stocks could very well see a second wave of increases.

But experienced pig cycle investors will always follow their playbook.

Memory stocks will have to prove themselves. Watch how they perform. If the rebound encounters sustained selling, wait for them to re-establish momentum.

What do you think — are we still early, or is it time to be cautious?