How long can the prosperity of storage last?

Author: Takashi Tamura

The semiconductor industry, especially the memory market, is currently experiencing an extraordinary expansion that can only be described as "explosive growth." Terms like "prosperous development" simply cannot adequately describe this phenomenon. A glance at the charts will leave anyone stunned by its sheer scale.

Figure 1 shows the three-month moving average shipment volumes of various semiconductors from 1991 to May 2026, based on data from the World Semiconductor Trade Statistics (WSTS). Over the past 30 years, despite fluctuations in the economic environment, the four major categories: microdevices, memory, logic devices, and analog devices, have shown a steady upward trend overall. This represents the "normal development process" of the semiconductor industry.

Figure 1. Three-month moving average shipment value of semiconductor products by type (as of April 2026) Source: Compiled by the author based on WSTS data

However, since around 2024, logic circuits, especially the recently emerging memory, have shown an almost vertical growth momentum, completely overturning traditional perceptions. The growth of memory, in particular, is astonishing, nearly breaking the peak of the chart. The speed of this growth is so rapid that the accumulated growth history of the past 30 years seems merely a warm-up.

Based on my years of close observation of the semiconductor market, I can confidently say that the memory market has never experienced such a steep growth. This can be regarded as a historic anomaly that is shaking the fundamental structure of the semiconductor industry.

It has grown tenfold in about a decade, with an annual growth rate as high as 285%, which is shocking. Figure 2 illustrates this explosive growth, focusing on the period from 2016 onwards. In 2016, the monthly shipment volume of MOS memory was about $5.6 billion. After a brief boom during the "memory bubble" of 2017-2018, the shipment volume plummeted to only $5.8 billion by early 2023 during the economic recession. This marked the lowest point of the semiconductor industry's downturn.

Figure 2. Three-month moving average shipment value of semiconductor products (by type) (April 2016 to April 2026) Source: Compiled by the author based on WSTS data

However, the subsequent recovery, or more precisely, the surge. As of May 2026, monthly shipments of memory have soared to $63.3 billion. Compared to the levels in 2016, it has increased more than elevenfold in just ten years. Compared to the low point in early 2023, it has grown approximately 10.7 times in just a few years.

Similarly, Figure 2 shows that the logic business, including NVIDIA GPUs, has also grown from $13.3 billion to $31.6 billion, clearly demonstrating the "NVIDIA GPU effect" and "AI effect." However, even the growth of the logic business appears relatively moderate compared to the explosive growth of the memory business.

From the perspective of growth rates, this anomaly is even more pronounced. Figure 3 shows the year-on-year growth rates of the same data, with memory recently reaching a year-on-year growth rate of 285%, a historic high.

Figure 3. Year-on-year growth rate of three-month moving average shipment volume of various types of semiconductors (as of April 2026) Source: Compiled by the author based on WSTS data

For reference, even at the peak of the last "memory bubble" (around 2017), the annual growth rate was about 60%. In contrast, you can see how astonishing the current growth rate is. Observing the other three categories in Figure 3, even logic circuits only have around 40%, while microcontrollers and analog circuits are only around 14% to 19%. In other words, only the growth of memory has truly reached "another level."

DRAM and NAND are experiencing explosive growth

From the perspective of the memory market, the main driving forces behind its explosive growth come from two major memory products: DRAM and NAND flash (hereinafter referred to as NAND). Figure 4 shows the quarterly overview of the global DRAM and NAND markets based on TrendForce data.

Figure 4 Global DRAM and NAND market (TrendForce's forecast for Q2 2026) Source: Compiled by the author based on TrendForce's DataTrack data.

At the beginning of 2023, during the semiconductor industry's recession, DRAM prices fell to a low of only $9.7 billion, and NAND prices also dropped to $8.7 billion. This was a dark period where all memory manufacturers were in the red and forced to cut production.

However, it is expected that by Q2 2026, the DRAM market size will leap to $145 billion, and the NAND market size will reach $81 billion. Compared to the low point in early 2023, the DRAM market size has grown about 15 times, and the NAND market size has grown about 9 times. When combined, the quarterly total will reach $226 billion, with an annual total exceeding $900 billion, a scale that seems unbelievable at first glance.

Once considered "cheap standards" or "commodities," memory is now preparing to seize the leadership of the semiconductor industry from microcomputers (such as logic and MPU).

The reason is the abnormal surge in memory and wafer prices

Why has the memory market expanded to such a scale? The key is that this explosive growth is not merely due to increased shipments; the biggest factor is the abnormal surge in memory prices themselves.

Figure 5 shows the trend of spot prices for DRAM (DDR5 16Gb 2Gx8) and NAND (1Tb TLC wafer). At the beginning of 2025, the spot price of DRAM was only $4.70. Recently, the price surged to $46.00, an increase of about tenfold. The price of NAND wafers also jumped from $2.40 to $25.00, also an increase of about tenfold.

Figure 5: Surge in spot prices of DRAM (DDR5 16Gb 2Gx8) and NAND (1Tb TLC wafer) Source: Compiled by the author based on TrendForce's DataTrack data

In other words, the primary reason for the tenfold growth in market size is not a tenfold increase in shipments, but rather the "unit price" soaring nearly tenfold. Even if memory sales remain unchanged, revenue will grow tenfold. This is the mechanism behind the explosive growth of the memory market size.

For memory manufacturers, this is an ideal environment. After all, prices rise automatically without significant capital investment, significantly improving profit margins. As will be discussed later, the soaring stock prices of memory manufacturers are also due to the rapid growth in profits brought about by this abnormal price increase.

Reasons for the price surge

So, why have memory prices surged? This is because demand far exceeds supply, but tracing the root of this demand, we find it is due to the astonishing investments made by hyperscale data center operators.

Figure 6 shows the capital expenditure trends of the four major hyperscale data center operators (Amazon, Google, Microsoft, and Meta). In 2015, the total capital expenditure of these four companies was only $21 billion. Even so, this was considered a "huge sum" at the time.

Figure 6: Capital expenditure of the top four hyperscale data center operators Source: Compiled by the author based on financial reports from each company and TrendForce data.

However, since OpenAI released ChatGPT in November 2022, the generative AI field has ushered in a wave of enthusiasm, and the capital investment curve has sharply risen. It is expected that by 2025, the total investment of these four companies will reach $355 billion, and by 2026, this figure is expected to reach an astonishing $755 billion. This represents an unprecedented investment boom, with the investment amount increasing about 36 times in just over a decade since 2015.

$755 billion far exceeds 120 trillion yen. Just four companies plan to invest an amount equivalent to the national budget (general account) of Japan within a year for data center and AI infrastructure construction. This illustrates how extraordinary this situation is.

AI data centers are like "black holes"

The ultimate destination of this massive investment is AI data centers. Figure 7 illustrates the current situation in a schematic form.

Figure 7: Why have DRAM and NAND flash prices surged?

As hyperscale data centers compete to invest in AI data centers, the semiconductors essential for AI learning and reasoning: namely, GPUs from companies like NVIDIA, high-bandwidth memory (HBM) used in GPUs, and large-capacity storage SSDs equipped with NAND flash, are being sucked into the "black hole" of AI data centers one after another.

Memory manufacturers prioritize the production of high-margin HBM and high-performance DRAM and NAND flash for data centers. This is a natural business decision. Consequently, production line capacity has been shifted to the AI field, resulting in a significant reduction in capacity available for other applications.

The industries most severely affected are DRAM and NAND flash used in digital consumer electronics such as personal computers, smartphones, and game consoles. There is a severe shortage of storage for these products, reaching a point of "complete inadequacy."

Due to limited supply capacity, if demand is concentrated solely on AI data centers, the storage for consumer electronics will inevitably be depleted. This will lead to fierce competition for limited supplies, further driving up prices. This is directly related to the previously mentioned abnormal surge in memory prices.

In fact, personal computer and smartphone manufacturers have already begun to sound the alarm, stating that they cannot obtain the required memory, and procurement costs have soared, forcing them to pass these costs onto product prices. Ironically, under the shadow of the rise of AI, the digital devices we use daily are becoming increasingly expensive and even scarce.

Semiconductor market forecast "completely wrong"

In 2023, I made a prediction about the global semiconductor market trends up to 2032, as shown in Figure 8. By analyzing the history of the semiconductor industry and combining the "PC effect," "Internet effect," "smartphone effect," and the upcoming "AI semiconductor effect," I predicted that the semiconductor market would grow at a rate of about "doubling every decade."

This prediction estimated that by 2032, the market size would reach approximately $1.2 trillion. At that time, I considered this a fairly optimistic forecast.

Figure 8: Global semiconductor market forecast for 2023. Data source: Based on WSTS data and author predictions.

However, it turns out that this prediction was completely wrong. In fact, it was wrong because it was too conservative. See Figure 9. According to the WSTS (World Semiconductor Statistics) spring forecast for 2026, the global semiconductor market size is expected to reach $630.5 billion in 2024, $795.6 billion in 2025, and exceed $1.5 trillion in 2026, reaching $1.5112 trillion. Furthermore, it is expected to grow rapidly to $1.9137 trillion by 2027, approaching $2 trillion.

Figure 9: Rapid expansion of the global semiconductor market due to the boom in AI, with predictions severely off target Source: Compiled by the author based on WSTS data and WSTS spring 2026 semiconductor forecast.

In just a few years, the market size has easily surpassed the author's previous prediction of $1.2 trillion for 2032. The 2032 forecast has already been realized by 2026. This does not mean that the previous prediction was "too optimistic," but rather that the disruptive power of the AI boom has completely overturned all past conventions in the semiconductor industry.

Figure 10 clearly indicates that this rapid growth is driven by memory (including DRAM and NAND) and logic circuits (including GPUs). Memory is expected to exceed $1 trillion by 2027, reflecting the explosive growth of DRAM + NAND mentioned earlier. The logic circuit market is also expected to exceed $500 billion.

Figure 10: Rapid growth of memory (including DRAM and NAND) and logic (including GPUs) Source: Compiled by the author based on WSTS data and WSTS spring 2026 semiconductor forecast.

On the other hand, the analog circuit and micro semiconductor fields have remained nearly stable. In other words, the overall growth of the semiconductor market is not balanced; rather, only these two AI-related fields are growing at an unprecedented speed, driving the growth of the entire market. This is an extremely distorted growth structure.

How long can this boom last?

Now, let’s answer the most concerning question: "How long will the boom brought about by the rise of AI last?"

The key to understanding this lies in the long history of the memory market. Figure 11 shows the annual growth rate of the memory market over nearly 35 years since 1991. This chart reflects the numerous ups and downs experienced by the semiconductor industry.

Figure 11: Annual growth rate of the memory market Source: Compiled by the author based on WSTS data and WSTS spring 2026 semiconductor forecast.

From the boom brought by the release of Windows 95 in 1995, to the IT bubble of 2000 and its subsequent burst, to the memory bubble of 2017-2018, the collapse of Lehman Brothers in 2008, and the boom brought by the COVID-19 pandemic since 2020, the semiconductor industry has experienced a rollercoaster of booms and busts.

What is noteworthy here is not the height of individual peaks, but the "duration" of sustained positive growth. A careful analysis of Figure 11 reveals a key fact: in any historical period, the longest duration of sustained annual positive growth in the memory market has only been five years. There has never been a case of six or seven consecutive years of positive growth in the past 35 years.

Why does the market tend to stabilize after five years? Because the memory market is essentially influenced by the "silicon cycle," which includes demand surges, price increases, companies increasing production investments, oversupply, and price crashes.

If the economy continues to prosper, companies will inevitably compete to invest in equipment, ultimately leading to oversupply and price crashes. This is the inherent operational mechanism dictated by the nature of memory as a product.

The current AI boom is expected to drive significant growth in the memory market, starting from the low point in 2023 and peaking in 2024. According to historical patterns, this boom should end by 2028 at the latest, and it may even peak in 2027.

While it is inevitable that some will say, "This time is different" or "AI is special," we must not forget that the same statements have been made during every past economic bubble.

The higher the peak, the deeper the valley

The history of the semiconductor industry also tells us another undeniable rule: "The higher the peak, the deeper the valley that follows."

If you look again at Figure 11, this rule becomes evident. The IT bubble peaked in 2000 with an annual growth rate exceeding 50%, but in the following year, it plummeted to a deep abyss of -49.5%. Similarly, the memory bubble of 2017-2018 experienced a peak of over 60%, followed by a sharp decline of 33% in 2019. In this industry, the higher the peak of prosperity, the more severe and prolonged the subsequent recession tends to be.

Let’s re-examine the current AI boom. An annual growth rate of 285% is an unprecedented peak, far exceeding any previous bubble. Its increase is so high that the peaks of the IT bubble and the memory bubble seem trivial in comparison.

Applying this principle, the conclusion is clear. This unprecedented boom is likely to be followed by a "valley" that is deeper and more severe than any that the semiconductor industry has experienced before. We should prepare for an extremely serious economic recession expected to begin around 2027-2028, a recession that cannot be overcome lightly.

The soaring stock prices of memory manufacturers and "billionaires"

Currently, the stock prices of memory manufacturers are soaring across the board. This is entirely expected, as rising prices have led to significant increases in their profits.

A symbolic example is Kioxia. The company's stock price has risen, allowing 600 investors to be considered "billionaires," gaining over 1 billion yen in unrealized profits from their holdings. The entire industry is immersed in an unprecedented celebratory atmosphere.

However, we need to remain calm now. Stock prices are like a mirror, reflecting people's expectations for the future. When these expectations are too high, if reality falls even slightly below expectations, stock prices can collapse instantly.

The higher the mountain, the deeper the valley. This principle applies to stock prices as well, not just market size. In every past bubble burst, there has been a pattern: billionaires who became wealthy at the peak of the bubble had to significantly reduce their assets as the economic recession approached.

Now is the time of economic prosperity, and it is crucial to prepare for the impending economic recession. This prosperity will not last forever. History shows that the positive growth of the memory market will last at most five years, after which a downturn period will follow (at least on the surface). The higher the peak, the deeper the valley. Therefore, I strongly urge companies currently enjoying prosperity: "While the boom continues, you should prepare for the economic recession."