Summary of Key Points
This article discusses how the global semiconductor market, particularly for memory chips (DRAM and NAND), has experienced unprecedented growth since the release of ChatGPT in 2022. The market size has tripled in just three years, while prices have increased tenfold in one and a half years. The underlying reason is the surge in computing demand driven by the AI revolution, with large technology companies (such as Amazon and Microsoft) aggressively building AI data centers that have led to a shortage of storage chips for consumer devices (computers and smartphones), resulting in soaring prices. However, this growth cannot continue indefinitely, and it is expected to peak between 2027 and 2028 due to historical trends and physical limitations.
Why the Storage Market Suddenly “Boomed”? – AI as the Trigger
The semiconductor market had been growing steadily, from $50 billion in 1990 to over $400 billion in 2020, driven by the popularity of computers, the internet, and smartphones. But with the launch of ChatGPT, the market saw a dramatic surge: $630.5 billion in 2024, expected to reach $1.5 trillion by 2026, and nearly $2 trillion by 2027.
The most significant growth has been in memory chips (DRAM and NAND), which saw their value drop from $92.3 billion in 2023 to an estimated $1.06 trillion by 2027, a tenfold increase in just three years. Even the growth of AI chips (such as NVIDIA GPUs) cannot keep up. This is because AI not only requires computing power but also massive amounts of storage to process and store data—just like a computer needs a powerful CPU, sufficient memory, and a hard drive to run complex games.
The Truth Behind the 10-Fold Price Increase: AI Data Centers as a “Storage Black Hole”
The size of the storage market is calculated by multiplying price by quantity. The recent explosion in prices was primarily due to soaring demand: DRAM prices increased from $4.7 in 2023 to $46, and NAND prices rose from $2.4 to $25, both showing a tenfold increase.
The reason for the price surge is the overwhelming demand from large technology companies (Amazon, Google, Microsoft, Meta) building AI data centers. Their investment in these data centers is expected to rise from $91 billion in 2020 to $755 billion by 2026, which is equivalent to Japan’s annual budget. These data centers act like “black holes,” absorbing all the storage chips (used for memory and storage).
On the supply side, chip manufacturers are limited in their production capacity, with most chips being allocated to AI data centers, leaving very few for consumer devices. This imbalance between supply and demand has led to skyrocketing prices.
Why Does AI Require So Much Storage? – The Explosion in Computing Demand
You might think that searching on Google or using ChatGPT is similar, as both involve entering questions and waiting for answers. However, the amount of computing required behind these tasks differs by a factor of 10,000 to 1 million!
- A simple Google search requires 1 billion to 10 billion floating-point operations, which can be handled by a standard CPU.
- GPT-5’s reasoning process, on the other hand, requires 100 trillion to 1000 trillion floating-point operations, necessitating 150 to 400 GPUs, resulting in energy consumption that is 30 to 100 times higher and costs that are 500 to 3000 times greater.
With one billion people using AI systems daily, the demand for storage is enormous—similar to switching from drinking from small cups to large buckets; the supply simply cannot keep up.
How Long Can This Growth Continue? – A Turning Point Expected in 2027-2028
The article suggests that this growth will not last forever due to two key limitations:
1. Historical Trends: The storage market has never experienced five consecutive years of positive growth. Despite high demand, prices have always dropped after a few years due to oversupply or declining demand (for example, the storage market crashed in 2018). With the AI boom starting in 2024, a reversal is likely between 2027 and 2028.
2. Physical Constraints: The supply of semiconductor chips cannot keep up with the exponential growth in demand. Building new factories takes several years, and there are limitations on resources such as electricity, water, and equipment. For instance, it takes 3 to 5 years to construct a new factory, and power grid expansions are even slower. Eventually, even with strong demand, there will not be enough chips available for production—these are the physical barriers that prevent continued growth.
Conclusion
The AI revolution has created a “super cycle” in the storage market, but it cannot continue indefinitely. We are currently at the steepest part of this upward trend, and we need to prepare for a potential turning point in the next few years. Either the storage market will return to its normal trajectory or the AI boom will slow down due to insufficient supply. For individuals, this may mean higher costs for computers and smartphones in the short term, but in the long run, the market will likely return to more rational levels.