The semiconductor memory market is entering a phase of a structural supercycle that defies conservative forecasts of an imminent peak. Analysts adhering to outdated models risk underestimating the scale of the current shift.
A recent forecast suggesting that the memory market's growth rate will peak year-over-year and that DRAM contract prices will hit a high in the fourth quarter of 2026 raises serious doubts. The key issue with such analysis is the use of a classic approach focused on the cyclical nature of consumer goods: PCs and smartphones. This method, which worked for decades, is now inadequate for reality.
What is the fundamental error?
The main driver of the modern market is not ordinary DRAM, but HBM (High Bandwidth Memory). HBM is a fundamentally different product. It is not a commodity that periodically goes into oversupply when new factories are launched. HBM is a strategic asset, "embedded" into the architecture of specific chips. Its production is so complex that simply "turning on" new capacity is impossible.
A telling example is one of the leading manufacturers, Micron. The company officially stated that all its HBM production capacity for 2026 is already fully sold out. Moreover, the shortage will persist at least until 2027. Meanwhile, Micron can only meet 50–67% of current demand from key clients in the medium term.
Financial indicators speak for themselves
Micron's results for the third fiscal quarter of 2026 leave no room for pessimistic scenarios. Revenue reached $41.46 billion, exceeding expectations by 346%. Gross margin set a new historical record at 84.9%. The forecast for the next quarter—$50 billion in revenue—also shows no signs of a slowdown.
Another marker of market strength is the signing of long-term contracts. In the last quarter, Micron signed 16 strategic agreements with clients. These are multi-year contracts with volume commitments and minimum prices, which completely changes the market structure, moving away from short-term spot transactions.
Demand for HBM is embedded in the roadmaps of all major AI accelerator manufacturers—Nvidia, AMD, and data center operators developing their own chips. According to TrendForce, the total revenue of the memory market in 2026 will grow by 134% to $552 billion. The HBM market will reach $100 billion by 2028, up from $35 billion in 2025.
My expertise: Forecasts of an imminent peak in the memory market are a classic case of analysts using a "thirty-year-old map" to navigate a completely new reality. The current cycle is not a cyclical spike but a structural shift driven by explosive demand for AI infrastructure. Investors should focus on the fundamental indicators of HBM manufacturers, not outdated macroeconomic models.