Morgan Stanley's forecast of an imminent peak in the memory market is facing serious criticism. Analysts at Milk Road AI argue that the bank is using outdated methodology, while real data points to the start of a supercycle, not the end of growth.
What is wrong with the classic approach?
Morgan Stanley believes that the growth rate of the memory market will peak on an annual basis, with contract prices for DRAM reaching a maximum in the fourth quarter of 2026. However, according to Milk Road AI experts, the bank is mistakenly applying a classic analysis method built on cycles of consumer PCs and smartphones. These segments are indeed cyclical and subject to inventory fluctuations.
The problem is that today's demand structure is fundamentally different. The key product is HBM (High Bandwidth Memory), not regular DRAM. The difference between them is fundamental. HBM is not a commodity that goes into oversupply every few years due to the launch of new factories. It is a strategic asset developed for specific chip architectures. HBM production is so complex that simply opening a new factory and starting production is impossible.
Reality vs. forecast: Micron data
Manufacturer Micron Technology itself has already confirmed that all its HBM production capacity for 2026 is fully sold out, and the shortage will persist at least until 2027. Moreover, the company can only meet 50–67% of key clients' needs in the medium term. This does not look like a "peak"—it is a classic sign of structural deficit.
Micron's financial results also refute the downturn scenario. The company's revenue for the third fiscal quarter of 2026 reached $41.46 billion, exceeding expectations by 346%. Gross margin hit a record high of 84.9%. The forecast for the next quarter—revenue of $50 billion—clearly does not fit the "peak" picture.
Why is this a supercycle?
Analysts point to several factors shaping the supercycle. First, the very nature of contracts has changed. In the last quarter alone, Micron signed 16 strategic agreements with clients—these are multi-year contracts with volume commitments and minimum prices, not short-term deals. Second, demand for HBM is embedded in the long-term plans of AI accelerator manufacturers: Nvidia, AMD, and all major data center operators developing their own chips.
The market's scale is also confirmed by TrendForce analysts' forecasts. According to their data, total memory market revenue in 2026 will grow by 134% to $552 billion, and the HBM market volume will reach $100 billion by 2028, up from $35 billion in 2025.
My opinion: Morgan Stanley's forecast is an attempt to apply a thirty-year-old template to a completely new reality. The memory market is experiencing not a cyclical upturn, but a fundamental shift driven by explosive AI demand. Those predicting its imminent peak are simply using the wrong map.