The Semiconductor Upcycle Is Being Driven by Memory Pricing, Not Unit Growth
The Semiconductor Upcycle Is Being Driven by Memory Pricing, Not Unit Growth
Dr. Robert CastellanoThu, May 14, 2026 at 1:41 AM UTC
0
Introduction
For decades, semiconductor cycles followed a fairly predictable sequence. Demand improved, companies bought more equipment to make more chips, unit shipments increased so that supply exceed demand, and pricing eventually weakened as oversupply returned to the market. Investors therefore learned to focus primarily on unit growth because pricing historically lagged demand rather than led it.
That relationship has now changed, particularly in memory.
The current semiconductor recovery is not being driven by a surge in semiconductor units. Instead, it is being driven by constrained supply initially, then by a shortage. In many respects, memory suppliers have inverted the traditional semiconductor cycle by limiting capacity additions as a result of increased Capex (capital expenditure) spend in recent years in anticipation of strong demand for AI and concomitant memory chips.
This did not happen accidentally. Following the severe downturn that began in 2022, memory manufacturers spent several years cutting capital expenditures, delaying wafer fab equipment purchases, and slowing capacity expansion. Suppliers became far more disciplined after watching prior cycles destroy profitability through overspending.
For example, in my March 18, 2024 Seeking Alpha article entitled "Micron's Fiscal Q2 2024 Earnings Preview With Eyes On HBMx," I noted:
The analyst who called NVIDIA in 2010 just named his top 10 stocks and SSNLF wasn't one of them.Get them here FREE.
"Memory Pricing and Capex
Memory Pricing Trends: The DRAM industry, including leaders like Micron and Samsung Electronics Co., Ltd. (OTCPK:SSNLF), has observed a strategic pullback in memory production capacity investments (capex), as shown in Chart 1.
Micron aims to allocate approximately 30% of its sales to capital expenditures and around 10% of its sales to research and development. These allocations are in line with the practices of industry peers.
Although a robust demand recovery is anticipated in the second half of 2024, SK hynix is not expected to pursue aggressive capital expenditure ("capex") increases this year. Instead, the company's focus seems to be on technology migration rather than expanding production capacity. SK Hynix has outlined a capex plan of W13 trillion for 2024, representing a nearly 50% year-over-year increase but only reaching 70% of the levels seen in 2022-23. The majority of this spending is likely to occur in the latter half of 2024.
For Samsung, capex on memory primarily focused on enhancing infrastructure at the Pyeongtaek facility in Korea and expanding production capacity for High Bandwidth Memory (HBM), DDR5, and other advanced nodes. Capex will likely drop to about W49 trillion in 2024.
This controlled approach to supply has beneficially impacted memory pricing."
Memory suppliers have yet to learn their lesson. But it doesn't matter this time around. The increased demand for memory chips with little or no Capex spend through 2025, which has resulted in a meteoric growth in ASPs (average selling prices), profit margins, and share price for Micron (MU) and other memory companies.
Memory Is Leading the Recovery
Chart 1 shows memory revenues, particularly DRAM, accelerating sharply following the 2023 trough and continuing higher throughout 2024 and into 2026. The recovery slope is far steeper than what would normally be expected from a simple inventory correction. Revenue growth has been driven primarily by pricing improvements tied to constrained supply and the rapid emergence of AI-related memory demand.
The result is that semiconductor revenue growth is increasingly being generated through higher ASPs rather than materially higher unit shipments. That distinction matters because it changes how investors should interpret the current upcycle and how sustainable it may ultimately become.
Chart 1: Memory Revenue (3mma) by Type - Millions of $
Chart 2 tells a very different story. Memory unit shipments improved only gradually and lagged the recovery in revenues by a meaningful margin. The three-month moving average smooths some of the short-term volatility, but the broader relationship remains clear: pricing recovered first while shipment growth followed later. The recent increase in unit shipments is due DRAM manufacturers tweaking out as much memory as possible and to an increase in capacity by Chinese Memory companies CXMT and YMTC.
Chart 2: Memory Units (3mma) by Type - Millions of Units
Part of this was simple caution. Suppliers were still dealing with the aftermath of the 2022-2023 collapse, one of the most severe downturns the memory industry following a 2018 peak. But unlike prior downturns, suppliers did not immediately rush back into aggressive expansion mode once pricing stabilized.
Advertisement
Instead, they constrained wafer additions, delayed greenfield expansion projects, and focused more heavily on technology migrations rather than pure capacity growth.
Complicating the supply-demand dynamics was the AI infrastructure demand, particularly for HBM (High Bandwidth Memory). One of the biggest drivers of the current memory cycle is the sharp increase in DRAM content required for AI servers using GPUs, CPUs, and custom ASIC accelerators. An 8-layer HBM stack uses eight DRAM dies while a 12-layer stack uses twelve, dramatically increasing memory consumption per system as hyperscalers deploy larger AI clusters. Last year alone, Nvidia shipped several million AI data center GPUs, creating enormous incremental DRAM demand at a time when memory suppliers were still limiting capacity growth. Data Center capacity has grown from 42 GW (gigawatts) in 2023 to 65 GW in 2025.
The broader semiconductor market is beginning to display some of the same characteristics.
Chart 3 illustrates that semiconductor revenues excluding memory continued trending upward through 2025 and into early 2026, although at a more measured pace than DRAM and NAND. Revenue growth is illustrated by the positive trend line (blue dotted line).
Chart 3: Semiconductor Revenues (ex Memory) (3mma) - $M
Meanwhile, Chart 4 shows semiconductor unit shipments excluding memory remaining relatively flat over the same period, as evidenced by the horizontal trend line.
Chart 4: Semiconductor Units (3mma) - M
Pricing Strength Is Being Driven by Supply Constraint
The implication is fairly straightforward. The industry is increasingly generating more revenue per unit rather than simply shipping more units. AI-related semiconductor content growth, richer product mix, and higher ASPs are contributing more heavily to revenue expansion than traditional volume growth.
Memory simply happens to be the most extreme example because supply discipline has been far more aggressive there.
The current pricing environment is not solely the result of strong demand. It is equally the consequence of several years of restrained capital spending.
During the downturn beginning in 2022, memory manufacturers sharply reduced capital expenditures after suffering severe pricing collapses and inventory corrections. Wafer fab equipment purchases slowed, utilization rates were intentionally reduced, and expansion projects were delayed across the industry. This is illustrated in Chart 5.
Chart 5 - DRAM Revenues Q1 2016-Q4 2025
Importantly, much of the spending that did occur was directed toward technology transitions rather than major new wafer capacity. Suppliers focused on node migrations and density improvements instead of aggressively adding wafer starts.
Investor Takeaway
The semiconductor recovery is increasingly being driven by pricing rather than unit demand growth, and memory remains the clearest example of that shift.
DRAM and NAND suppliers spent several years restricting capital expenditures following the 2022 downturn, preventing meaningful oversupply from returning to the market. When AI infrastructure demand accelerated, particularly for HBM, the industry lacked sufficient incremental capacity to respond quickly. Pricing therefore increased well before shipment growth materially recovered.
The broader semiconductor market is beginning to exhibit similar characteristics, although less dramatically than memory. Revenues are increasingly being supported by richer product mix, AI-related content expansion, and higher ASPs rather than by substantial increases in unit shipments.
For investors, this changes how semiconductor cycles should be evaluated. Traditional metrics tied solely to shipment growth no longer fully explain industry profitability. Supply discipline, capital intensity, and pricing behavior are becoming equally important indicators.
Looking ahead, the key issue is not whether semiconductor demand remains healthy. The more important question is whether suppliers maintain the supply restraint that is currently supporting pricing power.
For memory manufacturers Micron, Samsung (SSNLF), and SK hynix, this supercycle shows no signs of slowing because AI is not slowing, and Nvidia is not slowing as it continues its relentless 1-year cadence of introducing more powerful GPUs each year demanding more memory.
The analyst who called NVIDIA in 2010 just named his top 10 AI stocks
This analyst's 2025 picks are up 106% on average. He just named his top 10 stocks to buy in 2026. Get them here FREE.
Source: “AOL Money”
