Memory Market Up 11x in a Decade: Can the AI Cycle Break the Five-Year Curse?
nashnova research
Global memory shipments surged from $5.6 billion a month in 2016 to $63.3 billion by May 2026 — an 11-fold jump in a decade. Yet memory has never posted positive growth for more than five consecutive years; whether AI demand can break that pattern will define this cycle's fate.
Eleven times in ten years — how extreme is this memory boom?
WSTS data shows monthly memory shipments hit $63.3 billion, up more than 11× from roughly $5.6 billion in 2016 and about 10.7× from the early-2023 trough.
Year-on-year growth stands at 285%, dwarfing the ~60% peak during the 2017–2018 "memory bubble."
This means → no previous memory cycle comes close. This expansion has no historical benchmark.
Volume barely moved — so why did revenue jump tenfold?
The driver is unit-price spikes, not shipment volume. Per TrendForce, DRAM (DDR5 16Gb) spot prices rose from $4.70 in early 2025 to roughly $46.00 — about 10×.
NAND (1Tb TLC wafer) prices climbed from $2.40 to $25.00 over the same period — also about 10×.
In plain terms = even if not a single extra chip shipped, revenue would still have increased tenfold. This is a price-driven boom.
Who is footing the bill? Where is the hyperscalers' capex going?
Combined capital spending by Amazon, Google, Microsoft, and Meta is projected at $355 billion in 2025, jumping to $755 billion in 2026 — roughly 36× the $21 billion they spent in 2015.
AI data centers are absorbing capacity for GPUs, HBM — high-bandwidth memory, an ultra-fast memory designed specifically for AI chips — and high-capacity SSDs.
This means → AI is siphoning memory supply away from PCs and smartphones, creating severe shortages in consumer electronics and pushing prices up across the board.
Did the market just hit a forecast ceiling six years early?
WSTS's spring 2026 forecast puts the global semiconductor market above $1.5 trillion in 2026, approaching $2 trillion in 2027. Memory alone is expected to surpass $1 trillion in 2027.
That already exceeds the industry's earlier projection of ~$1.2 trillion by 2032 — effectively reaching the target six years ahead of schedule.
This reflects a highly lopsided growth structure: memory and logic (including GPUs) are surging while analog and micro-semiconductors are nearly flat. In plain terms = AI didn't lift the entire chip industry — it only lifted the two segments directly tied to it.
What is the "five-year curse"? What does history say?
Over the past 35 years, the longest streak of year-on-year positive growth in the memory market has been five years — six consecutive years has never happened.
The pattern is the classic "silicon cycle": demand surge → price rise → capacity race → oversupply → price crash.
The higher the peak, the deeper the fall — the 2000 IT bubble peaked above 50% annual growth, then collapsed 49.5% the next year; the 2017–2018 bubble peaked above 60%, followed by a 33% decline in 2019.
Can AI demand break that pattern?
If the current cycle began at the 2023 trough, historical precedent suggests positive growth ends by 2028 at the latest, with a possible peak in 2027.
Two variables are decisive: ① whether AI demand can keep outrunning expectations, and ② whether hyperscaler capex is actually sustained rather than cut back.
In plain terms = history says "five years up, then down." AI says "this time is different." Who is right? We could have an answer as early as 2027.
市场有风险，内容仅供研究参考，不构成投资建议。