
KEY POINTS
- DRAM contract prices have surged over 100% in 2026 as Samsung, SK Hynix, and Micron redirect wafer capacity to high-bandwidth memory for AI accelerators.
- HBM now consumes 23% of total DRAM wafer production, squeezing supply for consumer and enterprise memory and sending DDR5 module prices up 60% since September.
- Samsung warned in its April 30 earnings report that significant shortages will persist through at least 2027, meaning the memory trade has more room to run.
The three companies that control over 95% of global DRAM production have collectively decided that your laptop's RAM matters less than Nvidia's next AI server, and the price charts reflect it. Contract pricing for DDR5 has blown past $19.50 per unit, more than double the roughly $7 level from early 2025, according to industry supply chain data. Samsung's 32GB DDR5 modules now retail at $239, up from $149 in September, a 60% increase that is rippling through every tier of the PC and server supply chain.
The cause is straightforward. High-bandwidth memory chips used in AI accelerators generate three to five times more revenue per wafer than conventional DDR5. When the math is that clear, Samsung, SK Hynix, and Micron are not going to keep running commodity DRAM lines out of charity. HBM now accounts for 23% of total DRAM wafer starts, up from single digits two years ago, and every wafer converted to HBM is a wafer that does not produce traditional memory.
Micron Exits Consumer, Goes All-In on AI
Micron has taken the most aggressive stance of the three. The company recently exited the consumer memory market entirely, telling analysts on its last earnings call that its capacity is "sold out for 2026." That is not corporate hyperbole. Micron's revenue per bit from traditional DRAM is forecast to climb 54% year-over-year to $1.06 in 2026, but the company has calculated that every bit of HBM it can produce earns far more. For traders, this means Micron is no longer a cyclical memory play that whipsaws with PC shipment data. It has become a pure AI infrastructure bet, and the stock has reflected that transformation, becoming one of the hottest trades on Wall Street over the past 12 months.
SK Hynix is in the same position. The South Korean chipmaker reported during its October earnings call that HBM, DRAM, and NAND capacity are all "essentially sold out" for the full calendar year. Samsung, the largest of the three, has been slower to pivot but is catching up fast. In its full earnings report released April 30, memory division chief Kim Jaejune warned that "significant shortages" across memory products will continue through at least 2027.
The Downstream Squeeze
The shortage is not just a story about chipmakers printing money. It is creating real pain downstream. IDC flagged the crisis as a potential drag on smartphone and PC shipments in 2026, estimating that device OEMs face 15-20% higher bill-of-materials costs for memory alone. Server OEMs building non-AI infrastructure are getting squeezed even harder, as hyperscalers with AI budgets hoard whatever supply remains.
Samsung's revenue per bit from traditional DRAM is forecast to rise 116% year-over-year to $0.79, the steepest increase among the big three. SK Hynix is projected at a 78% jump to $0.70. These are not incremental pricing improvements. They represent a structural repricing of a commodity that most investors had written off as permanently deflationary.
The investment case splits into two trades. The direct play is the memory makers themselves, all of which are printing record margins on AI memory while simultaneously benefiting from scarcity pricing on legacy products. The indirect play is short the companies that absorb the cost: PC OEMs, budget smartphone makers, and any enterprise hardware vendor that cannot pass through 100%-plus memory cost increases to customers.
What Comes Next
The bull case for memory stocks rests on duration. If Samsung is right that shortages persist through 2027, current pricing power is not a one-quarter phenomenon but a multi-year structural shift. Bank of America's semiconductor team has raised its 2026 memory spending forecast to $633 billion, up from $216 billion last year, a number that captures both volume and pricing gains.
The risk is that AI capex eventually plateaus and the HBM-to-conventional conversion reverses, flooding the market with supply and crashing prices the way memory cycles always end. But with Alphabet raising full-year capex guidance to $180-190 billion and Microsoft pushing past $40 billion, the demand signal shows no sign of weakening. Traders should watch Micron's next earnings call for any shift in the "sold out" language. Until that changes, the memory shortage trade stays on.

