Memory makers hit 80% margins as AI demand squeezes PC and phone supply

Record profits for memory makers

Memory chip manufacturers are currently among the most profitable products in the technology sector. As MLQ.ai reported, companies like Micron, Sandisk, and SK hynix have each reported operating margins of about 80% in their latest quarters. This profitability comes as their revenues have surged between 328% and 379% compared to the previous year.

The primary driver of this windfall is the artificial intelligence (AI) boom. Servers built for Nvidia and AMD accelerators require massive amounts of high-bandwidth memory (HBM), large pools of conventional DRAM, and fast solid-state storage. Memory makers are intentionally steering their wafer production toward these high-demand AI customers to maximize profit.

Micron converted $54.23 billion in revenue into $37.70 billion of net income for the period ending September 3. Similarly, SK hynix reached an 80% margin, generating 83.3 trillion won in operating profit from 104.6 trillion won in revenue. These figures represent a massive shift in the industry's financial landscape, where memory is no longer a commodity but a high-margin priority.

The supply crunch for consumers

While memory makers are seeing record gains, the rest of the tech world is facing a supply squeeze. TrendForce forecasts that conventional DRAM contract prices will rise another 10%-15% this quarter, while NAND flash prices could jump 15%-20%. This follows significant price increases earlier in the year, creating a challenging environment for manufacturers of everyday electronics.

The shortage is caused by the physical requirements of AI hardware. HBM takes considerably more wafer area per bit than conventional DRAM, meaning every HBM stack a supplier ships takes capacity away from PCs, phones, and standard servers. TrendForce notes that suppliers are prioritizing capacity for servers and high-margin AI products over consumer needs.

This shift is expected to have long-term consequences for the personal computing market. TrendForce predicts that the availability of DRAM for PCs may drop in 2027 as production capacity is diverted toward server applications. Consequently, PC makers are already shipping smaller solid-state drives (SSDs) to help hold down prices for consumers who are not using AI-heavy hardware.

Apple absorbs the rising costs

Apple serves as a primary example of a major buyer currently absorbing these increased costs. The company's fiscal 2026 annual report notes that it is facing manufacturing hurdles and higher expenses caused by imbalances in global supply and demand. It specifically expects these trends to intensify for memory and storage components as the demand for AI infrastructure continues to grow.

The financial impact is already visible in Apple's margins. The company's September-quarter gross margin was about 46.9%, which was roughly one point lower than it was in June, primarily due to the rising cost of memory. Apple guided its December-quarter gross margin to 45%-46%, citing memory costs as a primary factor alongside the expense of its Siri AI ramp.

The scale of Apple's commitment to these components is massive. The company reported $89.7 billion of manufacturing purchase obligations as of September 26, with $82.7 billion of that amount payable within the next 12 months. Chief financial officer Kevan Parekh noted on a recent earnings call that these obligations rose by $33 billion in just one quarter, highlighting the pressure of securing components in a tight market.

Nvidia and TSMC move into memory

The demand for specialized memory is so high that even the largest customers are looking to move deeper into the manufacturing process. Nvidia recently unveiled NVHBM, a custom memory solution that places Nvidia's own memory controller in the base die. To produce this, Nvidia has asked Samsung, SK hynix, and Micron to use TSMC's 3-nanometer process for the base dies of its custom HBM.

This move represents a significant shift in how these companies collaborate. Samsung will partner with TSMC during the HBM manufacturing phase for the first time, while maintaining its own in-house process for standard HBM and other clients. TSMC has also started looking into creating 3D DRAM, which arranges memory cells in a vertical stack within a single die, at Nvidia's request.

While TSMC says it remains focused on its core business as an IC foundry, the move toward 3D DRAM suggests a desire to solve the looming capacity crisis. However, industry experts judge that 3D DRAM could not ease the current shortages for at least five years. In the meantime, companies like Qualcomm are exploring alternative designs, such as bonding DRAM layers directly onto a logic die, to bypass some of the HBM supply constraints.

Global competition and expansion

As Western companies focus on high-end AI memory, China is expanding its own capacity to fill the gap. China's largest flash maker, YMTC, aims to source up to 70% of the equipment for a planned fourth Wuhan fab from Chinese suppliers. This plant is intended to produce NAND with more than 300 layers, specifically targeting data centers and AI storage applications.

In the DRAM market, Chinese firms like CXMT are also expanding, though they currently hold a smaller share of the market. Counterpoint Research puts CXMT's DRAM bit-shipment share at about 9%, with a goal to increase capacity to 420,000 wafers a month by 2027. However, CXMT's HBM3 is not yet in mass production, positioning the company in the conventional DRAM market that Korean and U.S. suppliers are currently deprioritizing.

The expansion of these facilities faces significant geopolitical hurdles. Samsung's and SK hynix's China fabs require yearly government clearance for American hardware, and Washington has indicated it will not permit gear that significantly grows capacity in China. This creates a complex landscape where the world's most profitable memory components are caught between intense AI demand and tightening international trade controls.