The semiconductor industry has recently seen a dramatic reversal that caught many insiders off guard. 

 Just a few years ago, High Bandwidth Memory (HBM) was widely seen as the “golden chip of the AI era” and a key profit driver for the semiconductor industry. The reasoning was straightforward: HBM is essential for AI large language models, with complex manufacturing processes and high technical barriers. Securing HBM orders was widely viewed as a near-certain path to strong profitability.

 However, the latest market reports reveal a surprising shift: DDR5 memory is now delivering higher profit margins than HBM. 

 Record Profit Margins: DDR5 Leads with 81% 

 Micron Technology recently announced that its third-quarter profit margin is expected to hit 81%, up from a record 75% in the previous quarter. According to Micron CEO Sanjay Mehrotra, profit margins for non-HBM products have now overtaken those of HBM.

 And Micron isn’t alone. SK Group Chairman Chey Tae-won confirmed that HBM profit margins stand at roughly 60%, while standard DRAM chips are reaching 80%.

 In other words, the industry’s heavy focus on high-end HBM has inadvertently turned standard DDR5 chips into a more profitable segment—an outcome few could have predicted. 

 Understanding the Profit Shift: A Supply-Demand Imbalance 

 At the heart of this reversal is a classic supply-demand imbalance. 

 HBM is a high-end product with complex manufacturing and higher costs, while DDR5 is a standard, widely used memory chip with simpler production and lower costs. 

 During the AI boom, Nvidia and major cloud providers aggressively ramped up HBM orders, prompting Samsung, Micron, and SK hynix to allocate significant production capacity and R&D resources to HBM. 

 The issue is that wafer fab capacity is limited. Over-investing in HBM squeezed standard DRAM production, creating tight supply for DDR5. At the same time, DDR5’s manufacturing costs are much lower than HBM’s. As DDR5 prices surged due to constrained supply, profit margins soared. 

 In short, by doubling down on high-end HBM, the industry overlooked DDR5, which has unexpectedly become the more profitable segment. 

 Industry Strategies: Balancing Profit with Long-Term Stability 

 Despite achieving 81% profit margins, Micron has avoided a short-term strategy of focusing solely on high-margin HBM. Instead, it offers comprehensive AI server memory solutions that include both HBM and DDR5. The logic is clear: AI servers need both. Supplying only HBM would prevent customers from building complete systems, potentially driving them to competitors. 

 Micron has also signed five-year strategic agreements with key customers, covering both supply and co-development of future products. This ensures long-term collaboration regardless of memory price fluctuations. 

 Samsung has taken a more conservative route with what it calls “volume lock and price stability.” The company negotiates long-term contracts that require prepayment for three to five years of product supply. Prices fluctuate with the market but within predefined upper and lower bounds, effectively hedging against market cycles. 

 All eyes are now on SK hynix, the undisputed leader in HBM. SK was the first to disclose that DDR5 profits exceed those of HBM, and is expected to announce a DRAM price stabilization plan soon. The industry is watching closely: will SK cut prices and expand capacity, or maintain tight control to preserve high margins? Its decision is likely to set the tone for the global memory market. 

Risks Behind High Margins

 High profit margins are rarely permanent. Two major risks loom: 

 Opportunities for Competitors: Customers cannot absorb high prices indefinitely and may start looking for cheaper alternatives. Reports suggest Apple is considering sourcing memory for the next iPhone from Chinese suppliers to diversify its supply chain and reduce costs. If that happens, the dominance of the three major players could be challenged. 

 The Semiconductor Cycle: The industry follows a familiar pattern: price increases lead to capacity expansion, which eventually results in oversupply and price declines. The current tight supply is expected to last no more than two years. By 2027, new capacity will come online, likely triggering a price correction. The long-term contracts signed by major vendors are a way to lock in revenue while market conditions remain favorable.

 Conclusion

 The recent surge in DDR5 profitability serves as a reminder: in the AI era, HBM is not the only critical component driving the memory market. The industry’s previous focus on high-end products overlooked the enduring value of widely used standard memory. During periods of supply-demand imbalance, these standard products can deliver unexpectedly high returns. 

 The strategies of leading suppliers also highlight an important lesson: in the semiconductor market, long-term resilience matters more than short-term gains. Every long-term contract and strategic move is designed to ensure stability when market conditions inevitably shift. 

 All eyes are now on SK hynix’s pricing strategy—and how long current memory prices can hold. The next chapter in the global memory market is about to unfold.