The global memory chip market is at a pivotal moment. Driven by rapid growth in AI-related demand, prices for DRAM and HBM have surged to multi-decade highs, while supply constraints have reshaped the entire semiconductor landscape.
In this article, we explore what is driving the current upcycle, why it looks different from previous cycles and when the market may begin to turn.
Over the past 12 months, the Korean market has risen by almost 200%, driven by a significant wave of earnings upgrades and strong inflows into Korean memory producers, namely SK Hynix and Samsung Electronics. Looking ahead to 2027, both companies are expected to be among the largest profit generators in the world, with Samsung’s operating profit forecast to be second only to Nvidia’s.
Figure 1. Next twelve months price-to-book ratio (NTM P/B)
Figure 2. P/E and EPS growth for 2026 and 2027
This is clearly no longer a niche or under-the-radar theme, but rather a key consideration and firmly on the agenda of global investors. Consequently, we are fielding a growing number of questions on the subject and thought it would be useful to share our current view on the memory cycle.
For the past 9 months, our positioning has mostly comprised large positions in SK Hynix and Square (the holding company for Hynix, trading at a 50% discount). However, we have recently started to gradually reduce these positions, since the memory space is cyclical, even though the cycle may be longer and stronger this time. SK Hynix is currently trading at 2.5 times book value, which is 43% above the previous peak in 2021. It is difficult to predict what the impact of AI on computing usage is, but it is likely that previous forecasts will be revised upwards. Hence, the demand for AI-related chips is likely to continue to increase. We have redeployed capital to more niche areas of the AI industry, such as copper-clad laminate and high-end circuit board materials. Our new additions all have a common denominator: they dominate a very niche product for which there is a shortage of supply. This implies that prices will continue to rise and that there will be less competition compared to the more commoditised memory space.
We are at a truly historic point in the cycle
The memory cycle bottomed in Q3 2024, with prices exploding from Q3 2025, driven by huge demand for HBM, which fed into regular memory chip production, as well as a massive supply-demand gap for mainstream memory.
The price of mainstream DRAM products has reached a 25-year high, with unprecedented year-on-year increases. As of mid-March 2026, the 16Gb DDR5 spot price reached USD 39.3 million per unit equivalent, up 50% QoQ and 683% YoY. Meanwhile, the 16Gb DDR4 spot price (the previous generation) hit USD 77.3, up 2,537% YoY, far exceeding the previous cycle peak of approximately USD 10 in October 2017.
As most memory is sold on a contract basis (see Figure 4), it is important to stress that this explosive rally has driven an unprecedented widening of the spread between the spot and contract markets. As of mid-March 2026, the spot price of 16Gb DDR4 was trading at a premium of over 150% to its contract price, while the spot price of 16Gb DDR5 was trading at a premium of over 30%. This gap is directly fuelled by the severe shortage of mainstream DRAM in the contract supply chain.
The current 16Gb DDR5 spot price of USD 39.3 is 7–12x higher than the normalised price of USD 3-5 per unit. The 16Gb DDR4 spot price of USD 77.3 is 18–37x higher than the normalised level of USD 2–4 and is nearly 8x higher than the previous cycle peak in 2017.
Figure 3. Memory prices over time
Figure 4. A memory primer
The cause of the price surge is an all-time low in inventory levels across the supply chain. Chipmakers currently have only three to four weeks of stock, which is well below the usual two months. Furthermore, factories are operating at full capacity with no room to ramp up production in the near term.
While hyperscalers with high capital expenditure requirements are relatively price-insensitive, these sky-high costs are having a devastating impact on phone and wider consumer electronics brands. They are being forced to hike prices, cut specifications, delay launches and take a hit on sales. China’s smartphone shipments fell by 13% YoY to 16 million units in February 2026, marking a 21% MoM decline. Rising memory costs were cited as a key factor capping end demand. For Xiaomi, OPPO, Vivo, and other Chinese original equipment manufacturers (OEMs), memory costs as a percentage of the total bill of materials have surged from 15% in 2024 to over 35% in Q1 2026. Many vendors have been forced to either raise end-product prices, reduce memory specifications in mid- to low-end models, or accept compressed gross margins. All of these factors have contributed to weakening shipment momentum.
HBM vs DRAM
Over 95% of global HBM sales come from multi-quarter or annual contracts with leading AI chip designers, such as NVIDIA and AMD, as well as hyperscalers. There is a negligible spot market. Pricing is negotiated for periods ranging from six to twelve months, with minimal sensitivity to short-term fluctuations in the spot market for mainstream DRAM.
SK Hynix, the global leader in HBM, achieves an operating margin of over 80% for its HBM business compared to around 60% for mainstream DRAM. This creates a strong incentive for manufacturers to prioritise HBM production. Consequently, chipmakers are shifting a significant portion of their wafer capacity from regular DRAM chips to HBM, which is directly causing the shortage in DRAM. Furthermore, ramping up HBM production takes far longer, meaning this shortage will not ease in 2026.
When does the cycle typically turn, and when will it turn this time?
Such strong performance makes it easy to forget that Hynix nearly went bankrupt in 2022 and was forced to issue convertible bonds to fund a huge operating loss. However, the memory market is a classic cyclical industry with a predictable pattern tied to factory spending, supply, demand and pricing. Historically, a full cycle lasts three to four years: twelve to eighteen months of growth, followed by eighteen to twenty-four months of decline. The cycle turns when new capacity floods the market, high prices kill demand and inventory piles up excessively.
However, based on current supply and demand, we expect the cycle to, at least for HBM, to only start turning downward around Q3 2027, although price growth will slow. On the supply side, SK Hynix has confirmed that capacity expansion is constrained by a shortage of cleanroom space, a bottleneck that is unlikely to be resolved before mid- to late 2027. Demand, however, is the more consequential variable. The cycle could turn earlier should AI capital expenditure slow, for example if one of the major hyperscalers withdraws from the AGI race. Conversely, it could extend further if memory demand from agentic AI continues at its current pace. Anthropic’s annualised revenue was USD 9 billion at the end of 2025, but is now reported to be USD 30 billion, reflecting the speed at which enterprises are deploying agentic AI. Demand for computing power and memory will inevitably follow suit.
As this becomes more widely recognised, we anticipate yet another leg up in memory stocks. Importantly, earnings revisions have kept pace with share price performance, meaning these names have not re-rated on a price-to-earnings basis. The more conventional valuation measure for this sector, price-to-book, does point to stretched levels, though returns on equity are equally unprecedented. From a risk management point of view, we have been trimming positions accordingly, but we remain overweight and expect to do so for the foreseeable future.
Figure 5. Hynix 2027 earnings revisions and stock price
Figure 6. Hynix 12-month forward PE ratio
Broadening out of the AI trade?
Indeed, the more compelling upside may lie in less prominent companies that play equally vital roles across AI value chains. From an equity standpoint, not every segment is interesting, but there is real value to be found in identifying those with the tightest supply-demand dynamics and market structures dominated by a small number of disciplined players.
Elite Materials
Elite Materials is a leading Taiwan-based manufacturer of high-end copper-clad laminates (CCL), specialising in AI-grade CCL solutions that serve as the core material for high-performance printed circuit boards (PCBs). As one of few long-term certified suppliers to top global chipmakers, including NVIDIA and AMD, as well as leading ASIC designers, the company’s products are indispensable for advanced AI servers, accelerator cards and high-speed computing hardware.
We expect to see further rounds of price increases and hence earnings upgrades in the coming quarters, driven by strong demand, technological upgrades and the passing on of raw material prices. This will continue to drive improvements in sales and profitability across the industry.
Its YTD share price has delivered exceptional performance, beating TSMC by 80% and the Taiwanese market by 85%. This rally is underpinned by 100% production capacity utilisation, explosive AI-driven demand, and locked-in long-term orders from top-tier clients. With a current trading price of 41x 1-year forward P/E and expected EPS growth of 125% (0.33x PEG), it is poised for sustained growth in 2026-2027 through capacity expansion and ongoing global AI capital expenditure tailwinds.
Unimicron
Unimicron is a leading Taiwanese manufacturer of essential circuit board materials for advanced chips. The company specialises in ABF film, a vital component that enables high-end AI graphics chips, HBM memory and modern chip designs to work efficiently.
Its year-to-date (YTD) share price has achieved remarkable outperformance, surpassing TSMC by 130% and the Taiwan market by 135%. This rally is driven by full-capacity production, surging order volumes that far exceed current output and rapidly expanding gross margins, all amid a tight supply-demand balance. Trading at 40x 1-year forward P/E with 74% expected EPS growth (0.54x PEG), Unimicron is well-positioned for sustained growth in 2026–2027 through targeted ABF capacity expansion and increasing demand for advanced packaging.
