NAND， Suddenly getting cold?

In the past decade, NAND flash memory has been regarded as a star in the storage industry. Whether it is the popularization of smartphones, the trend of PC replacement, or the rise of cloud computing, NAND has played a crucial role. Its high growth has also driven a global semiconductor industry expansion boom, forming several oligopolistic patterns such as Samsung, SK Hynix, Micron, and Kaixia.

However, since 2024, the industry atmosphere has taken a sharp turn for the worse. NAND prices have experienced drastic fluctuations, putting pressure on corporate profits. Major manufacturers have coincidentally slowed down production expansion or even reduced investment, marking the arrival of a new cycle.

At the same time, the rise of AI and high bandwidth memory (HBM) has gradually shifted the market's focus to the DRAM field, redefining the position of NAND in the storage industry landscape. The adjustments made by major giants in NAND research and development production indicate that this industry has shifted from "high-speed expansion" to "cautious investment", and a slow yet intense change is quietly taking place.

The slowdown and transfer of the two giants in South Korea

As a long-term dominant player in the global NAND market, Samsung has been known for its aggressive investments in the past.

At this year's International Solid State Circuit Conference, Samsung announced the upcoming release of its V10 (10th generation) NAND flash memory. It is reported that Samsung's V10 NAND flash memory has over 400 active layers and an interface speed of 5.6 GT/s. It also adopts Samsung's exclusive and pioneering Hybrid Bonded Peripheral Unit (CoP) architecture.

However, Samsung's V10 NAND mass production journey has not been smooth. It was originally expected to start mass production by the end of this year, but in June, there were reports in South Korea that Samsung was expected to conduct an evaluation of its supply chain composition before the second half of this year, and the actual large-scale production investment may not be carried out until the first half of next year.

According to reports, as of June this year, Samsung has not yet finalized its supply chain for NAND core devices, including etching. The reason is that the market demand for high-level stacked NAND is unclear, and the cost-effectiveness issues brought by the introduction of new processes hinder investment promotion.

The so-called "etching" refers to the process of removing unwanted substances on a wafer. In the past, when etching channel holes (similar to small holes), it was necessary to perform at a low temperature environment of about -20 ℃ to -30 ℃; But in V10 NAND, it is expected to require an ultra-low temperature environment of -60 ℃ to -70 ℃. The lower the temperature, the weaker the chemical reactivity, and the more precise etching can be achieved without the need for additional protective film.

To this end, Samsung Electronics has introduced ultra-low temperature etching equipment from major front-end equipment manufacturers such as Lam Research in the United States and Tokyo Electric (TEL) in Japan, and conducted trial production and quality evaluation. However, the actual evaluation results show that ultra-low temperature etching technology is still difficult to directly apply to mass production. So Samsung is negotiating with Lam Research and TEL to try to conduct equipment evaluation again by increasing some etching temperatures.

It is understood that the investment cost brought by the introduction of new devices is also considered one of the important reasons for Samsung's delay in investing in V10 NAND mass production. Samsung still mostly uses Lam Research's equipment in NAND etching processes. If TEL is included in the supply chain, it means diversification of equipment, but at the same time, it will face the problem of decreased usage of existing Lam Research equipment and the need to improve compatibility between the two devices.

The difficult production of V10 NAND ultimately affected Samsung's plans for upgrading other production lines.

According to South Korean media reports, Samsung Electronics has been promoting conversion investments in the Pingze P1 factory and Xi'an NAND factory since the beginning of this year, with the core being to shift the originally mass-produced 6th and 7th generation NAND to 8th and 9th generations. Compared to fully constructing a new production line, the cost of conversion investment is lower, and partial renovation and continued use of existing equipment result in higher efficiency.

However, the conversion investment speed of the most advanced NAND is slowing down recently. Although the 8th generation NAND conversion of P1 factory is proceeding as planned, there have been reports of a delay in the investment for the 9th generation NAND conversion, which was originally scheduled to start as early as the second quarter of this year.

The situation at the Xi'an factory is similar. The X1 production line, which is undergoing the 8th generation conversion, is nearing completion, while the X2 production line, which is undergoing the 9th generation conversion, only plans to invest in a monthly scale of 5000 wafers in the third quarter of this year, which is almost the minimum scale required for mass production of memory products.

A semiconductor industry insider said, 'Samsung Electronics plans to continue mass producing older generation NAND such as V6 on its X2 production line before the first quarter of next year, and the true implementation of the 9th generation conversion will not take until at least the middle of next year,' explaining that 'this is because the demand for advanced NAND is still sluggish.'.

However, Samsung remains cautious about investing in next-generation NAND and related technologies. It had originally considered applying hybrid bonding technology to V9 NAND on the Xi'an X2 production line, but has recently decided to put it on hold.

Behind the delay in NAND technology is a lack of market confidence and a reconsideration of capital utilization. Compared to Samsung's high-intensity investment in HBM and DRAM, its conservative attitude in the NAND field is more prominent.

Equally lacking in confidence is SK Hynix, which made it clear during the conference call that its NAND business will remain cautious and prioritize profitability.

It is worth noting that SK Hynix completed the acquisition of Intel's NAND business this year and is currently restructuring its factory and related assets in Dalian, China. However, the facility investment for its second factory, which began construction about three years ago, is still on hold.

Even before acquiring Intel's NAND business, SK Hynix had been discussing building a second factory in Dalian. In 2022, the company held a groundbreaking ceremony in Dalian and announced plans to continue expanding its 3D NAND production capacity in China.

The industry initially expected SK Hynix to introduce infrastructure, including cleanrooms, according to its construction plan starting in mid-2023. However, due to geopolitical factors making it difficult to introduce advanced semiconductor manufacturing equipment into China, and the weak performance of the NAND market, the investment plan has been postponed.

Korean media pointed out that SK Hynix Dalian Second Factory has not made any equipment investment in the past three years since its foundation.

A semiconductor industry insider said, "Although SK Hynix has obtained the 'Certified End User (VEU)' qualification in the United States and investment restrictions in China have been relaxed, the NAND business still faces great uncertainty due to the economic downturn and the entry of newcomers from China. ”He added, "Therefore, there have been no discussions on the specific investment for the new factory in Dalian at present

In fact, over the past few years, despite acquiring Intel's NAND business, SK Hynix has still been at a relative disadvantage in this field, and now the AI driven HBM demand has exploded, allowing it to rediscover its core growth points.

At present, SK Hynix holds a leading position in the HBM field, almost monopolizing NVIDIA's AI acceleration card supply chain, and its profitability has significantly improved. In contrast, its V10 NAND development progress is relatively lagging behind, and the company has also focused its resources on advanced DRAM and HBM internally. This strategic shift clearly indicates that NAND is gradually being marginalized within SK Hynix and will not be the main investment direction, at least in the short term.

American strategic contraction, Japanese pressure, domestic breakthrough

Micron's actions also confirm the difficulties faced by the NAND industry.

In August of this year, Micron announced that due to the continued weak financial performance of mobile NAND products in the market and the slowdown in growth compared to other NAND opportunities, Micron will cease the development of future mobile NAND products globally, including terminating the development of UFS5 (fifth generation universal flash storage).

Micron pointed out that this decision only affects global mobile NAND product development work. Micron will continue to develop and support other NAND solutions, including SSD, automotive, and other terminal NAND solutions. At the same time, Micron will continue to support the mobile DRAM market and provide DRAM product combinations.

Faced with profit pressure, Micron directly withdrew from competition with Samsung and SK Hynix in the consumer market. The company will shift its focus to the enterprise SSD, automotive, and data center markets, which, although not as large as smartphones, have more stable demand and higher profit margins.

At the same time, Micron has significantly increased its research and development, as well as capacity building, in HBM and DRAM, in an attempt to carve out a bigger share of the pie in the AI era. According to its financial report, since the second half of 2024, Micron has continuously raised its revenue and gross profit margin guidance, resulting in significantly better market performance than its NAND business. This differentiation pattern of "NAND sluggish, DRAM strong" reflects Micron's strategy of shrinking NAND to free up resources for future AI memory.

Compared to Samsung and Hynix, the situation of Armor Hero is even more difficult. As the world's third-largest NAND supplier, Kaixia has long relied on technology and production capacity cooperation with Western Digital. However, the merger between the two companies, which has been brewing for many years, has yet to materialize, making it difficult for Kaixia to achieve economies of scale that can compete with Korean giants.

The lack of scale advantage means that Kaixia is in a passive position in cost control, technological iteration, and market discourse power. At the same time, the drastic fluctuations in NAND prices directly dragged down Kaixia's financial performance, causing its performance to hover on the edge of profit and loss for years.

Faced with the pressure of the capital market, Kaixia needs to maintain basic R&D and production investment to maintain competitiveness, but lacks sufficient financial flexibility to promote larger scale expansion or technological upgrades, falling into the dilemma of "unable to advance and unable to retreat".

On the other hand, China's Yangtze River storage has chosen to increase investment against the trend and embarked on a differentiated development path. With the support of domestic market demand, Changjiang Storage has been able to maintain its expansion momentum during the global industry's widespread "scale reduction" phase, winning a strategic window for future competition.

However, from a global perspective, the NAND market is still dominated by oligopolies, with established manufacturers such as Samsung, SK Hynix, Micron, and Kaixia still firmly controlling most of their production capacity and channel resources. Even though the Yangtze River storage is rapidly catching up, it is difficult to completely shake this structure in the short term. The overall atmosphere remains cautious, with companies placing greater emphasis on risk control and capital returns in investment and expansion, awaiting the arrival of the next growth cycle with clear demand.

The 'cold wave' of equipment manufacturers

The most direct victims of the slowdown in NAND investment are undoubtedly semiconductor equipment manufacturers. In the past few years, local equipment companies in South Korea, such as SEMES (a subsidiary of Samsung), Jusung Engineering, Dongjin Semichem, etc., have relied heavily on Samsung and Hynix's expansion projects to maintain growth. Now, with the two giants delaying the mass production plans of advanced NAND projects, related equipment orders have significantly declined, and cash flow and profitability have been severely impacted.

Global device giants have not been spared either. ASML's EUV lithography machines mainly serve the logic and DRAM processes, and their DUV equipment shipments are under pressure during low NAND demand; TEPCO Electronics (TEL) has a high dependence on NAND business in the fields of thin film deposition and etching, and delayed orders directly affect its performance; Applied Materials and Lam Research have felt the most obvious impact in the deposition and etching processes, especially in the demand for ALD, CVD, and high aspect ratio etching equipment for high-level NAND. This has forced both companies to lower their shipment expectations. Kokusai Electric、 Japanese companies such as Hitachi High Tech and SCREEN Semiconductor have also been hit by chain reactions.

Faced with this dilemma, equipment manufacturers have to accelerate their "pivot". With the explosive demand for AI driven HBM and high-performance DRAM, as well as the continuous increase in advanced logic processes (3nm/2nm), equipment giants are gradually shifting Jusung Engineering's focus from NAND to logic chips and HBM/DRAM related devices to fill the NAND investment gap. Dongdian Electronics has repeatedly emphasized the stability of its DRAM business in its financial report, while Kelei and Applied Materials are actively promoting the layout of advanced logic and wafer level packaging equipment.

At the same time, the popularization of the "conversion investment" model is also changing the industry ecology. More and more wafer fabs are choosing to purchase second-hand equipment and carry out remanufacturing or renovation to extend the service life of the equipment, in order to reduce overall capital expenditures. This trend has driven a rapid rise in the second-hand equipment market, leading to business growth for companies such as Surplus GLOBAL, AG Semiconductor, MTM, and others. However, this has further compressed the market demand for new devices, causing equipment manufacturers that previously relied on large-scale NAND expansion to feel an unprecedented chill in the short term.

NAND， How to break the deadlock

The NAND market entering a downturn is not due to a single reason, but a combination of multiple factors:

On the one hand, the global shipment growth of smartphones is sluggish, the PC replacement cycle is prolonged, and the demand for traditional mobile NAND is under long-term pressure. On the other hand, the explosive growth in demand for HBM and DDR5 driven by AI has tilted capital and R&D manpower towards these emerging fields, and NAND is gradually being marginalized.

In addition, NAND is also facing capital pressure. Over the past three years, the total investment in the semiconductor industry has reached a historic high, forcing manufacturers to control costs and invest limited resources in businesses with faster returns.

Finally, we cannot ignore the technological challenges. Currently, the number of NAND stack layers is approaching the limit of 400, and process difficulty, yield issues, and cost control have become bottlenecks, greatly increasing investment risks.

In the foreseeable future, the NAND market will continue to maintain a low speed, investment will be cautious, and manufacturers will allocate more resources to DRAM and HBM.

However, in the medium term, with the popularity of AI training, edge computing, high-capacity SSDs and other scenarios, NAND may rediscover growth opportunities. For example, in the storage systems required for future generative AI, high-capacity and low-cost NAND may have unique value.

Of course, NAND manufacturers are also looking for breakthrough opportunities like HBM, such as the recent collaboration between Sandisk and SK Hynix aimed at standardizing "high bandwidth flash" (HBF). HBF is a memory technology based on NAND flash memory, built into a package similar to HBM. This marks the first solid step in the industry towards integrating flash memory and DRAM like bandwidth into a single stack, with the potential to fundamentally change the way AI models access and process data on a large scale.

But in the long run, the dual breakthroughs in technology and market will truly determine whether the NAND industry can reach the next turning point. On the one hand, new processes such as hybrid bonding and>400 layer stacking need to mature in order to achieve a qualitative leap; On the other hand, new application bursts must emerge on the demand side in order to free NAND from the fate of marginalization. Otherwise, this former 'star storage' is likely to continue hovering on the edge of the industry.