According to the latest memory spot price trends released by TrendForce, overall DRAM spot market demand has not further heated up, and transaction volume has further shrunk. NAND Flash, affected by pessimistic demand, is experiencing a lack of enthusiasm in market price inquiry transactions. For more details:

DRAM Spot Market:

In the spot market, demand has not risen further for chips, and the overall transaction volume continues to shrink. Most module houses have been slow to reduce inventory, so the situation with spot trading has not been ideal. However, procurement momentum is relatively healthy for server DRAM RDIMMs mainly because DRAM suppliers have significantly increased contract prices for these products, thereby causing some increase in demand in the spot market.

Overall, various DRAM products are currently not consistent in terms of price increases and declines. Further observations on the demand situations across different application segments are necessary to determine their price trends. The average spot price of mainstream chips (i.e., DDR4 1Gx8 2666MT/s) rose by 0.21% from US$1.945 last week to US$1.949 this week.

NAND Flash Spot Market:

Price inquiries have not been robust under a poor outlook on demand as the spot market has started manifesting room for negotiations with channel traders, or even module houses, experiencing an excessive inventory, alongside the obstruction on a further hike of 3D wafer prices. Spot prices of 512Gb TLC wafers have dropped by 0.03% this week, arriving at US$3.764.

According to a Reuters, despite the U.S. expanding export controls on advanced artificial intelligence (AI) chips to China last year, Chinese universities and research institutions have recently acquired high-end AI chips from Nvidia through distributors.

Reviewing hundreds of bidding documents, Reuters found that since the U.S. expanded chip export controls on November 17 last year, ten Chinese entities have acquired Nvidia’s advanced chips embedded in server products produced by U.S. firms Supermicro, Dell, and Taiwanese company Gigabyte Technology.

Based on this Reuters report, bidding documents not reported from November 20 last year to February 28 this year show that Chinese institutions such as the Chinese Academy of Sciences, Shandong Artificial Intelligence Institute, Hubei Earthquake Administration, Shandong University, Southwest University, a technology investment company owned by the Heilongjiang Provincial Government, a state-owned aerospace research center, and a space science center have purchased these server products from distributors, which include some of Nvidia’s most advanced chips.

In response, a Nvidia spokesperson told Reuters that the products involved in these bids were exported before the ban was implemented in the United States. The spokesperson stated that the report does not imply that Nvidia or any of its partners violated export control regulations, and the proportion of these products in global sales is negligible. Nvidia complies with U.S. regulatory standards.

Both Supermicro and Dell stated that they would investigate and take action if any third-party illegal exports or re-exports are found. Gigabyte, the Taiwanese company mentioned in the report, told the Central News Agency that it has fully complied with relevant regulations since the chip ban took effect on November 17 last year, and has not shipped any restricted products to China. Gigabyte reiterated its strict adherence to relevant Taiwanese laws and international embargo regulations, stating that there has been no violation of any embargo regulations.

In 2023, the United States further restricted Chinese businesses from acquiring high-end AI chips. At that time, Nvidia responded by launching a China-specific version, the H20. TrendForce also presented relevant data for the Chinese market, indicating that Chinese CSP companies, including ByteDance, Baidu, Alibaba, and Tencent (BBAT), accounted for approximately 6.3% of high-end AI server shipments in 2023. Considering the ban and subsequent risks, it is estimated that the proportion in 2024 may be less than 4%.

(Photo credit: NVIDIA)

As AI semiconductor competition intensifies, the wafer foundry industry faces new challenges due to stagnant demand and excess capacity. The battle for dominance in the high-bandwidth memory (HBM) market is also escalating.

According to Business Korea, Samsung has extended the operation time of its Taylor plant from the end of 2024 to 2026, possibly adjusting the investment pace in consideration of the foundry market.

Regarding the global wafer foundry industry outlook for this year, TSMC President C.C. Wei stated that the growth of the global foundry industry this year has been revised from the previous earnings call of 20% to mid-teens.

Currently, TSMC’s two fabs in Arizona, USA, are scheduled to commence production in 2025 and 2028, respectively; the Kumamoto fab in Japan has started operations in February, and the second fab will start production before 2027. Intel, on the other hand, plans to establish new foundries in the United States, Europe, and Israel. The activation of these new fabs has raised concerns in the market about oversupply issues.

Contrastingly, in the HBM market, crucial for AI chips, SK Hynix and TSMC have formed an alliance, intensifying the competition between this alliance and Samsung.

SK Hynix announced on April 19th that the company has recently signed a memorandum of understanding with TSMC for collaboration to produce next-generation HBM and enhance logic and HBM integration through advanced packaging technology. The company plans to proceed with the development of HBM4, or the sixth generation of the HBM family, slated to be mass produced from 2026, through this initiative.

Looking at Samsung’s developments in the HBM, Samsung Electronics successfully developed the industry’s first highest-capacity 12-layer HBM3E in February, attempting to regain market dominance. In the second quarter, along with the 8-layer product, it will supply to Nvidia. The next goal is to launch the 16-layer HBM4.

Per TrendForce’s data, the three major HBM manufacturers held market shares are as follows: In 2023, SK Hynix and Samsung each held around 47.5%, while Micron’s share was roughly 5%. Still, forecasts indicate that SK Hynix’s market share in 2024 will increase to 52.5%, while Samsung’s will decrease to 42.4%.

In line with the same report from Business Korea, despite the decline in foundry demand, Samsung Electronics and SK Hynix’s profit prospects are expected to improve compared to last year.

As per estimates cited in the report from investment institutions, SK Hynix’s first-quarter revenue is expected to reach 12.1021 trillion won, with an operating profit of 1.7654 trillion won, and the operating profit for the entire year of 2024 is expected to exceed 21 trillion won; Samsung Electronics’ DS division’s performance is improving, with the first-quarter operating profit expected to be between 700 billion and 1.8 trillion won, and the overall operating profit for 2024 is expected to be around 35 trillion won.

(Photo credit: SK Hynix)

Recently, according to sources from an official platform of Shanghai Construction No.4 (Group), Huahong Group has topped out the main building of FAB9 of the Huahong Manufacturing (Wuxi) project. This is reportedly the second phase project with a total construction area of about 530,000㎡, which is projected to construct a 12-inch characteristic process production line with a monthly capacity of 83,000 wafers.

It can be seen that the 12-inch wafer field is greeted by high enthusiasm. Previously, companies such as Renesas Electronics, PSMC, TSMC, and UMC have all announced plans to build new 12-inch fabs. As for China, as per earlier industry reports, 12-inch fabs scheduled to start production in 2024 include CR Micro, ZenSemi, and CanSemi, all located in Guangdong Province.

• Investment Enthusiasm in 12-Inch Fabs Unabated

On April 11, Renesas officially restarted its previously closed factory in Kofu. It was announced in 2022 that Renesas would invest JPY 90 billion to convert the plant into a 12-inch fab in a bid to meet the increasing demand in the power semiconductor sector.

The fab has a cleanroom with an area of 18,000㎡, and it will start mass production of IGBTs, power MOSFETs, and other power devices in 2025, which is expected to double Renesas’ overall power semiconductor capacity.

On March 13, PSMC and India-based Tata Group held a groundbreaking ceremony for their joint 12-inch fab. The fab comes with a total investment of INR 910 billion (Around USD 11 billion), which is estimated to produce a monthly capacity of 50,000 wafers, covering various mature nodes such as 28nm, 40nm, 55nm, 90nm, and 110nm.

On February 24, TSMC’s Japan Kumamoto fab (JASM) was officially opened, marking TSMC’s first plant (Fab23) in Japan. TrendForce stated that the plant will possess a total capacity of up to 40~50Kwpm in the future, with the process mainly focusing on 22/28nm and a small amount of 12/16nm. This will pave the way for developing the main process of the Kumamoto Fab2 later.

Previously, TSMC had announced that in response to customer demand growth, construction of JASM’s second fab in Japan is planned to commence at the end of 2024 and start operation in late 2027. Media reports stated that TSMC would invest JPY 2 trillion in the second fab in Kumamoto, which will adopt advanced processes of 6nm and 7nm. The monthly total capacity of 12-inch wafers in JASM Kumamoto fab is expected to exceed 100,000 pieces in the future.

In January, it was reported that UMC’s new fab in Singapore is scheduled to complete construction by mid-2024 and start mass production in early 2025. UMC said that to meet the demand for capacity construction, its board of directors approved a capital budget execution proposal of USD 39.8 million. The first phase of the new fab, with a total investment of USD 5 billion, is expected to deliver a monthly capacity of 30,000 wafers, providing 22/28nm processes.

UMC has been operating the 12-inch fab in Singapore for over 20 years. In February 2022, UMC’s board of directors approved the plan to expand a new advanced fab in the Fab12i area in Singapore. At that time, UMC expected the new fab to start mass production in late 2024, but the latest news indicates that the date of mass production will be in early 2025.

• Several Guangdong-Based 12-Inch Fabs Set to Start Production in 2024

According to a local official report from Zengcheng, Guangdong, ZenSemi held a lithography machine introduction ceremony for the project of 12-inch advanced intelligent sensors and characteristic process wafers mass production lines, marking that the project has smoothly entered the debugging and production preparation phase.

It is reported that the first phase of the project covers an area of 370 acres, with a planned investment of CNY 37 billion. It is expected to start production in June 2024, with the first batch of high-yield products scheduled to be completed in late December and delivered to customers.

According to official information from CanSemi, the third phase of CanSemi’s project will establish a 12-inch integrated circuit analog characteristic process production line with a capacity of 40,000 wafers per month. Currently, the first and second phases have been put into production successively, and the company are accelerating the construction of the third phase project, striving to achieve a fixed asset investment of over CNY 4 billion in 2024 and ensure the third phase to complete construction and start production in 2024.

According to the Wechat Account “Binhai Baoan”, Huahong’ 12-inch characteristic process integrated circuit production line project is also expected to start production this year.

It is reported that the first phase of Huahong’ 12-inch power chip production line project has a total investment of CNY  22 billion, with a total construction area of 238,000㎡, and an annual production capacity of 480,000 wafers after completion. The products will mainly be used in automotive electronics, new energy, industrial control, consumer electronics, and other fields.

As per stats from Wechat Account “Global Semiconductor Observation”, there are currently 31 operating 12-inch fabs in China (Inclusive of under-construction 12-inch fixed-capacity fabs), with a total monthly capacity of approximately 1.189 million wafers.

Compared with the planned monthly capacity of 2.17 million wafers, the capacity utilization rate of these fabs is close to 54.48%, indicating significant expansion potential. Considering construction and future planning, it is estimated that China will add 24 new 12-inch fabs in the next five years, with a planned monthly capacity of 2.223 million wafers.

Assuming all planned 12-inch foundries achieve full production, the total monthly capacity of 12-inch wafers in China will exceed 4.14 million wafers in late 2026, representing a 248.19% increase in capacity utilization compared to the present.

Read more

• [News] SK Hynix Rumored to Propose Collaboration with Kioxia for HBM Production in Japan
• NAND Flash Industry Revenue Grows 24.5% in Q4 2023, Expected to Increase Another 20% in Q1, Says TrendForce

(Photo credit: TSMC)

As the Apple Worldwide Developers Conference (WWDC) in June approaches, recent rumors about Apple’s AI research have resurfaced. According to reports from MacRumors and Tom’s Guide, Apple is reportedly developing a large language model (LLM) comparable to ChatGPT that can run directly on devices without relying on cloud platforms.

In late February of this year, Apple reportedly decided to terminate its electric car development project “Project Titan” initiated a decade ago and redirected research funds and resources into the field of generative AI. This move has drawn significant attention to Apple’s activities in the AI sector.

Moreover, MacRumors also reports that Apple’s AI research team, led by John Giannandrea, began developing a conversational AI software, known today as a large language model, four years ago. It is understood that Apple’s proprietary large language model has been trained with over 200 billion parameters, making it more powerful than ChatGPT 3.5.

Previously, Apple disclosed that the iOS 18 operating system, set to launch this year, will incorporate AI capabilities. Recently, tech website Tom’s Guide speculated further that iOS 18 could execute large language models directly on Apple devices. However, whether Apple’s large language model can be successfully integrated into various Apple software services remains to be seen.

Using Apple’s voice assistant Siri as an example, at an AI summit held by Apple in February last year, employees were informed that Siri would integrate a large language model in the future. However, former Siri engineer John Burkey revealed to The New York Times that Siri’s programming is quite complex, requiring six weeks to rebuild the database for each new sentence added.

On the other hand, amid Apple’s AI research facing challenges, interest in its Vision Pro headset device has also begun to wane, with recent sales cooling rapidly. As per a report by Mark Gurman from Bloomberg, he has indicated that demands for Vision Pro demos are way down at Apple stores, and sales of Vision Pro at some stores have dropped from a few units per day to a few units per week.

Read more

• [News] Apple Reportedly Releases Supply Chain List, Nanya Technology, Lotes, and Novatek Not Included
• [News] Apple Reportedly Plans to Team Up with Baidu, Sparking AI Server Deployment Trend

(Photo credit: Apple)