According to a report from Liberty Times, citing The Chosun Daily, Samsung Electronics has announced a significant personnel adjustment following its November 27 presidential-level appointments, reportedly cutting around 100 senior executives from its semiconductor (DS) division—an unprecedented reduction of senior leadership in the company’s history, as the report notes.

The report highlights growing concern that these departing executives might join Chinese semiconductor companies, increasing the risk of technology leaks. Chinese semiconductor firms are reportedly offering salaries three to four times higher as part of aggressive recruitment efforts targeting employees from Samsung Electronics and SK Hynix, as the report indicates.

Additionally, the report points out that many of the executives being laid off are in their 50s and 60s. This has raised particular concerns about intellectual property leaks, as they are likely to seek new employment opportunities.

The report, citing sources, also reveals that Chinese semiconductor companies have historically contacted South Korean tech talent discreetly through intermediaries but are now openly offering salaries three to four times higher to attract employees. Sources indicate that these Chinese companies are conducting large-scale recruitment, often hiring through foreign entities to disguise their Chinese affiliations.

To address this issue, efforts to retain experienced professionals in South Korea are ongoing, as the report highlights. Ahn Ki-hyun, Executive Director of the Korea Semiconductor Industry Association, outlined initiatives such as creating roles for patent examiners and strengthening collaborations between academia and industry. However, despite these measures, Ahn acknowledged that completely preventing engineers from seeking opportunities abroad—especially in China—remains a significant challenge.

Read more

• [News] Samsung Reshuffles Leadership: New Semiconductor Chief, Foundry CTO Role Added
• [News] Samsung Advances Year-End Reshuffle to Tackle Crisis, Device Solutions Division Hit Hardest

(Photo credit: Samsung)

While NVIDIA may be better known for its latest advancements in AI accelerators such as Blackwell, the tech giant’s automotive chips are expected to seen explosive growth in 2025, with TSMC and IC design house MediaTek poised to seize this business opportunity, according to a report from Commercial Times.

China’s SAIC Zhiji and Geely among the First Adopters

According to Commercial Times, China’s high-end smart electric vehicle brands, such as SAIC Zhiji and Geely’s Zeekr, will be among the first clients to adopt NVIDIA’s latest Drive Thor chip. On the other hand, major car manufacturers in China, including BYD and Li Auto, also plan to fully transition to NVIDIA’s Thor platform by the end of 2025 or early 2026, as per the report.

NVIDIA’s Thor platform, according to the report, is currently known to be available in five versions based on computing power. It uses the Blackwell GPU architecture, with a significant performance boost (Tops) compared to the previous generation Orin.

Notably, NVIDIA is advancing its entire product line to be manufactured with next-gen processes, according to Commercial Times. In addition to the upcoming 5090 GPU, which will be built on TSMC’s 4nm process and set to launch early next year, NVIDIA’s Thor is planned to be migrated to 4nm from 7nm, the report suggests.

MediaTek’s Pairing Chips to Debut by 2026

Citing sources from the supply chain, the report notes that Taiwanese IC design house MediaTek plans to launch smart cockpit chips that pairs with NVIDIA’s Thor platform, which are likely to begin appearing in certain car models by 2026.

It is worth noting that MediaTek is reportedly preparing to reserve TSMC’s CoWoS-R capacity by the end of 2025 specifically for the upcoming automotive SoCs, according to Commercial Times.

The report further notes that MediaTek’s smart cockpit chips are likely to come in four variants: CX-1, CY-1, CM-1, and CV-1. The CX-1 is rumored to be built on the 3nm process.

It is also noteworthy that NVIDIA not only collaborates with MediaTek on automotive chips. The two companies are reportedly teaming up, targeting to launch their AI PC chip in the first half of 2025, as per Chinese media outlet MyDrivers.

Read more

• [News] NVIDIA Reportedly to Launch RTX 5090D Specifically for China, Bypassing U.S. Restrictions
• [News] NVIDIA Q3 Data Center Revenue Hits Record, as Jensen Huang Denies Blackwell Overheating Rumors

(Photo credit: NVIDIA)

According to a report from UDN News, U.S. President-elect Donald Trump’s proposed 25% tariff on imports from Mexico has sent shockwaves through the server manufacturing industry. The report highlights that Taiwanese server ODMs are swiftly implementing contingency measures, with major players like Foxconn and Quanta increasing their investments in the U.S. Meanwhile, manufacturers such as Wistron, Wiwynn, and Inventec are actively preparing similar strategic responses.

The report, citing industry sources, reveals that Trump’s tariff proposal has led global cloud service providers (CSPs) and server brands to demand alternative production plans from Taiwanese manufacturers. These alternatives emphasize capabilities outside of Mexico, with a preference for “Made in the USA” options. This urgency has driven manufacturers to activate their “Plan B” strategies, intensifying investments in U.S. operations.

Previously, many leading manufacturers had established facilities in Mexico to efficiently serve North American clients. However, the report notes that Trump’s tariff proposal has placed AI server manufacturers on high alert, as Mexico is currently the largest exporter of servers to the U.S. If the tariffs are implemented, the AI server supply chain would face significant disruption.

Foxconn: New Factory Plan in the U.S.

According to the report, Foxconn, a leading player in the AI server manufacturing sector, has responded swiftly by approving a new factory plan in the U.S. The company is set to invest USD 33.03 million to acquire land and facilities in Houston, Texas, to manufacture AI servers.

Foxconn is carefully monitoring potential changes in U.S. trade policy under Trump to assess the need for further expansion in the U.S. The report cites Foxconn Chairman Young Liu, who stated that while clients might need to relocate production, Foxconn’s globally advanced and flexible layout positions the company to handle such shifts more effectively than its competitors.

Quanta: Expanding Operations Across the U.S.

As for Quanta’s accelerated efforts to expand in the U.S., the report notes that the company’s board has approved an additional USD 230 million investment in its Tennessee subsidiary, QMN, bringing Quanta’s total U.S. investment this year to USD 1.23 billion.

Quanta’s Executive Vice President Mike Yang disclosed that the company has production facilities on both the East and West Coasts of the U.S., ranging from L6 (motherboard assembly) to L10 (server assembly and testing), as the report indicates. This diversified capacity allows Quanta to address the varying needs of its customers across regions, and the report notes that the company intends to maintain its focus on U.S. investments.

Wiwynn and Inventec: Flexible Contingency Plans

Meanwhile, Wiwynn and Inventec are closely monitoring policy developments while maintaining close communication with clients and preparing flexible contingency measures. The report cites Inventec’s General Manager, Jack Tsai, who highlighted that the company operates facilities in various global locations. If the U.S. imposes high tariffs on Mexican imports, Inventec is prepared to shift production to its factories in Thailand or Taiwan, according to the report.

Read more

• [News] Trump’s Tariffs Likely to Hit Taiwan’s AI Server Supply Chain, with Foxconn among the Impacted
• [News] Taiwanese AI Supply Chain Expands in Southeast Asia Amid Investments from CSPs

(Photo credit: Foxconn)

As demand for artificial intelligence continues to surge, the industry’s appetite for high-performance computing chips has grown increasingly intense. According to a report by the Commercial Times, NVIDIA’s next-generation Blackwell architecture chip, the GB200, has encountered new technical hurdles in its mass production plans. In response, CSP provider Microsoft is reportedly scaling back its orders.

Sources within the supply chain cited by Commercial Times reveal that the issue lies in the backplane connection design. The testing yield for cartridge connectors provided by U.S. Tier-1 supplier Amphenol has been suboptimal, potentially delaying mass production until March 2025.

The GB200 chips employ TSMC’s cutting-edge CoWoS-L advanced packaging technology, incorporating a highly complex cabinet design. However, this complexity has led to various challenges, including overheating in chip design, leakage issues in UQDs, and now, insufficient yield rates for copper cables. While NVIDIA announced during its recent earnings call that Blackwell production is fully underway, supply constraints remain a pressing issue that the company is working to resolve with its partners.

The same report, citing supply chain sources, attributes the issue to a newly developed cartridge connector module. The significant specification upgrade of the GB200 has increased production complexity, resulting in poor yield rates and failed testing, creating a major bottleneck.

NVIDIA is actively seeking alternative suppliers, but issues such as patent restrictions and capacity ramp-up delays are expected to prolong resolution efforts. While the report notes that chip production schedules remain unaffected, supply chain checks indicate that Microsoft has already cut its orders for NVIDIA by 40%, reallocating some to the GB300 chips set for release in mid-2025.

(Photo credit: NVIDIA)

When the names of Samsung and SK hynix have been brought up together, their long-standing rivalry in the HBM market may be the first thing that comes to mind. However, the two memory giants now might be surprisingly forming an alliance, as they are joining forces to standardize LPDDR6-PIM products amid the AI boom, as per Business Korea.

According to the report, the growing adoption of on-device AI technology is driving the push toward commercializing PIM memory. On-device AI performs computations directly within devices such as smartphones, offering advantages over cloud-based AI services like ChatGPT, the report notes.

Collaboration on Standardization with JEDEC

The collaboration between Samsung and SK hynix on “Low Power Double Data Rate 6 (LPDDR6)-Processing In Memory (PIM)” products reportedly involves the preparation to register the standardization with the Joint Electron Device Engineering Council (JEDEC), according to Business Korea.

Citing a Samsung Electronics representative, the report indicates that the two companies are currently formulating plans for the standardization timeline.

It is worth noting that both companies have introduced PIM-related products, though their commercialization has seen limited success, the report suggests. Samsung introduced the industry’s first HBM-PIM (Aquabolt-XL) in 2021, which incorporates the AI processing function into Samsung’s HBM2 Aquabolt, according to its press release. The company also launched LPDDR5 products with PIM capabilities and pursued standardization efforts.

(credit: Samsung)

On the other hand, SK hynix launched Graphics DRAM (GDDR6)-PIM in 2022, as it plans to introduce a technology that combines GDDR6-AiM with AI chips in collaboration with AI chip company Sapeon, according to its press release.

(credit: SK hynix)

However, developing products based on their respective standards makes it challenging for the industry to adopt a unified standard, Business Korea indicates. This has prompted the two companies to collaborate on standardization efforts for PIM products prior to their commercial release, the report states.

According to Business Korea, unlike traditional memory, LPDDR with PIM requires unique technical considerations, such as “Internal Bandwidth,” which measures bandwidth within the memory itself, in contrast to the “External Bandwidth” between the processor and memory that is critical for conventional memory.

Ongoing discussions are said to be focused on determining the appropriate specifications for each standardization item, the report adds.

Read more

• [News] Samsung and SK hynix Face Uncertainties amid Trump’s Potential CHIPS Act Subsidy Review

(Photo credit: Samsung)