Samsung Defends Talent and AI Norms as Foldables Go Mainstream

The September 14 morning briefing from China summarized the scramble for semiconductor talent, AI codes of conduct, and trends in foldable devices. In the report by TMTPost’s official account, multiple developments related to devices, components, and artificial intelligence were covered alongside industrial economy operations and financial statistics. Financial statistics showing August M2 up 7.5% year-on-year were also included, highlighting stabilization of manufacturing and consumption as a policy issue. For Japanese engineers and product planners, it provides material to grasp supply chains, device form factors, and operating conditions for foundation models at the same time.

Intensifying Battle for Semiconductor Talent

in South Korea

As U.S. memory companies step up hiring in South Korea, Samsung Electronics is countering with higher compensation and stock grants. According to reports, Micron is recruiting HBM design engineers in South Korea, while SanDisk is hiring in Seoul for NAND system architecture and core design, as well as solid-state drive firmware development. The background is said to be growing demand for high-bandwidth memory for AI data centers, high-performance NAND, and enterprise SSDs. HBM determines the bandwidth of AI accelerators, while NAND is directly linked to storing training materials and inference records. As a result, talent in circuit design, implementation, and control software has become key to the supply chain.

Samsung’s headcount at its South Korean headquarters in the first half fell by about 300 year-on-year to 129,200. Meanwhile, employee compensation spending rose 6.8% year-on-year to 8.01 trillion won. The company promised to grant 1.777 million shares of common stock to employees in the first half, with the stated purpose of securing key talent and boosting motivation. The fair value at the time was about 370.6 billion won, with some shares to be vested continuously through 2028. The coexistence of declining headcount and rising compensation indicates selective retention. The long-term stock vesting can be seen as a means to curb knowledge outflow from job changes.

This scramble is not merely a wage competition. Stacking and bonding for HBM and higher layer counts for NAND require accumulated process knowledge. In areas where design assets and mass-production conditions are intertwined, the move of a single skilled expert could affect yields. The move by U.S. companies to expand bases in South Korea can be assessed as aiming to shorten the distance between design and mass production. Samsung’s use of stock as a non-cash tie can also be seen as designed to encourage long-term tenure.

Aiming for Leadership in Advanced Packaging

with Organic Bridges

Samsung Electro-Mechanics and Qualcomm are reported to be jointly developing organic bridge technology for AI chips. The two companies are said to have been conducting R&D and verification for more than a year. Samsung Electro-Mechanics is also developing 2.1D packaging substrates based on organic bridges in parallel with other customers. The aim is to broaden options for advanced packaging that connects multiple chips at high density. In AI accelerators, bandwidth between compute chips and memory devices determines performance, making wiring technology on the substrate side critical.

As a technical target, the goal is reportedly to embed organic bridges with five to seven redistribution layers into FC-BGA substrates, refining line width and spacing to 1.5 to 2 micrometers. The technology was positioned as a potential alternative to Intel’s EMIB silicon bridge packaging technology. The approach using organic materials can be expected to offer advantages in cost and supply compared with silicon interposers. Stable formation of fine wiring and warpage control will be key to mass production. The fact that the verification period has exceeded one year indicates an emphasis on ensuring reliability.

The collaboration with Qualcomm can be seen as targeting both mobile and on-device AI processing and data center-side accelerators. If packaging substrates become standardized, reuse of design assets will become easier. Samsung Electro-Mechanics’ system of parallel development for multiple customers can be assessed as intending to avoid dependence on specific applications. Packaging attracts less attention than front-end miniaturization, but it determines yields and costs at the implementation stage. In future disclosures, the timing of mass production and the range of applicable products will be the focus.

Smartphone Assistants and Foldable Model Trends

On September 14, the consumer version of the Doubao smartphone assistant was officially announced. The first model to carry it is the Nubia NaviX Ultra, now accepting pre-orders and set for official release on September 16. It supports multiple activation methods, including voice and an AI key, enabling invocation suited to the usage scenario. An official video explained that compared with the technical preview version announced late last year, emphasis was placed on stability and everyday ease of use. The success of assistant functions will be decided not only by response accuracy but also by latency from activation to execution and by few false triggers.

For foldable devices, a post by noted reporter Mark Gurman drew attention. In a post on September 13 local time, he expressed the view that Apple’s first foldable iPhone Duo will push foldables into the mainstream. He mentioned the possibility of expanding the lineup over the next few years with a large-screen iPhone Duo Max, a lower-spec Duo SE, and a vertically folding Duo mini. Regarding the name, while it was internally called Ultra for a time, one reason Duo was ultimately chosen was that, like Air and Pro, its three letters make it easy to add suffixes such as Max and SE.

As for the adoption outlook, it was suggested that with falling technology costs and expanding use cases, foldables could account for more than half of new sales within the next 10 years. They are currently said to account for about 2% of global shipments. While creases, thickness, and hinge durability remain challenges, demand for document work and video viewing on large screens is clear. If larger models like the Duo Max materialize, the boundary between mobile phones and small tablets will blur further. Development of parts supply and repair systems can be seen as determining the pace of adoption.

Moves to Improve AI Safety and Codes of Conduct

AI company Anthropic is still reported to be planning an initial public offering in 2026. It is said to have selected Nasdaq as its listing venue. Although debate over AI safety, product liability, and regulation is intensifying, there is currently no change to the listing plan. The view was expressed that undergoing high-level disclosure and external oversight as a listed company would rather help improve transparency on safety. While private fundraising can cover needs for now, staying away from capital markets long-term was seen as unrealistic.

Microsoft published an interim AI model code of conduct, placing restrictions on its in-house developed models. It included prohibitions on assisting weapons manufacturing or helping obtain hazardous materials, and on promoting unhealthy eating habits, violence, and generating explicit sexual content. Models should follow users’ goals and must not hold their own goals or conceal inappropriate behavior. AI chief Mustafa Suleyman said external input will be sought before updating the code, and the new rules will guide model development from 2027 onward. The move comes as Anthropic and OpenAI executives call for slowing the pace of frontier model development.

The following remarks by OpenAI CEO Sam Altman were introduced.

The world should trust that U.S. companies developing increasingly powerful AI will act responsibly, especially now as the pace of technological progress accelerates. OpenAI welcomes the development of a federal framework for frontier AI and wants to establish unified safety requirements. Companies can begin efforts to provide such trust without waiting for antitrust exemptions or legislation. Shifting to safety-focused AI development and evaluation requires new tools. OpenAI looks forward to working with other industry participants to jointly develop common standards on misalignment, monitoring, and safety.

As a view on productivity, remarks by Nobel economics laureate Philippe Aghion were introduced.

AI is expected to lift productivity growth by about an additional 1.08 points per year over the next decade, with potential impact possibly exceeding the IT revolution. However, whether this technological potential can be converted into sustained growth depends on whether competition policy, education systems, and labor market institutions can keep pace together. 0.68 points is only a lower bound. Further considering AI’s effect in spurring new idea creation, productivity growth would be boosted by at least about 0.4 points more per year, for a total potential increase of about 1.08 points. However, technological breakthroughs do not automatically lead to sustained growth. AI certainly has great growth potential, but whether it can be fully realized also depends on whether competition issues can be handled well.

Developing norms and disclosure can be assessed as an attempt to curb accidents and misuse without slowing the pace of development competition. Incorporating external input into the process can be seen as having the effect of reducing post-deployment correction costs.

Accelerating Electrification and Consumption

Stimulus Trends

BYD Executive Vice President Li Ke expressed the view that with the spread of fast-charging technology, gasoline vehicles no longer have a future in China. He said the shift to electrification in other parts of the world may take several years, but the final outcome will be the same. He also revealed that the company plans to launch models equipped with all-solid-state batteries in 2027. For all-solid-state batteries, balancing safety and density is the challenge, and establishing mass-production conditions will determine value. Specifying 2027 as the timing can be seen as influencing investment decisions by parts companies and charging networks.

On the domestic policy front, Xin Guobin, vice minister of the Ministry of Industry and Information Technology, chaired a symposium on industrial economic operations for selected provinces in Beijing. Competent departments from 11 provinces — Hebei, Jiangsu, Zhejiang, Anhui, Fujian, Shandong, Henan, Hubei, Hunan, Guangdong, and Sichuan — participated to assess the current situation and discuss approaches to economic work for 2027. It was noted that since the start of the year the industrial economy has advanced steadily, moving toward new fields and higher added value. Policies outlined included unclogging supply-demand linkages, integrating large, medium and small enterprises, developing and promoting use of the AI industry, managing production capacity, and securing a baseline for safe development. Focus on key industries and key regions was emphasized for the fourth quarter and 2027.

On the consumption side, the Ministry of Commerce and seven other departments announced an action plan to promote consumption of household smart devices. Details of targets and support measures were not shown in this summary, but the aim can be assessed as boosting replacement demand for connected devices in homes. Wider adoption of smart home appliances presupposes unification of communication standards and maintenance systems. The simultaneous presentation of industrial policy and consumption stimulus can be seen as intending to ensure stability from both supply and demand sides. Trends in devices, semiconductors, and models will be implemented within this policy framework.

Editorial Opinion

In the short term, rising talent acquisition costs could spill over into memory and packaging costs. Samsung’s pay increases and stock grants are seen as raising hiring terms at other companies and prompting reorganization of HBM and NAND development structures. Microsoft’s interim code is assessed as becoming the premise for model development from 2027 onward, increasing review and documentation steps. The launch of the consumer version of the Doubao assistant is seen as setting a benchmark for on-device AI usability.

From a long-term perspective, leadership in packaging and memory devices is assessed as determining profit distribution across the supply chain. If organic bridges reach mass production, the cost structure of AI accelerators will change, with spillover to on-device applications. If Anthropic’s listing materializes, its level of safety disclosure is assessed as becoming the industry-wide standard. The planned 2027 introduction of all-solid-state batteries is seen as changing designs in both automotive and stationary fields.

As questions from the editors, whether disclosure and norms can coexist with development speed is at issue. Transparency through listing enhances trust, but where is the boundary with competitive secrets? Will stock compensation that restrains talent mobility lead to fixation of knowledge? In a future where foldables account for half of devices, who will handle repair and reuse systems?

References

• “【钛晨报】央行8月重要金融数据出炉，M2同比增长7.5%；美光、闪迪在韩争抢芯片人才，三星加薪并扩大股票奖励；比亚迪高管称燃油车没有未来”, by 钛媒体官方账号 — 钛媒体, 2026-09-14T23:20:00.000Z (ARR)
• Source URL: https://www.tmtpost.com/8139577.html