The fact that Vietnam’s VinFast has partnered with Israel’s Autobrains to deploy L2++ autonomous driving with AI semiconductors at its core in mass-produced EVs is an important move that shows the reality of automotive semiconductors and autonomous driving.

VINFAST AND AUTOBRAINS ANNOUNCE STRATEGIC PARTNERSHIP ON DEVELOPING AUTONOMOUS DRIVING TECHNOLOGY AND AFFORDABLE ROBO-CAR

🟧VinFast and Autobrains jointly develop L2++ autonomous driving starting from VF 8 and VF 9

VinFast announced a strategic partnership with Autobrains, an Israeli autonomous driving AI company, for the research and development of enhanced Level 2++ (L2++) autonomous driving technology.
This technology is being validated, starting with the company’s flagship SUVs, the VF 8 and VF 9. VF 8 is a mid-size electric SUV mainly for the North American and European markets, while VF 9 is a larger, high-value-added full-size electric SUV and is the core model of VinFast.

• Highlights
  of the announcement: Pilot
  testing of L2++ autonomous driving system with VF 8/VF 9 begins – “Robo-Car” concept using 7 cameras and approximately 20 TOPS of AI semiconductors
• Technical and strategic features
  : Camera-centric design
  that does not use LiDAR or high-precision HD maps, and adopts an efficiency-oriented AI SoC (System on a Chip) that does not rely on high computing performance.
• Expected market, application
  , advanced driver assistance (ADAS)
  for mass-produced EVs, and future development to Level 4 (L4) autonomous driving

🟧 Semiconductor-led design transformation against the backdrop of higher costs of autonomous driving

In the field of autonomous driving, methods equipped with a large number of LiDAR and radar and using high-performance semiconductors in the hundreds of TOPS class have been attracting attention in recent years, but rising vehicle costs and increased power consumption have become issues for mass production.
The initiative between VinFast and Autobrains is characterized by the fact that Autobrains’ unique Agentic AI (Agentic AI) aims to achieve advanced driver assistance even with relatively compact AI semiconductors of about 20 TOPS, based on the premise of efficient AI processing that executes only the necessary inferences.

This is different from NVIDIA’s ultra-high-performance SoC route, and is in the same direction as Mobileye and Qualcomm’s low-power, production-oriented ADAS semiconductor strategy.

🟧 Conclusion

The partnership between VinFast and Autobrains is an attempt to bring autonomous driving into a “realistic price range” by integrating AI semiconductors and algorithms to optimize the mass-produced EVs VF 8 and VF 9.

We will pay attention to whether this approach will spread to other EV manufacturers and predict the mainstream of automotive semiconductors and autonomous driving technology in the future.

Taiwan Semiconductor Manufacturing Co., Ltd. (TSMC) is considering a shift from the originally planned 6~7 nanometers (nm) to a more advanced “4 nanometer” process for its second plant under construction in Kikuyo Town, Kumamoto Prefecture, and it has been found that there is a change in construction trends.

TSMC mulls making more advanced chips at 2nd Japan plant to feed AI boom

🟧TSMC Kumamoto Plant 2 Planning Review and Changes in Construction Trends

Taiwan Semiconductor Manufacturing Co., Ltd. (TSMC) is building its second plant (JAIM Plant 2) in Kikuyo Town, Kumamoto Prefecture. Initially, the factory was mainly intended for manufacturing logic semiconductors with a circuit width of 6~7 nanometers, but it is reported that they are considering switching production to a more advanced “4 nanometer” process against the backdrop of a surge in demand for advanced technologies such as for AI (artificial intelligence) data centers.

• Review of production targets
  Initially, the company planned to focus on 6nm to 7nm, but it is reported that it is also considering advanced processes in the 4nm class. This is a strategic decision to respond to the growing demand for high-performance chips for AI.
• Movement at the construction site At the site of the second plant, which started construction in
  October, a large number of cranes and heavy machinery were in operation until around November, but the number has decreased sharply since then, and there is information that there will be no equipment introduction in 2026. This has spread the view that “construction is temporarily suspended”.
• Impact on the start of operation Regarding the
  scheduled start of operation in 2027, it has been pointed out that there is a possibility of extending the schedule due to design reviews and policy changes.

🟧 Why is the shift to “4 nano” attracting attention?

The nanometer (nm) of a semiconductor is a measure of the circuit width, and the smaller the number, the finer and more high-performance the process. Conventional 6~7nm processes have been widely used in smartphone SoCs and automotive semiconductors, but the 4nm class is becoming mainstream for high-performance processors for generative AI and large-scale data centers. For this reason, TSMC’s review of its plans for the second Kumamoto plant is due to changes in the market environment, with global demand for AI rapidly increasing.

There is also a view that chips for AI and high-performance computing (HPC) require high EUV (extreme ultraviolet exposure) equipment and packaging technology, which may require optimization of factory design and equipment specifications. As a result, the change of plan can be seen as a strategy to increase flexibility in the competition of manufacturing technologies.

🟧 Conclusion

TSMC Kumamoto Plant 2 is likely to be strategically redesigned from a 6-7nm plan to 4nm manufacturing, and the movement of construction sites is also changing. This reflects a global trend that responds to the growing demand for AI and data centers.

In Japan, in addition to TSMC, Micron is promoting EUV mass production of advanced DRAM in Hiroshima, and Rapidus is taking on the challenge of developing EUV-based development with next-generation logic. If multiple factories that actually handle EUV are set up and the number of engineers who experience processes, equipment, and yields increases, it is expected that semiconductor productivity and technology absorption capacity in Japan as a whole will further increase.

Denso’s new “Integrated Mobility Computer” is a strategic product that combines vehicle intelligence with energy efficiency.

Access Denied

www.denso.com

🟧 Denso to launch “integrated mobility computer” equipped with proprietary SoC in 2029

Denso has announced the next-generation in-vehicle computer, the Integrated Mobility Computer. As the “brain of the car” that controls various functions of automobiles, it aims to be put into practical use and mass production in 2029.

Equipped with a high-performance SoC for integrated control: A proprietary SoC that combines multiple semiconductors into a single chip. Achieves high computing performance while consuming low power.

Produced at an unmanned factory: Produced at the new Yoshiaki Manufacturing Plant in Nishio City, Aichi Prefecture, which is scheduled to be completed in 2027. A next-generation line with 24-hour unmanned operation has been introduced.

Application of heat treatment technology: Utilizing thermal management technology cultivated in the air conditioning field for semiconductor cooling. Aim for stable operation even in high-temperature environments.

Shinnosuke Hayashi, President of Denso, said at the briefing, “We will develop original high-performance semiconductors that can withstand harsh environments and challenge the future with the power of technology.”

🟧SDV and the flow of manufacturing in automotive semiconductors

The background is the shift to software-defined vehicles (SDVs) in the automotive industry. Designs that integrate a large number of ECUs that have been scattered throughout the vehicle and expand their functions with software updates are becoming mainstream. Such integrated control requires massive data processing and high-speed communication, and reliable automotive SoCs are key.

In Japan and abroad, Bosch and Continental, as well as Tesla and Chinese EV manufacturers, are accelerating the development of similar integrated computers. Denso’s strengths lie in its automotive reliability and thermal design technology, as well as its in-house production of its own semiconductors. This is expected to achieve both a supply chain with less external dependencies and computing performance optimized for vehicle design.

In addition, the unmanned factory to be built by the company is a “smart factory” equipped with AI-based quality inspection and an autonomous logistics system, and is a concept that supports the SDV era with both hardware development and production innovation.

🟧Summary

Denso’s “Integrated Mobility Computer” is a strategic product that aims to reshape in-vehicle computing. By combining our proprietary SoC and heat treatment technology, we pursue energy saving, high performance, and high reliability at the same time. The completion of the unmanned factory as a mass production base also symbolizes the company’s “fusion of mobility and manufacturing.”

Certainly, Denso’s move this time is a major turning point in stepping into “car brain = semiconductor”. The semiconductor business has higher added value and higher profit margins than automotive parts, but it is also an area with high development costs and risks.

Reports that Intel is considering acquiring AI semiconductor startup “Sambanova Systems” indicate that the company is trying to make a new move to make a comeback in the AI era.

🟧 Intel considers acquiring AI chip emerging “Sambanova”

It was reported that Intel is considering acquiring AI semiconductor startup SambaNova Systems. Preliminary negotiations have already begun, and the acquisition is likely to be below its valuation at the time of the 2021 financing (about $5 billion).

• Sambanova is an AI chip development company founded in 2017 by professors at Stanford University, headquartered in Palo Alto, California.
• The company’s flagship product, the SN40L, uses a unique Reconfigurable Dataflow Architecture and specializes in high-speed inference for large language models (LLMs) and generative AI.
• Unlike NVIDIA GPUs, they are characterized by significantly improving power efficiency by pre-optimizing the flow of computing and minimizing data movement.

🟧AI chip competition is back in the era of “design power”

The background of this move is the intensification of competition in the AI hardware market. Currently, much of the learning and inference of generative AI is dominated by NVIDIA GPUs, but GPUs are general-purpose designs, making it difficult to optimize for each AI model.

Intel is re-challenging the AI market with its self-developed Gaudi series, but the reality is that the gap with NVIDIA in terms of software ecosystem and performance is still large. Sambanova’s data-flowing architecture is promising in that it can “dynamically optimize” hardware specifically for AI models, and could complement Intel’s AI strategy.

In addition, since 2024, AI-specific chip companies such as Cerebras and Groq have also emerged, and the AI accelerator market is taking on a new battle for the “third option, neither CPU nor GPU”.

🟧Summary

Intel’s consideration of acquiring Sambanova symbolizes that the battle for leadership in AI chip development has entered a new phase. There is a possibility that it will open a wind hole in the composition of the GPU-strong system, and if it is realized, the power map of the AI semiconductor market will change significantly.

It will be interesting to see how Sambanova’s technology will fit into Intel’s AI strategy. While the involvement of CEO Rip Vu Tan in both companies is curious, if it is a pure integration focused on technical synergies, it could greatly boost Intel’s reinstatement in the AI field.

JASM, a subsidiary of TSMC, has signed a site agreement with Kikuyo Town, Kumamoto Prefecture for the location of its second plant, and has begun construction in earnest as a domestic mass production base for 6nm generation semiconductors.

TSMC Kumamoto Plant 2 Starts Construction on 24th Subsidiary JAIM and Kikuyo Town Sign Location Agreement|Kumamoto Daily Shimbun

🟧TSMC Kumamoto officially starts construction of its second plant and expands to mass production of 6nm, with a total investment of 3.4 trillion yen

JASM, the Japan subsidiary of Taiwan Semiconductor Manufacturing Co., Ltd. (TSMC), has signed a site agreement with Kikuyo Town, Kumamoto Prefecture for the construction of its second plant. The new plant is scheduled to start operations in December 2027 and will produce advanced 6-nanometer generation logic semiconductors. With a focus on autonomous driving, advanced automotive control, and industrial equipment applications, this move is to further strengthen the mass production system of advanced semiconductors in Japan.

• Investment: Approximately $13.9 billion (approximately 2.1 trillion yen). The Ministry of Economy, Trade and Industry supports a huge amount of subsidies.
• Production technology: Adopt 6nm process, higher performance than the first factory (12/16nm).
• Employment scale: 1,700 people will increase to a total of 3,400 people.

The second factory is located to the east of the first factory and has a construction area of about 69,000 square meters. TSMC as a whole has invested a total of 22.5 billion dollars (about 3.4 trillion yen) in the first and second factories, and the government has solidified a support policy of 1.2 trillion yen.

🟧 Why is advanced logic production in Japan expanding now?

The construction of JASM’s second factory aims to diversify geopolitical risks in the semiconductor supply chain and increase the independence of domestic production. As automobiles become more electrified and autonomous around the world, the demand for high-performance chips for automotive applications is rapidly increasing. On the other hand, as dependence on Taiwan, South Korea, and the United States continues, the Japan government is supporting TSMC in a way that encourages the “return of manufacturing bases to Japan.”

The second plant will move to a higher-value-added advanced node and aim to be able to respond to long-term market fluctuations. In terms of competition, Lapidus (Hokkaido) is also accelerating its movement towards 2nm prototypes, and the movement to revive semiconductors in Japan is in full swing.

🟧Summary

TSMC’s second plant in Kumamoto will be the most advanced manufacturing line in Japan for the 6nm generation, and it is a project that symbolizes the restructuring of the domestic supply chain with government support. The ripple effect on the local economy is also significant, and expectations are high for the accumulation of related companies and human resource development.

👉 I feel that TSMC’s expansion in Kumamoto will be the starting point for redrawing the “semiconductor map of Japan” again.

Alibaba Cloud’s new technology “Aegaeon” successfully reduced the number of GPUs required for AI inference, representing China’s AI competitiveness.

SOSP 2025: The 31st Symposium on Operating Systems Principlessigops.org

🟧Aegaeon: Alibaba’s New Technology to Reduce GPU Usage by 82%

A research team from Chinese internet giant Alibaba Group and Peking University has announced a new cloud-based technology called “Aegaeon” that efficiently manages GPU resources. The announcement was made at SOSP 2025, an international conference held in Seoul, South Korea, and has attracted the attention of the cloud industry and investors.

• Significantly reduced GPU usage: Conducted more than three months of testing on Alibaba Cloud, reducing the number of NVIDIA “H20” GPUs used from 1,192 to 213, approximately 82%.
• Visualizing LLM operating costs: Aegaeon claims in its paper that it has revealed for the first time the “invisible cost structure” of large language model (LLM) operations.
• Efficiency of AI services: Adopt a mechanism that shares GPUs on a “token-based” basis rather than on a per-model basis, and runs many models at the same time. This dramatically reduced AI inference costs.

🟧GPU shortage and US-China AI competition promoted a “computing efficiency revolution”

The background to this technological development is the shortage of GPUs due to US semiconductor export restrictions to China. Chinese companies are highly dependent on NVIDIA GPUs, and U.S. export controls have limited the supply of modern chips such as the H100 and B200. For this reason, the Chinese are focusing on “technology that maximizes performance with limited GPUs“, and Aegaeon is also positioned in this trend.

　また、アリババは自社のAIモデル「Qwen（通義千問）」シリーズを急速に拡大しており、推論処理の負荷は増大中です。GPU効率を高めるAegaeonは、単なる節約策ではなく、AI事業の持続的な競争力確保のためのインフラ戦略といえます。アリババはすでに自社製AI半導体の開発も進めており、ソフトとハード両面からAIコスト構造の再設計を進めています。

🟧まとめ

Alibaba’s “Aegaeon” has emerged as a core of China’s AI cloud strategy as a technology that fundamentally overhauls GPU efficiency in AI inference. The result of reducing GPU usage by as much as 82% is an impact that cloud providers cannot ignore.

💬Aegaeon is not a cloud optimization, but a scheduling technology that devises “how to distribute and share AI calculations”. Inference has great effects, but learning has many communication and synchronization constraints, making it difficult to apply. In that regard, NVIDIA’s dominance is likely to continue for some time.

As automotive semiconductors become increasingly complex, major players such as GlobalFoundries and Infineon are participating in the chiplet development program led by Belgian research institute imec to realize new architectures for the software-defined vehicle era.

First foundry and suppliers join Automotive Chiplet Program | imec

🟧 imec leads the new trend of automotive semiconductors, chiplets

imec, a leading Belgian research institute, has announced that it has welcomed GlobalFoundries (GF) as a foundry partner for its Automotive Chiplet Program (ACP). In addition, companies in the semiconductor and autonomous driving fields such as Infineon, Silicon Box, STATS ChipPAC, and TIER IV have also announced their participation, accelerating the new design trend of automotive semiconductors.

• Key takeaways: GF will leverage its manufacturing sites and advanced processes in the U.S., Europe, and Asia to support the development and mass production of automotive chiplets.
• Technical features: The chiplet architecture is a structure that combines small chips divided by function, making it scalable, flexible, and cost-effective.
• Expected markets and applications: The main focus is on high-performance computing systems installed in next-generation software-defined vehicles (SDVs), autonomous driving, ADAS, and in-vehicle infotainment.

Although this ACP is in the basic research stage, it aims to establish an architecture and packaging technology that meets the “safety, reliability, and durability” of automotive-grade applications, and is attracting attention from both the automotive and semiconductor industries.

🟧 Is it a chiplet now? —The restructuring required by the SDV era

In recent years, the automotive industry has undergone a significant shift from “hardware-centric” to “software-centric.” With the spread of electrification and autonomous driving, in-car computers are now required to perform enormous amounts of processing, such as AI calculations, sensor integration, and network control. However, conventional monolithic chips (large-scale SoCs with a single component) have apparent limitations in terms of development costs, heat generation, design time, and reliability.

That’s where the chiplet system came in. Functions such as compute, communication, memory, and power control can be manufactured as independent chips and connected in a package, allowing for the flexibility to combine optimal processes. This increases development speed and reusability, making it easier to adapt to the evolution cycle of automotive systems.

🟧Summary

It marks a turning point in the evolution of automotive semiconductors from a “single chip” to a “chiplet configuration”. With manufacturing and design leaders such as GlobalFoundries and Infineon coming together on imec’s research platform, the development of semiconductors at the heart of next-generation vehicles has begun to move on a global scale.

Going forward, the focus will be on which automakers will adopt this new chiplet platform and apply it to which areas (ADAS, AI control, and in-vehicle communications).

The Dutch government’s control of Chinese semiconductor manufacturer Nexperia for security reasons has affected the supply network of European automakers such as BMW.

Minister of Economic Affairs invokes Goods Availability Act

🟧 Dutch government’s intervention in Nexperia affects the automobile supply network

The Dutch Ministry of Economy took control of Nexperia under the emergency law to prevent technology leakage. Harmful decisions can be stopped, while production will continue. Nexperia was acquired by Wing Tech after the spin-off from NXP and supplies basic semiconductors such as transistors and diodes. BMW said that some supply networks were affected and that it would maintain production while assessing risks.

🟧 Economic security and US-China friction

This intervention was carried out in the context of European economic security and technological friction between the United States and China. The Extraordinary Court found that Nexperia’s management was flawed and threatened with technology leakage, and the government applied a law that allowed companies to intervene.

Wingtech is subject to U.S. export controls, and China has banned the export of certain Nexperia products, and regulations in both the United States and China have increased the supply risk of basic semiconductors.

🟧Summary

Geopolitical risks surrounding the basic components of semiconductors are affecting the European automotive industry. Disruptions in the supply of discrete semiconductors could lead to production halts, forcing companies to urgently diversify their supply networks and manage risk.

With government intervention and restrictions expected to continue due to US-China friction, efforts to strengthen supply chains are likely to accelerate.

Oracle has adopted AMD’s latest AI GPU “MI450” on a large scale and has taken steps to diversify its AI infrastructure. This is a move to challenge NVIDIA’s strong system in both cloud and AI development.

Oracle and AMD Expand Partnership to Help Customers Achieve Next-Generation AI Scale

www.amd.com

🟧 Oracle to Adopt 50,000 MI450 AMD’s Next-Generation AI Semiconductors

Oracle announced that it will introduce AMD (Advanced Micro Devices) next-generation AI semiconductor “MI450” in data centers on a scale of 50,000 units from the second half of 2026. The announcement was made at Oracle CloudWorld, a technology event held in Las Vegas, and was a symbolic move for cloud strategy in the AI era.

• Oracle will adopt AMD’s MI450 GPU in its cloud (OCI) and make it available to customers as AI computing resources.
• The MI450 adopts AMD’s new “CDNA4” architecture, which is optimized for large-scale AI models with high bandwidth with HBM3E memory and support for FP4/FP8 operations.
• The company plans to expand its introduction after 2027, aiming to streamline everything from AI training to inference on the cloud.

🟧 Strategies to challenge NVIDIA’s dominant market — Diversification of AI semiconductors accelerates

In the AI semiconductor market, NVIDIA still has an overwhelming market share. GPUs such as the H100 and B200, which support generative AI and large language models (LLMs), are always in short supply, and cloud providers and AI development companies are struggling to procure them. The oracle’s movement is the first step in changing this structure.

AMD’s MI450 is said to be competitive in both computing efficiency and cost performance, and is attracting attention as an “alternative to NVIDIA”. In addition, AMD uses a rack-scale integrated design called “Helios” to provide optimized data center configurations in combination with EPYC CPUs and Pensando DPUs. Incorporating this design into Oracle’s cloud infrastructure simultaneously improves AI compute performance and reduces operational costs.

On the other hand, OpenAI, which provides AI platforms, has also revealed plans to procure additional AMD GPUs, and the relationship between the two companies is deepening through a large contract with Oracle (5 years and $300 billion in cloud use).

🟧Summary

Oracle’s decision to adopt AMD’s MI450 marks a turning point in its AI cloud strategy. For AMD, it was also a major achievement for the first time that a major cloud operator adopted next-generation GPUs.

💬 Intel and Originally, it would have been a competitor to AMD and NVIDIA with “Falcon Shores” and “Gaudi 3”, but the delay in development schedule and the lack of maturity of the software infrastructure are resonating.

Intel has started production of semiconductors for personal computers using the latest 2-nanometer-class “18A” process at its new plant in Arizona. As an advanced process originating in the United States, Intel aims to return to technology-driven power.

🟧 Intel Starts Production of Semiconductors for PCs in 18A Process

Intel has officially started semiconductor production using a new 2nm-class process “Intel 18A” at its state-of-the-art factory “Fab 52” in Chandler, Arizona, USA.
The first product is “Intel Core Ultra Series 3 (development code name: Panther Lake)“, which is scheduled to be mass-produced from the end of 2025.

• A new transistor structure “RibbonFET” is adopted. The gate all-around structure improves current controllability and switching efficiency, significantly reducing leakage current.
• The world’s first introduction of back-side power supply technology “PowerVia”. By supplying power from the back of the chip, it isolates the signal wiring and reduces noise and power loss.
• Performance and efficiency: Compared to the previous generation “Intel 3”, the performance/power consumption ratio has been improved by approximately 15%, and the chip density has also increased by 30%.

The 18A generation adopts an XPU configuration that integrates CPU, GPU, and AI accelerator, and supports a variety of processing, including AI calculations, on a single chip. In addition to PC applications, the next-generation server Xeon 6+ (Clearwater Forest) will also be produced at the plant.

🟧 Why Intel is rushing mass production at its Arizona plant

In the background, there is a national strategy to regain semiconductor manufacturing sovereignty to the United States. While Taiwan’s TSMC and South Korea’s Samsung plan to mass-produce 2nm, Intel aims to be the only company that can manufacture state-of-the-art nodes in the United States.

In addition, Intel has been slow to acquire customers in recent years in its foundry business (manufacturing chips designed by other companies). The launch of Fab 52 in Arizona is a step forward in the fight against AI demand and the cloud market. While the company is delaying the construction of its plant in Ohio, it is concentrating its resources on Arizona and prioritizing the establishment of mass production technology and yield improvements centered around the 18A.

In the area of manufacturing know-how and quality stability, where TSMC has an advantage, Intel is poised to leverage its advanced packaging technology “Foveros” to achieve both flexibility and high performance through 3D stacking of multiple chiplets.

🟧Summary

Intel’s “18A” process is the first 2-nanometer node to be developed and manufactured in the United States, and is a technology that symbolizes the geopolitical restructuring of the semiconductor industry.
With Fab 52 fully operational, U.S.-origin advanced manufacturing will once again have a strong presence in a wide range of fields, from AI PCs to cloud and robotics.

In the future, it is expected that AI-specialized chips using 18A technology and contract manufacturing projects will expand. 2026 is likely to be the year to measure how far Intel can approach the “miniaturization competition” with TSMC.