December 2025 marks the end of an unprecedented and turbulent year for the semiconductor and AI industry. The “supercycle” triggered by the generative AI boom has gone beyond mere supply and demand waves and has evolved into a struggle for supremacy among nations and a shift in technology paradigms. From the vast amount of news observed this month, we have extracted the 10 most important topics that will influence the industry in the future and explain them in detail.
- NVIDIA H200 Export Approval to China and the “Pax Silica” Attack
The U.S. government has conditionally approved the export of NVIDIA’s H200 high-performance AI chip to China. This is expected to be subject to a 25% fee (tariff) and is seen as part of a new technology trade strategy under the Trump administration. On the other hand, some in the U.S. Congress are concerned about China’s growing AI capabilities and have called for a 30-month ban on exports of advanced chips, including the H200 and Blackwell, revealing a lack of alignment within the administration. In response, the Chinese side has accelerated its “self-reliance and self-reinforcement” movement by adding domestically produced chips from Huawei, Cambricon (Cambricon), and others to its procurement list; some believe that the maturation of Huawei’s Ascend 910C and other products has prompted the relaxation of US export controls. NVIDIA itself has also been forced to strengthen its compliance efforts, including the introduction of chip tracking technology, following the detection of smuggling networks through Southeast Asia. The struggle for technological supremacy between the U.S. and China has shifted from a mere blockade to a more complex management competition involving tariffs and tracking technology, and a new geopolitical order centered on semiconductors, known as “Pax Silica” (peace through silicon), is being sought to be formed. - The arrival of the “memory supercycle” and the ripple effects of rising prices
The explosion in demand for AI servers has brought the memory market into a powerful boom period known as the “supercycle.” The supply shortage of HBM (High-Bandwidth Memory) is particularly severe, with Micron revealing that its HBM supply quota for 2026 has already been sold out. This impact is spreading to general-purpose DRAM and NAND flash, and DRAM prices are expected to skyrocket toward 2026. Samsung and SK Hynix are planning significant price increases ahead of the transition to the next generation of HBM3E and HBM4. This cost increase is beginning to be passed on to end products, with PC makers such as Dell and Lenovo signaling product price increases starting in early 2026. In addition, rising memory costs are expected to weigh on shipments in the smartphone market as well, creating a divide between companies benefiting from AI and hardware makers struggling with rising costs, The company’s decision to scale back and withdraw its consumer brand Crucial and focus its resources on the enterprise is emblematic of this dramatic shift in market structure. - TSMC Achieves 2nm Process and Stands by “N-2” Rule
TSMC, the world’s largest foundry, achieved its cutting-edge 2nm process technology milestone on schedule, despite geopolitical and natural disaster risks such as military exercises by China and the M7.0 earthquake in Taiwan. While preparations for volume production at the Kaohsiung plant are progressing, the Taiwanese government’s “N-2” rule (overseas plants must remain two generations ahead of the home country in technology) was reaffirmed for overseas expansion. This may restrict TSMC’s Arizona plant from providing cutting-edge technology to customers such as Apple and NVIDIA who wish to manufacture cutting-edge chips in the US. In addition, TSMC is reportedly considering converting its second Kumamoto plant from its original plan (6nm/7nm) to a more advanced 4nm process in response to surging demand for AI. However, the Arizona plant in the U.S. is facing bureaucratic procedural hurdles, highlighting the reality that decentralization of global manufacturing sites is not a straightforward process. With its overwhelming technological advantage, TSMC’s dominance in the foundry market is unshakable, with plans to raise prices for its leading-edge processes between 2026 and 2029. - Samsung Strikes Back: 2nm “Exynos 2600” and HBM4 onslaught
Samsung Electronics, which had reportedly been struggling in both foundry and memory, showed signs of a turnaround toward the end of the year. The company unveiled the specifications of its Exynos 2600 mobile processor, the industry’s first to use the 2nm Gate-All-Around (GAA) process, and showed that it is ready for mass production. The chip offers a 113% improvement in AI performance over the previous generation and is expected to be used in the next Galaxy series. In the memory area, the company is poised to challenge SK Hynix’s solo run, having reportedly outperformed the competition in NVIDIA’s tests of its next-generation “HBM4”. Furthermore, SK Hynix is regaining technological credibility after a “redesign,” with a major contract with AI inference chip startup Groq and the possibility of acquiring a share of HBM supply from NVIDIA. The company is also strengthening its ties with Tesla and xAI, and is making a concerted effort to regain its position in the AI semiconductor market by 2026. - Japan Strengthens Support for Rapidus and Advanced Packaging Technology
Support from the Japanese government and financial community for Rapidus, Japan’s cutting-edge domestic semiconductor project, has entered a new phase. The government has put in place a framework that guarantees 80% of the financing for Rapidus and is strongly encouraging private banks to provide financing. In addition, the government has secured investments and technical cooperation from SoftBank, Canon, and others, and is solidifying the framework for mass production of 2nm chips by 2027. On the technology side, Rapidus demonstrated its packaging technology (glass interposer) using a huge 600mm square glass substrate as the key to competing with TSMC. This is the key to “chiplet” technology, in which multiple chips are connected to raise performance, and demonstrates the company’s willingness to participate in the world’s most advanced competition not only in the front-end process (miniaturization) but also in the back-end process (mounting). Specific preparations for mass production are steadily underway, including the installation of IBM’s manufacturing execution system at the Hokkaido plant. - SK Hynix AI Strategy: HBM4 Advance and U.S. Expansion
HBM market leader SK Hynix is adjusting its HBM4 mass production schedule in line with NVIDIA’s plans for its next-generation AI accelerator, Rubin. The company expects to move forward from its original plan and begin full-scale mass production in early 2026, and has begun shipping final samples through its collaboration with TSMC. In general-purpose DRAM, the company is also shifting to a new technology that increases the capacity per cell by a factor of 10. In addition, the company is considering building an advanced packaging (2.5D) line for AI chips in the US. This is intended to close the physical distance between the company and its customers, U.S. big techs such as NVIDIA and Amazon, and to compete with TSMC in a “turnkey” strategy of providing HBM and packaging in one package. There is also a move to open a new office in Seattle to strengthen its presence in the U.S. market in both manufacturing and development. - Power and Infrastructure Challenges for AI Data Centers and “Photovoltaic Convergence”
As AI models grow in size, data center power consumption is exploding, and energy supply is becoming an industry bottleneck. In Michigan, USA, OpenAI and Oracle’s massive data center plans have caused friction with the community over power and environmental issues. Against this backdrop, there is a growing interest in “optoelectronic fusion (silicon photonics)” as a technology that can dramatically improve power efficiency. Marvell Technology acquired Celestial AI, which has optical interconnect technology, and the Taiwanese government has made silicon photonics a pillar of its national AI strategy. The Japanese government has also decided to provide massive subsidies for data centers that utilize renewable energy and nuclear power, and is promoting the development of a set of energy infrastructure and computing infrastructure. 2026 will be the year when the evolution of AI will be tested to see how it can break through energy constraints. - China’s Semiconductor Independence and the Advancement of “EUV” Technology
Even under severe U.S. sanctions, China’s semiconductor industry has stubbornly survived and continues to evolve. Shanghai Microelectronics Equipment (SMEE) has reportedly developed the country’s first prototype extreme ultraviolet (EUV) lithography equipment, suggesting the long-term potential to break ASML’s monopoly. In addition, GPU makers Moore Threads and Biren Technology are seeking IPOs (initial public offerings) to circumvent U.S. regulations and accelerate development by raising capital. Foundries such as SMIC are raising prices of mature processes (legacy nodes) and securing profitability on the back of domestic demand. While U.S. think tanks warn of “loopholes” in China’s use of multi-patterning technology to produce regulated advanced chips, China is strengthening its resistance to supply chain decoupling by investing heavily through state funds and integrating its domestic ecosystem. - Intel’s Struggles and the Future of its Foundry Business
Intel, the former semiconductor champion, is at a critical point in its business restructuring. The company is facing mixed results and challenges in the separation and spin-off of its foundry business, which has been underway under CEO Gelsinger, and in the race to miniaturize its manufacturing processes (e.g., 18A). While there is positive news such as the introduction of the latest high-NA EUV equipment and the partnership with India’s Tata Group, there are reports that major customers such as NVIDIA have decided not to adopt Intel’s manufacturing processes, and trust has not been restored. Furthermore, the company has stepped up its political lobbying efforts, including the appointment of the Trump administration’s economic advisor as head of government affairs, and has been busy securing subsidies under the CHIPS Act. At one point there were rumors of a major merger, but Intel is decisively selling off some businesses and cutting costs while seeking to restructure on its own. 2026 will be the watershed year for whether Intel can remain a major player or go down the path of dismantlement. - The Rise of “Physical AI” and Edge Computing
The trend in AI is expanding from large-scale language models (LLMs) in the cloud to “Physical AI” that performs physical actions in the real world and “Edge AI” that performs processing at the terminal side. The Ministry of Economy, Trade and Industry (METI) has decided to provide 1 trillion yen in support for the development of domestically produced AI for robots, and SoftBank and other companies are focusing on this field. Humanoid robots, self-driving cars, and industrial robots are emerging as new applications for AI. The acquisition of a RISC-V technology company by Qualcomm and Apple’s next-generation chips with enhanced inferencing capabilities are examples of how the main battlefield for AI is “learning. 2026 is expected to be the first year in which AI will step out of the digital world and transform physical labor and services.
