chips

OpenAI's General Manager for South Korea, Harrison Kim, stated at a press conference in Seoul on Wednesday: "One of the areas where we have made the most significant progress and received the most recognition in our collaboration with Samsung Electronics is the joint production and research for the next-generation chips we are developing."Kim also mentioned that Samsung Electronics is "one of the largest enterprises globally applying ChatGPT at scale," and pointed out that employees in the company, both in Korea and overseas, are using ChatGPT in research and development, marketing, and sales. Kim stated that as the chip technology required for faster and more complex computations continues to upgrade, the demand for memory chips will also keep growing, making collaboration with Korean semiconductor suppliers a priority for OpenAI's local office.

According to Bloomberg, the top executive of Kioxia Holdings denied the possibility of establishing closer cooperation with competitors and shareholder SK Hynix, stating that they will control the rapid rise in storage chip prices to avoid impacting long-term demand in the artificial intelligence sector.Currently, storage chip manufacturers are expanding production capacity to meet the surge in orders from artificial intelligence service providers, leading to a significant increase in related chip prices. Some investors believe that forming manufacturing partnerships may help reduce the risk of large-scale capital expenditures.

According to DIGITIMES, the IC design and semiconductor industry generally believes that the duration of the current tight supply and demand for mature process chips will be longer. Most industry opinions suggest that the industry will need to actively compete for capacity at least until 2028, and price increases may become the norm in the industry. Some even believe that by 2030, the supply and demand relationship for mature process chips may not be able to stabilize.Industry insiders in IC design admit that the overall market situation for end products is not particularly hot; it is just that the demand related to cloud AI is excessively large and continues to expand.

Malaysia is leaning towards relying on Huawei Technologies' artificial intelligence chips to power its sovereign AI program. The Southeast Asian country is seriously evaluating the use of Huawei AI hardware as the backbone of a 2 billion ringgit (approximately 494 million USD) initiative aimed at giving the country greater control over national data.Insiders say it is currently unclear how many chips Malaysia will purchase, and these individuals requested anonymity due to unauthorized disclosure. This move faces strong opposition from the administration of U.S. President Donald Trump.

Samsung Electronics' wafer foundry division has recently surpassed 50% in the allocation of production capacity for the sixth-generation high bandwidth memory HBM4 substrate chips in the 4-nanometer process. The industry states that this division will invest more than half of its 4-nanometer wafers into HBM4 substrate chip manufacturing in the second half of this year to expand supply to companies like NVIDIA, Broadcom, and AMD.Samsung initiated mass production and shipment of HBM4 in February this year and plans to increase supply starting from the third quarter, significantly increasing the related wafer input since mid-year. This chip is based on Samsung's wafer foundry 4-nanometer (SF4) process. As of last month, over 50% of its total 4-nanometer capacity has been allocated to HBM4 substrate chips. According to insiders, Samsung's 4-nanometer process is currently at full capacity, with HBM4 substrate chips accounting for about 50% to 60%, and this high proportion will be maintained in the second half of the year.Samsung's wafer foundry production lines at Pyeongtaek Plant 2 and Plant 3 are mass-producing processes ranging from 4 to 7 nanometers, with a monthly capacity of about 30,000 wafers for the 4-nanometer process. Based on this, the wafer input related to HBM4 substrate chips is estimated to be about 15,000 wafers per month. During the second quarter earnings call, Samsung stated that it expects HBM4 revenue to increase more than threefold quarter-over-quarter in the third quarter, with HBM4 revenue in the second half expected to exceed 60% of the company's overall HBM revenue.

NVIDIA CEO Jensen Huang stated that GPT-6 Astra was trained using over 100,000 NVIDIA Grace Blackwell NVLink72 chips. It took only 4 years from ChatGPT to o1, and then to Astra. AGI has arrived. Congratulations to the OpenAI team.

According to a report by Bloomberg on September 4, the artificial intelligence company DeepSeek plans to deploy at least 160,000 top accelerators from Huawei Technologies Co. in a large data center being built in Inner Mongolia, potentially forming one of the largest known clusters of Huawei AI chips and advancing China's efforts to replace Nvidia.DeepSeek intends to use Huawei's next-generation Ascend 950DT chips to run its models, with the installation timeline depending on Huawei's production capacity. Sources say that the company currently does not plan to use the 950DT chips for training, although Huawei has designed and marketed these processors for more demanding training functions. DeepSeek had previously attempted to train models using Huawei chips but has so far relied on Nvidia accelerators to complete this critical step.

According to South Korean media Greened.kr, Samsung Electronics and Arm have officially launched a joint research and development project for edge AI chips. Arm approved a one-time engineering fee at the end of August for the development of the next-generation edge AI SoC chip and officially started the project with Samsung Electronics.The project adopts a collaborative model where Arm provides core technology, and Samsung Electronics integrates and implements it, aiming to jointly develop customized chips for end customers.

U.S. Secretary of Commerce Gina Raimondo said on Wednesday that the Trump administration is considering a new round of import tariffs on semiconductors, which could further increase taxes on foreign chips to attract more semiconductor manufacturing to the United States.However, these tariffs may be implemented alongside tax relief measures for investments in domestic manufacturing companies, following the approach previously taken by the Trump administration in the pharmaceutical industry.Raimondo stated, "This is the principle that the semiconductor industry should follow: If you produce in the U.S., we will give you tariff relief; if you do not produce in the U.S., you will pay tariffs. This is a very wise approach and it works."Trump had previously imposed a 25% tariff on some advanced semiconductors and ordered additional import tariffs on these chips in January of this year after the U.S. Department of Commerce completed relevant trade investigations. However, the scope of the new round of tariffs may expand from semiconductors themselves to products containing chips, thereby affecting imports of data center servers and consumer electronics.

The research institution Sigmaintell has released a storage price indicator showing that in the third quarter of 2026, the overall supply and demand in the global storage market remains tight. Prices for all categories of DRAM (Dynamic Random Access Memory) and NAND Flash (flash memory) continue to rise month-on-month, but the rate of increase has significantly slowed compared to the second quarter, with the month-on-month increase for some storage chips narrowing to within 10%.

According to Bloomberg, investment firm Dragoneer Investment Group is raising approximately $1 billion for a new continuation fund using equity from its star assets such as OpenAI, SpaceX, Databricks, and wealth management firm Creative Planning as "chips." However, sources indicate that Dragoneer's initial fundraising target was $2 billion, and it is currently unclear what specific assets are included in the final $1 billion fund raised.It is reported that Dragoneer hopes to enhance the fund's appeal to investors by establishing a continuation fund that allows for the continued holding of some existing assets for a longer period. This fundraising also reflects that, as the valuations of popular private tech companies like OpenAI and SpaceX continue to attract market attention, holding early or private equity in these companies is becoming an important chip for asset management firms in fundraising and asset restructuring.

LG Electronics will entrust Samsung Electronics to manufacture AI home appliance chips to reduce reliance on TSMC. CoAsia SEMI will develop two IoT chips for LG, with a total project cost of 31 billion Korean won (approximately 1.51 billion RMB), lasting from July 2026 to December 2030.This project is part of the South Korean government's "K-On-Device AI Semiconductor Technology Development" program, which has a total budget of approximately 800 billion Korean won.

According to The Information, the Trump administration is studying a new regulatory rule for artificial intelligence chips aimed at limiting the remote acquisition of computing resources for advanced AI chips in the United States.The report states that after Trump begins his second term in 2025, the White House announced the withdrawal of the AI Diffusion Rule introduced during the Biden administration. Currently, a small team within the U.S. Department of Commerce is working on a smaller, more targeted alternative plan to restrict geopolitical competitors from accessing advanced AI chips through remote methods such as cloud computing.The rule is still in the drafting stage, and specific restrictions have not yet been clarified, but it indicates that the U.S. government may re-strengthen regulations on the cross-border acquisition of advanced AI computing power, focusing on the possibility of circumventing physical chip export restrictions through remote computing services.

According to Nikkei Asia, SK Hynix CEO Kwak Noh-Jung stated that the global shortage of memory chips is expected to last until the end of 2030, with a low risk of oversupply. He pointed out that in the AI era, memory chips are no longer just "commodities," and there are currently no signs of oversupply or industry downturn; demand from AI customers remains strong.SK Hynix is a major supplier to NVIDIA. The report stated that the company remains open to further expanding its investments in the U.S. after launching its first HBM chip packaging plant in Indiana.

According to Bloomberg, the U.S. government is investigating Apex Logistics, headquartered in Singapore, for allegedly playing a role in smuggling NVIDIA AI chips to China. Sources say this could become the first action against a transportation company involved in illegal semiconductor trade.Authorities are verifying whether Apex (a unit of shipping giant Kuehne+Nagel) violated U.S. export restrictions by transporting AI systems assembled by Super Micro Computer. NVIDIA chips are regarded as the industry standard for training and running AI models, and companies like Super Micro Computer assemble them into data center servers.

Zhipu released the lightweight flagship model GLM-5.3 Flash, aimed at competing with DeepSeek V4 Flash, and utilizes over 100,000 domestic chip clusters to provide inference services. The model has a parameter scale of approximately 300B, about 40% of GLM-5.3, and is roughly on par with DeepSeek V4 Flash; it adopts the MoE architecture, with actual activated parameters around 18B, possesses native multimodal capabilities, supports image and video input, and is the first new model with multimodal capabilities following Zhipu's focus on coding strategy.The comprehensive evaluation score exceeds that of the previous flagship GLM-5.2, is on par with Claude Opus 4.8, and is higher than the relevant scores of DeepSeek V4 Pro official version. The pricing is 0.8 yuan per million tokens for input, 2.8 yuan for output, and 0.23 yuan for cache hits, approximately one-tenth of GLM-5.3, with a half-price discount offered for the first two weeks after launch. The model is fully compatible with domestic computing power, online traffic is handled by domestic chips, and Zhipu claims that the cluster hardware efficiency and unit token cost have reached levels comparable to mainstream NVIDIA GPUs; this model was previously tested anonymously on overseas platforms under the code name Ox Alpha.

Amazon and NVIDIA announced that Amazon will deploy approximately 2 million high-end NVIDIA chips in 2027 and 2028 for training and running artificial intelligence models. Additionally, Amazon stated in March this year that it would install 1 million NVIDIA chips in AWS data centers. The chips involved in this transaction include Blackwell Ultra, Rubin, and Rubin Ultra graphics cards.

Apple (AAPL.O) has launched a new Mac mini, equipped with M6 and M5 Pro chips, significantly enhancing its artificial intelligence performance and further strengthening the capabilities of this leading desktop computer in continuous intelligent computing.

Lei Jun announced that Xiaomi's new generation of the Xuanjie family of chips, Xuanjie O3, Xuanjie O100, and Xuanjie D100, have all completed back-plate verification. Xuanjie O3 is an AI flagship SoC aimed at the highest-end products and will debut on the Xiaomi 18 Fold; Xuanjie O100 is a 1.22TB/s high-bandwidth AI acceleration chip, and Xuanjie D100 is the first domestic 3nm intelligent driving high-performance AI chip, with the latter two set to officially launch next year. The three chips cover the AI computing power needs of Xiaomi's entire ecosystem of people, vehicles, and homes.Xuanjie O3 uses a 3nm process, has an area of 133mm², and a transistor count of 24 billion, which is a 26% increase over the previous generation, with an AnTuTu extreme score reaching 5.22 million. The CPU features a ten-core all-large-core design, with a maximum frequency of 4.35GHz, Geekbench single-core score of 3945, and multi-core score exceeding 15000; the GPU has 16 cores and supports ray tracing, with significant improvements in multiple graphics performance metrics and a 64% reduction in power consumption compared to the previous generation. This chip is the industry's first to support LPDDR6, with an NPU tensor computing power of 200TOPS, and is specifically optimized for large language model inference.Xuanjie O100 is the world's first 6nm 3D wafer-level stacked AI acceleration chip, utilizing wafer-on-wafer packaging and hybrid bonding, with two layers of AI-specific high-speed DRAM and one layer of high-performance NPU stacked vertically, achieving a bandwidth of 1.22TB/s. Together with O3, it can achieve a maximum local AI inference speed of 330 Tokens/s. Xiaomi has restarted its large chip research and development for over five years, with a cumulative investment exceeding 21 billion yuan and a chip team of nearly 3000 people.

According to The Information, citing two NVIDIA employees, NVIDIA plans to ship a small batch of a Groq technology-based LPU (Language Processing Unit) to Chinese customers before the end of the year, with several Chinese customers already placing orders. The report states that this chip does not require a downgrade in specifications and can be used with processors available in the Chinese market through software adjustments.NVIDIA subsequently denied the report, stating it is "incorrect," and that the company currently has no LPU sales in China, nor is there a "China-specific LPU" in the product roadmap. In March of this year, Reuters also cited informed sources reporting that NVIDIA was preparing to sell Groq chips in China, to which Jensen Huang later stated in an interview with Punchbowl that the report was "completely wrong," clearly indicating that Groq chips would not enter China. It is recommended to rely on official statements.