読書リサーチレーダー投資フレームワーク
ログイン / 新規登録
すべて技術製品投資株式市場その他

読書

2026-10-28154 投稿

更新

Reports: Alibaba, Tencent, ByteDance Negotiate Up to 10-Year Memory Supply Deals with Samsung and SK Hynix

トピック · 中国AI内存长约重大
2026-10-28 12:29 GMT+8

According to a report by South Korea's Maeil Business Newspaper on October 27, 2026, Chinese tech giants including Alibaba, Tencent, and ByteDance are negotiating long-term supply agreements (LTAs) of up to 10 years with Samsung Electronics and SK Hynix to secure DRAM and NAND flash memory for their AI data centers.

The report notes that while Chinese domestic memory manufacturer ChangXin Memory Technologies (CXMT) is rapidly expanding capacity, Korean suppliers still hold an advantage in high-performance DDR and LPDDR technologies required for AI servers, as well as in mass-production stability. An executive at a server company stated that AI data centers demand extremely high reliability for 24/7 operation, and there remains a significant performance gap between CXMT's server DRAM and products from Samsung and SK Hynix.

Additionally, US-based Micron Technology announced its exit from the Chinese market in 2025 following government restrictions, further pushing Chinese firms toward Korean suppliers. SemiAnalysis statistics indicate China has built at least 24GW of AI data center capacity, second only to the US's 56GW, with another ~20GW under construction or planned. Tight supply has also driven the adoption of LTAs; Samsung plans to allocate at least 70% of its capacity to LTA customers, and KB Securities estimates that approximately 80% of Samsung's planned 2027 memory output is already locked in through long-term contracts.

ソース:INSIDE

リサーチ

Korean Security Firm PaaSoo AI Report: 77.3% of Respondents in Companies Banning Generative AI Still Use Personal Accounts for Work

トピック · 韩国影子AI使用クイックテイク
2026-10-28 13:14 GMT+8

South Korean data security software company PaaSoo AI (파수 AI) released a report titled 'Intelligence Outside Control – Status of Shadow AI Usage and Response in Domestic Organizations' on October 28, 2026. The survey revealed that 77.3% of respondents in organizations that strictly prohibit the use of generative artificial intelligence still access AI services via personal accounts during work hours. This unauthorized 'Shadow AI' usage poses significant security risks.

The survey targeted 151 AI, IT, and security officials from major South Korean companies and institutions. Data showed that 29.1% of organizations allow autonomous use, 22.5% permit only designated services, and 14.6% enforce a principle-based block. Among those enforcing blocks, the vast majority of employees bypass restrictions. Key drivers for Shadow AI prevalence include 'lack of clear policies and guidelines' (28.5%) and 'intent to avoid leaving usage records in company systems' (25.3%).

Currently, only 19.2% of organizations can monitor and control employee usage of personal account AI services. Respondents cited 'internal core data being used for training third-party models' (28.3%) as their top concern, followed by violations of regulations like the Personal Information Protection Act and GDPR (24.0%), and difficulty in incident response due to untraceable usage history (23.7%). PaaSoo AI recommends addressing this challenge by providing secure internal AI tools, ensuring usage visibility, and establishing clear policies.

ソース:미래를 보는 창 - 전자신문

リサーチ

Cloudflare Discloses Workers AI Data Leak: SGLang Bug Caused 155 Cross-Tenant Mix-ups

重大
2026-10-28 12:24 GMT+8

US cloud provider Cloudflare published the investigation results of a data leak in its Workers AI service on October 28. The report states that between August 27 and September 9, the GLM-5.3 model mixed other customers' data into 155 responses. The issue stemmed from cache overflow during the processing of extremely long inputs. Cloudflare completed the fix on September 9, confirming that other Workers AI models were unaffected and no evidence of malicious exploitation was found.

The incident was reported by Gadi Evron, a researcher at security firm Knostic, on September 8. The investigation identified the root cause as a combination of Cloudflare's deployment settings and a bug in the open-source inference software SGLang. Cloudflare set the maximum input length for GLM-5.3 at approximately 1.05 million text units, but the GPU cache could only hold about 680,000 units. When inputs exceeded capacity, SGLang should have rejected the request but continued processing due to the bug, causing the receiving end to misread residual data from other customers left in memory and mix it into the response.

Cloudflare admitted that it knew before launch that the cache capacity did not match the configured input limit, but at the time only checked service status and performance without verifying the correctness of the response content. Additionally, users had reported related vulnerabilities in SGLang as early as July, but Cloudflare did not notice until after the incident occurred. Currently, Cloudflare has increased the GPU cache capacity to approximately 2.04 million text units and corrected the SGLang logic. The company recommends that enterprises using the same architecture update to version v0.5.22 released on October 19 or higher, and verify that cache capacity is sufficient to handle the maximum input length.

ソース:iThome

リサーチ

US NIST Sets Retirement Timeline for RSA, ECDSA and Other Traditional Public-Key Algorithms

構造的
2026-10-28 12:26 GMT+8

The US National Institute of Standards and Technology (NIST) released document IR 8547, "Transitioning to Post-Quantum Cryptography Standards," on October 27, 2026, setting a concrete retirement timeline for traditional public-key algorithms. The document lists algorithms to be phased out, including RSA, ECDSA and EdDSA for digital signatures, and RSA, Diffie-Hellman and MQV for key establishment. NIST plans to mark these as Deprecated after 2030 and Disallowed in NIST standards and guidelines after 2035.

US federal high-sensitivity systems face earlier deadlines. Under Executive Order 14412 issued in June 2026, federal High Value Assets and High Impact Systems—excluding national security systems—must complete post-quantum key establishment migration by end of 2030 and digital signature transition by end of 2031.

NIST emphasized that migration is not merely algorithm replacement but involves the entire information system and cryptographic supply chain. Network security protocols such as TLS, SSH and IPsec must support new key establishment and authentication mechanisms; libraries like OpenSSL must support post-quantum algorithms; Hardware Security Modules (HSMs) and Trusted Platform Modules (TPMs) may require firmware updates or redesigns; and Public Key Infrastructure (PKI) must adjust certificate issuance, validation and revocation to support post-quantum digital signatures.

NIST lists hybrid cryptographic mechanisms—combining traditional and post-quantum algorithms—as a transitional option to address compatibility needs, but warns that hybrid approaches may increase implementation complexity and security costs. Given that cryptographic algorithms historically require 10 to 20 years to move from standardization to full deployment, NIST recommends that governments and enterprises promptly inventory algorithms in use, assess system dependencies and vendor support capabilities, and plan migration priorities based on data sensitivity.

ソース:iThome

リサーチ

Korean SNS Data Firm Featuring Launches AI Agent to Automate Influencer Marketing Workflows

トピック · Featuringクイックテイク
2026-10-28 10:32 GMT+8

South Korean SNS data analytics firm Featuring unveiled its 'Featuring AI Agent' on October 28 during its inaugural marketing conference, FUTURING 26, held in Gangnam, Seoul. The tool is designed to automate the entire influencer marketing lifecycle—from candidate selection and strategy formulation to budget allocation and performance evaluation—through natural language dialogue.

According to live demonstrations, marketers can input commands such as "Find YouTubers similar to Channel A with over 3% engagement for a cream PPL campaign." The AI agent then filters candidates based on platform, target country, follower size, and demographic criteria. The system automatically flags and excludes accounts with abnormally high engagement rates, compares expected advertising costs against reach efficiency within a specified budget, and even drafts collaboration proposals and final performance report presentations. If actual performance data is not connected, the system explicitly labels outputs as "sample data."

Choi Hwa-seop, CTO of Featuring, stated that the core value of vertical AI agents lies in deep industry understanding, reliable data foundations, and the ability to translate recommendations into actionable execution. This launch marks Featuring's transition from a pure data analysis tool to an automated execution platform, aiming to address industry pain points such as high manual screening costs and fragmented data in influencer marketing.

ソース:스타트업 생태계의 모든 것 \'플래텀(Platum)\'

リサーチ

Samsung Gets MFDS Approval for AI Sleep Apnea Detection with 3-Level Severity Grading

トピック · 三星睡眠呼吸暂停检测重大
2026-10-28 08:52 GMT+8

Samsung Electronics announced on October 28 that its upgraded sleep apnea detection feature has obtained medical device approval from South Korea's Ministry of Food and Drug Safety (MFDS). The feature utilizes artificial intelligence algorithms to precisely analyze the number of breathing disturbances per hour during sleep, upgrading the results from simple presence/absence detection to three severity levels: mild, occasional, and frequent.

The feature had previously received approval from the U.S. Food and Drug Administration (FDA) in July this year. The newly approved functionality is scheduled to launch in South Korea next month (November), initially available on the Galaxy Watch Ultra2, Galaxy Watch9, and Galaxy Ring. The Galaxy Ring is noted for its lightweight design and comfort, which aids in stable sleep data collection.

Samsung stated it will gradually expand the feature to existing models like the Galaxy Watch7 and Watch8 running Wear OS 7.0 or higher via future software updates. Users need to measure via the Samsung Health Monitor app; detecting signs requires at least two measurements of over five hours of sleep within ten days.

Untreated sleep apnea can lead to hypertension or stroke. Choi Jong-min, Executive Vice President of the Health Development Group at Samsung's Mobile eXperience (MX) Division, emphasized the goal of helping users recognize hard-to-detect risk signals in daily life. Currently, Samsung's sleep apnea feature is available in 84 countries worldwide, including South Korea, the US, and Brazil.

ソース:zdnet.co.kr

リサーチ

South Korean Navigation Platform TMAP Signs MOU with Incheon City and KARI for AI Connected Car Development

クイックテイク
2026-10-28 10:11 GMT+8

On October 28, 2026, South Korean navigation platform TMAP Mobility announced that it had signed a memorandum of understanding with Incheon Metropolitan City, the Korea Automotive Research Institute (KARI), and related companies. The agreement aims to jointly develop AI connected car technologies and build the foundation for the future mobility industry.

The participating entities include Incheon City, KARI, and ten private companies. These parties will collaborate on innovative projects for AI connected car technology development, support research task planning, and promote industrial growth. They also plan to establish Cheongna as a core hub for this sector, forming a practical working committee if necessary to discuss detailed initiatives.

TMAP Mobility stated it would leverage approximately 20 years of accumulated traffic data and navigation operation expertise to explore the application possibilities of AI connected car technologies with its partners and seek opportunities for developing new mobility services. Yoo Jung-hwa, Data Business Director at TMAP Mobility, expressed commitment to helping Incheon leap forward as a future mobility hub.

ソース:zdnet.co.kr

リサーチ

Qualcomm and Anduril Announce Joint Development of 'Bluechip' Military Laptops with TSMC Arizona Chips

トピック · Bluechipクイックテイク
2026-10-28 10:16 GMT+8

US chipmaker Qualcomm and American defense technology firm Anduril announced on October 27, 2026, that they are jointly developing a computer product line named 'Bluechip,' targeting the US military, government agencies, and public safety sectors. The first products will feature Qualcomm's Snapdragon X series system-on-chips (SoCs).

According to the disclosed production plan, the core chips will be manufactured by Taiwan Semiconductor Manufacturing Co. (TSMC) at its facility in Arizona, USA. Anduril is responsible for the laptop design and manufacturing, with circuit board assembly taking place at its 'Arsenal-1' facility in Ohio. External cases and battery components will also be sourced through a US-based supply chain. Qualcomm stated that over 90% of the component procurement costs for the current Bluechip prototypes come from suppliers in the US and allied nations.

The companies reported that the prototype for the first product has been completed and they expect to begin accepting orders next year. This initiative aims to secure a stable supply of computing equipment to US government and military entities even if overseas supply chains are disrupted. The Bluechip product line is also planned to expand to computing devices beyond laptops in the future.

ソース:zdnet.co.kr

リサーチ

Robotics Firm Realbotix Launches 'Echo' AI Companion Robot Priced Up to Nearly $60,000

クイックテイク
2026-10-28 10:09 GMT+8

According to ZDNet Korea, robotics company Realbotix launched its artificial intelligence companion robot named 'Echo' on October 27, 2026. The product features a highly realistic appearance with silicone-based skin and articulated joints, while integrating with major conversational AI platforms such as ChatGPT, Claude, and Gemini.

The Echo series includes tabletop and full-body models. The entry-level 'Echo Tabletop' stands about 81 cm tall, weighs 11 kg, and has 17 joints. The full-body female model is approximately 182 cm tall, the male model is 195 cm, both weighing around 40 kg, with the upper body offering 39 degrees of freedom. While Realbotix describes the full-body model's movements as flexible, the report notes it does not yet achieve human-level range of motion.

Pricing starts at $11,950 for the base model and reaches $59,950 for the top-tier configuration. Customers must pay a $199 reservation deposit and a monthly software subscription fee of $20 after purchase.

Realbotix plans to limit initial production to 250 units, comprising 225 tabletop robots and 25 full-body robots, with deliveries expected to begin in the second half of 2027. An optional 'Vinci' system enhances computer vision and on-device processing, supporting facial recognition, gaze tracking, and sub-second latency.

ソース:zdnet.co.kr

リサーチ

AI Marketing Firm Appier Appoints Sanam Lou as VP of Global Marketing, Plans London Office for EMEA Expansion

トピック · Appier欧洲扩张计划クイックテイク
2026-10-28 08:52 GMT+8

AI marketing technology company Appier has announced the appointment of Sanam Lou as Vice President of Global Marketing to enhance its competitive edge in international markets. According to a report by Korean tech media outlet ZDNet Korea on October 28, 2026, Lou will oversee Appier's global marketing organization across Europe, Middle East, and Africa (EMEA), Asia-Pacific, and the United States.

As part of this personnel change, Appier also plans to establish a new office in London, UK, to formally target the high-growth-potential EMEA and US markets. This move indicates the company's intention to optimize its global marketing strategy through local hubs, aiming to translate marketing efforts into tangible business growth.

Sanam Lou brings over 25 years of expertise in technology, strategy, and business growth. She previously served as a Fractional CMO and executive coach at Beyond Founding and led EMEA marketing at machine learning ad platform Moloco. Chih-Han Yu, CEO and co-founder of Appier, stated that Lou's extensive experience and strategic vision will help connect technological innovation with evolving market demands.

ソース:zdnet.co.kr

リサーチ

AI Glasses Maker Even Realities Launches Camera-Less Fashion Model G2 Brown2 at $649

クイックテイク
2026-10-28 09:10 GMT+8

AI hardware company Even Realities has recently launched the G2 Brown2 smart glasses along with accompanying accessories. According to ZDNet Korea, the product line aims to address the bulkiness and awkward design often associated with traditional smart glasses due to their built-in cameras.

The G2 Brown2 features a color scheme inspired by leopard print and comes in two frame styles: round and rectangular. Its appearance is indistinguishable from ordinary fashion eyewear. To achieve this slim profile, the model omits the camera module, focusing instead on audio interaction and visual assistance functions.

In terms of features, the G2 Brown2 supports real-time translation, navigation, AI assistant conversations, and note-taking. After users wake the device with "Hey Even," the system utilizes Perplexity and Even's proprietary Large Language Models (LLMs) to provide answers in over 22 languages, and can transcribe or summarize conversations in real time. Additionally, it integrates walking and cycling navigation guidance based on Google Maps.

The launch also includes polarized clips attachable to the lenses (G2 Clip, $149) and stainless steel portable chains (G2 Chain, $79). The G2 Brown2 smart glasses themselves are priced at $649. This product strategy indicates that some AI wearable manufacturers are attempting to gain mainstream consumer aesthetic acceptance by sacrificing certain visual capture capabilities.

ソース:zdnet.co.kr

リサーチ

BrainChip Announces Sequora Processor IP for Running Generative AI and LLMs on Battery-Powered Edge Devices

クイックテイク
2026-10-28 19:00 GMT+8

US-based semiconductor IP company BrainChip announced the launch of its 'Sequora' processor intellectual property (IP) core on October 28, 2026. The IP is designed to directly execute generative AI and Large Language Model (LLM) tasks on battery-powered edge devices, such as earphones and sensor nodes.

A key technical feature of Sequora is its optimization for State Space Models (SSMs). Unlike traditional Transformer architectures where memory requirements scale with sequence length, SSMs represent input history with a fixed-size state. This allows Sequora to maintain constant memory usage regardless of sequence length, making it suitable for resource-constrained edge computing scenarios.

BrainChip stated that Sequora complements its existing Akida neuromorphic processor IP: Akida handles streaming sensor data like vibration, audio, and radar, while Sequora manages on-device LLM inference, voice-to-voice dialogue, and AI agent tasks. Sequora uses a shared multiply-accumulate (MAC) array for both signal processing and neural network operations, eliminating the need for a separate DSP preprocessing stage, and supports software updates for new algorithms.

Sequora is currently available as licensable semiconductor IP, provided in fully synthesizable RTL format along with verification IP, synthesis scripts, and a Software Development Kit (SDK). This release marks a recent development in specialized edge AI chip IPs addressing the challenges of localizing large model deployment.

ソース:EDN Japan

リサーチ
次へ →
ログイン / 新規登録
読書リサーチレーダー投資フレームワーク