LeituraPesquisaRadarFramework de investimento
Entrar / Cadastrar
TudoTecnologiaProdutosInvestimentoBolsaOutros

Leitura

2026-10-28154 posts

Atualizar

South Korean Semiconductor Equipment Maker AP Systems Strategically Invests in SD Optics to Integrate MEMS Optical Inspection for HBM and TGV Markets

Tópico · AP系统与SD光学合作Visão rápida
2026-10-28 16:55 GMT+8

AP Systems, a South Korean manufacturer of semiconductor and display panel equipment, announced on October 28 that it has made a strategic equity investment in SD Optics, a provider of ultra-precision optical inspection solutions. AP Systems stated that it will participate in SD Optics' third-party allotment capital increase by subscribing to newly issued convertible preferred shares (CPS). This move aims to internalize SD Optics' core optical technology into AP Systems' equipment and establish a partnership to explore new markets.

SD Optics is the first company globally to commercialize 'Mirror Array Lens System' (MALS) technology based on Micro-Electro-Mechanical Systems (MEMS). This technology controls the movement of micro-mirror arrays at ultra-high speeds (12,000 Hz), enabling real-time focus adjustment without mechanical vibration. It is suitable for 3D shape measurement of samples with large steps and ultra-high-speed vision inspection. Currently, Zeiss microscope systems have adopted this technology, and SD Optics supplies inspection modules and equipment to companies in the semiconductor packaging, glass substrate, and display sectors.

Under the cooperation plan, the two companies will jointly develop next-generation standalone metrology/inspection equipment and process/inspection composite equipment. The target market focuses on advanced semiconductor packaging fields such as High Bandwidth Memory (HBM), Fan-Out Panel-Level Packaging (FO-PLP), and Through-Glass Vias (TGV) for semiconductor glass substrates. AP Systems also plans to equip its main semiconductor and display devices with focus tracking and 3D line scan modules based on SD Optics' MALS technology to enable real-time measurement of surface topography, step height, and residues before and after processes.

Fontes:zdnet.co.kr

Pesquisa

Report: Samsung Commands Up to 25% Price Premium Over SK Hynix and Micron in HBM Market

Tópico · HBM定价竞争Material
2026-10-28 10:10 GMT+8

According to a report by TechNews citing South Korean media, the competitive landscape in the High Bandwidth Memory (HBM) market has shifted. Samsung Electronics is currently quoting HBM at approximately $4.50 per unit, significantly higher than Micron's $4.00 and SK Hynix's $3.60 to $3.80, representing a premium of up to 25%.

Industry analysis suggests that Samsung's ability to command this high premium stems not from product performance but from supply capability. Data center clients prioritize suppliers who can deliver large-scale, stable production capacity on time, making them willing to pay extra to ensure supply security. This marks Samsung's transition from an HBM laggard a year ago to a market leader with strong pricing power.

Following this news, Samsung's stock rose more sharply than SK Hynix's after the latter announced its best-ever Q3 2026 results. The report also notes that as AI applications expand, demand for GPU HBM capacity has surged, while supply-side constraints remain due to the multi-year timeline required to build new wafer fabs, indicating the current memory supercycle is still in its early stages.

Fontes:TechNews 科技新報

Pesquisa

Amazon Launches Three New Echo Devices: Smart Alarm Clock, Hub Speaker, and Camera-Equipped Studio

Tópico · 亚马逊配送眼镜Visão rápida
2026-10-28 22:17 GMT+8

US technology company Amazon launched three new Echo series smart speakers on October 28, 2026, aiming to expand the application of its generative AI voice assistant, Alexa Plus, in home scenarios. The devices include the Echo Time, a smart alarm clock designed for bedrooms; the updated Echo Dot, which serves as a smart home hub; and the Echo Studio, now equipped with a camera.

The Echo Time features a 3.2-inch display and a sensor suite called "Omnisense." It tracks sleep cycles by detecting room temperature, presence, and breathing via ultrasound, waking users during light sleep phases within 10 to 60 minutes of the set alarm time. It also simulates sunrise lighting and reduces screen brightness and blue light at night.

The new Echo Dot includes an AZ3 chip for local processing to improve response times. For the first time, it integrates a Smart Home Hub function, allowing direct connection to compatible devices without third-party bridges. Within Amazon's Eero router networks, it can also act as a Wi-Fi repeater.

The Echo Studio adds a built-in camera to support Alexa Plus's visual intelligence features, such as identifying objects or explaining game rules. However, the device lacks a physical lens cover; users must rely on a button to mute microphones and disable the camera, with visual signals indicating data uploads. All models are available immediately, priced between €104.99 and €259.99.

Fontes:t3n.de

Pesquisa

Merge CEO Shensi Ding: Executive AI Fluency Sets Organizational Ceiling

Material
2026-10-28 19:51 GMT+8

OpenAI published an interview with Shensi Ding, co-founder and CEO of Merge, focusing on the role of leadership in AI adoption. Ding posits that "If you're an executive, you are the ceiling for AI fluency for your organization," suggesting that a company's ability to leverage AI is bounded by its leaders' understanding and willingness to engage directly.

Ding manages the non-R&D side of Merge but actively codes herself. In the episode, she describes using Codex to help redesign a feature by leveraging the product’s existing design system. She notes that the tool worked through tasks and filled in details she hadn't specified, highlighting a practical workflow where AI assists in bridging high-level intent and technical execution.

She also mentions that her non-engineering teammates are building tools for their own workflows. The resource offers readers a glimpse into how a CEO maintains technical literacy and uses AI agents like Codex not just for oversight, but for direct contribution to product development.

Fontes:youtube.com

Pesquisa

Unitree Reports First Post-IPO Earnings: Revenue Up Nearly 70%, Net Profit Falls Due to AI R&D Surge

Tópico · 宇树科技财务表现Material
2026-10-28 21:38 GMT+8

Chinese robotics manufacturer Unitree Robotics released its first quarterly financial report following its listing on the Shanghai Stock Exchange on October 28. The data shows that for the third quarter ending September 30, the company's revenue reached 665 million yuan (approximately US$99.7 million), a year-on-year increase of 69.93%. However, net profit excluding non-recurring items fell 17.42% to 105 million yuan.

The primary driver of the profit decline was a sharp increase in research and development spending. Unitree's R&D expenses jumped 249.58% year-on-year to 127.04 million yuan, accounting for about 19% of revenue. According to filings with the Shanghai Stock Exchange, this investment is focused on developing the software "brains" of its robots and associated computing power, aiming to enhance the intelligence of its industry-leading robotic hardware.

Despite the pressure on single-quarter profitability, Unitree's overall performance for the first three quarters remained strong. Net profit attributable to shareholders reached 387 million yuan, a significant year-on-year increase of 270.54%. The company attributed this growth partly to revenue expansion and partly to a high base of "significant" one-off expenses incurred during the same period last year. Excluding these non-recurring items, net profit for the first three quarters decreased by 18.78% year-on-year.

In terms of market reaction, Unitree shares rose 8.74% to 695.90 yuan before the earnings release, fueled by optimism in the Chinese robotics sector following recent breakthroughs in AI models like OpenAI's GPT-6 Astra, which demonstrated surprising robotics manipulation capabilities. Prior to this rally, the stock had plummeted more than 60% from its peak after its August trading debut.

Financial health indicators show that the IPO provided a massive injection of cash. As of September 30, Unitree's cash and cash equivalents surged to 7.354 billion yuan from 1.419 billion yuan at the end of 2025. However, cost of sales grew faster than revenue, increasing by approximately 78% to 836 million yuan for the first three quarters, reflecting the cost pressures involved in hardware manufacturing and AI integration.

Fontes:South China Morning Post

Pesquisa

UK Defence AI Company Mind Foundry CEO Lindsay Chadwick: Consolidated Three Business Lines into One, Focusing Solely on Defence

Tópico · Mind FoundryMaterial
2026-10-28 21:30 GMT+8

In late 2025, when Lindsay Chadwick assumed the role of CEO at Mind Foundry, the machine learning company—founded in 2016 by University of Oxford professors Stephen Roberts and Maike Osborne—operated across three distinct sectors: insurance, infrastructure, and defence. Her first major decision was to consolidate these three areas into one, focusing exclusively on defence and national security.

Chadwick explained this choice to Tech Funding News as the result of overlapping conditions: mission, technical fit, and national need. She argued that the company’s talent, products, and capital must be concentrated in the area where it has the "strongest right to win." This decision was not easy; she acknowledged it meant forgoing numerous opportunities and making difficult choices that affected talented employees who had made significant contributions to the company. Prior to this role, Chadwick served as a lawyer advising fast-growing venture-backed startups and university spinouts, and later participated in establishing the UK’s Advanced Research and Invention Agency (ARIA). These experiences gave her a close-up view of "turning ambitious science into organizations capable of delivering real-world impact."

Following this strategic focus, Mind Foundry’s product line became clear. Its flagship product, NIGHTINGALE, is being tested jointly with the UK Ministry of Defence, the Royal Navy, and Thales. It uses machine learning to help sonar operators separate meaningful signals from background noise and prioritize them. A sister product, SENTRY, integrates data from multiple sensors to create a unified situational picture. Chadwick emphasized that the purpose of these tools is not to replace human decision-making, but to reduce cognitive load and enable faster, better-informed judgments. She stated that leadership sometimes means choosing clarity over comfort.

Chadwick holds a firm stance on technological sovereignty. She argues that defence AI is not a standard software business: it processes sensitive sensor data such as sonar and radar, and systems require continuous updates and maintenance after deployment. This means the technology must remain within the country. In her words, the UK must understand and control the technology upon which critical decisions depend. She noted that the same technological approach could also be applied to protecting cables and pipelines beneath British waters in the future.

A policy window has emerged synchronously with this strategic focus. The UK government announced a £298 billion defence investment plan over the next four years, which includes a new £200 million AI investment fund and a £1.6 billion UK Defence Innovation fund through 2030. By late June, UK and European defence tech startups had raised $12.3 billion, nearly double the previous year. Chadwick described this as creating significant opportunity, but noted that the key question remains whether ambition can quickly translate into deployed capability.

The team responded positively to this directional choice. Chadwick revealed that during the company’s restructuring, more than half of the staff invested their own money, including former colleagues and friends who had already left. She described this support as a powerful vote of confidence in the company’s direction and future prospects.

Fontes:Tech Funding News

Pesquisa

South Korean Software Firm Yuracle Signs MOU with KT to Join Global AI Distribution Platform

Tópico · Yuracle与KT合作Visão rápida
2026-10-28 09:20 GMT+8

South Korean enterprise software company Yuracle announced on October 28 that it has signed a strategic memorandum of understanding (MOU) with KT, South Korea's leading telecommunications operator, to jointly expand their domestic and international enterprise artificial intelligence (AI) platform businesses. Under the agreement, Yuracle will participate as a partner in KT's newly launched global B2B AI distribution platform, 'Token Factory'. This platform aims to distribute AI services to domestic and overseas markets across industries such as finance, public sectors, manufacturing, and healthcare, and already includes participation from global AI firms like OpenRouter, Cast AI, and Timely AI.

The two companies will also collaborate technically by integrating Yuracle's AI coding assistant into KT's 'NPU LLM Station' solution. The 'NPU LLM Station' is an on-premise AI infrastructure package developed by KT, which integrates neural processing units (NPUs) from South Korean chipmaker Rebellions, large language models (LLMs), and KT's business-specific AI agents. According to KT, this solution is designed to reduce deployment costs by over 50% and power consumption by more than 25% compared to traditional GPU servers.

Yuracle stated that this integration will enable AI-assisted code writing and development using internal corporate data, even in closed network environments where external cloud connections are restricted. Both firms plan to leverage their respective sales networks to conduct joint marketing and sales efforts targeting financial institutions, government agencies, and large manufacturing enterprises that prioritize security and infrastructure operating costs. Kwon Tae-il, CEO of Yuracle, said in a statement that the company would combine its proven enterprise deployment capabilities with this strong partnership to establish new standards in the enterprise AI market both domestically and abroad.

Fontes:zdnet.co.kr

Pesquisa

Runpod Engineer Chris Vela Demos BenchPod: Choosing vLLM Configs from Data

Tópico · vLLM配置优化Visão rápida
2026-10-28 17:00 GMT+8

Most teams spin up a new open model with the default vLLM recipe and hope for the best. Runpod forward-deployed engineer Chris Vela demonstrates in an AI Engineer video how to choose configs from data instead, starting with the use case and concurrency, rather than relying on guesswork.

Vela demos BenchPod, a benchmarking dashboard built on top of vLLM that tests configs live without restarts. He runs the Kimi K3 model on Nvidia B300 hardware and shows how tuning the 'max batch tokens' parameter cuts inter-token latency.

The episode also covers the transition from R&D to production and asks whether you need a frontier model at all. For AI practitioners focused on inference infrastructure optimization, this specific workflow offers a practical reference point.

Fontes:youtube.com

Pesquisa

German Toy Maker Playmobil's Parent Company Announces Plan to Convert Former Factory into 100MW AI Data Center Campus

Visão rápida
2026-10-28 21:27 GMT+8

According to reports by German technology media heise online and WirtschaftsWoche, Horst Brandstätter Group, the parent company of the well-known toy brand Playmobil, announced on October 28, 2026, that it plans to build a high-performance computing campus on the grounds of its former factory in Dietenhofen, Germany. The campus is designed with a capacity of 100 megawatts and has potential for further expansion.

Horst Brandstätter Group stated that the company will provide the infrastructure, while partners can manage their respective applications independently. The offerings range from simply providing space with corresponding network connections to delivering finished computing power. This business segment is named "Playmobil digital" as a new, independent branch within the corporate group. The team responsible for the construction and operation of the campus has more than 25 years of experience in the areas of data centers and IT.

The Dietenhofen site was previously the only production plant in Germany where Playmobil manufactured toys. It was closed at the end of June this year, with production relocated abroad, resulting in the loss of 350 jobs. Matthias Fauser, a board member of Horst Brandstätter Group, said that the company wants to enable one of the most important infrastructures of the digital future where Playmobil was produced for decades.

Regarding power supply, the site already has a connection to a 110-kilovolt transmission line, and the grid operator N-ERGIE in Nuremberg has confirmed the necessary network capacities. However, this allocation of power has a direct impact on the neighborhood: according to BR, a planned Anthropic data center in nearby Hirschneuses currently lacks the necessary network capacity because the available power has been committed to the Playmobil site.

Fontes:heise online

Pesquisa

Morgan Stanley Macro Strategist Matthew Hornbach and Chief U.S. Economist Michael Gapen Analyze Fed Strategy to Skip a Hike While Signaling Future Tightening

Tópico · 美联储货币政策Material
2026-10-28 20:25 GMT+8

Morgan Stanley published a video interview titled "To Hike or Not to Hike" on October 28, 2026. The episode features discussions by Morgan Stanley's Global Head of Macro Strategy, Matthew Hornbach, and its Chief U.S. Economist, Michael Gapen.

According to the publisher's episode description, the two experts focus on three core topics: what factors could accelerate the pace of monetary tightening; what lies ahead for U.S. Treasury yields; and how the Federal Reserve can choose to skip a rate hike at this stage while using communication strategies to signal that further hikes are still on the table.

For professional readers tracking macro interest rate paths and market expectations, this resource offers a specific perspective from the core research team of a major global investment bank. Although a full transcript was not obtained, the identities of the guests and the concrete policy mechanism discussed (balancing a pause with forward guidance) constitute clear selection value.

Fontes:youtube.com

Pesquisa

Hurun Rich List: ByteDance Founder Zhang Yiming Tops Ranking with 670 Billion Yuan; DeepSeek's Liang Wenfeng Enters Top Ten

Tópico · 胡润百富榜Material
2026-10-27 13:43 GMT+8

The Hurun Research Institute released the '2026 Hurun Rich List' on October 27, 2026. The ranking shows that driven by progress in its AI business and higher company valuations, ByteDance founder Zhang Yiming saw his wealth increase by 200 billion yuan, reaching 670 billion yuan to become China's richest person for the second time and setting a new record for the top spot. Nongfu Spring founder Zhong Shanshan fell to second place with 410 billion yuan.

Notably, among the 263 newcomers to the list, the 'Rookie of the Year' is DeepSeek founder Liang Wenfeng, who entered at seventh place with a wealth of 280 billion yuan on his first appearance, reflecting the rapid accumulation of personal wealth driven by the AI large model boom. Additionally, Wang Xingxing of Unitree Robotics, Wang Zhiyong of HKC, and Zhipu AI founders Liu Debing and Tang Jie also ranked highly among the new entrants.

In terms of industry distribution, this year 10.4% of the listed entrepreneurs are involved in the AI sector with AI as their primary direction, whereas no entrepreneurs with AI as their main business appeared on lists prior to 2019. This data indicates that AI has become one of the core drivers of wealth growth for China's top entrepreneurs.

Fontes:MoneyDJ理財網

Pesquisa

Shenzhen's DiffuSpace Open-Sources Clover-35B Model, Claims Lead Over Autoregressive Peers in Agent Benchmarks

Tópico · Clover-35BMaterial
2026-10-28 19:35 GMT+8

On October 28, 2026, Shenzhen-based AI firm DiffuSpace announced the release and open-sourcing of its latest diffusion language model, Clover-35B-A3B-Preview. The model utilizes a Mixture-of-Experts (MoE) architecture with 35 billion total parameters and 3 billion active parameters, supporting a thinking mode.

According to benchmark results published by the company, Clover-35B-A3B-Preview ranked first among same-level models in six agentic coding and tool-use tasks, including SWE-bench Verified and Terminal-Bench2.1. Its performance reportedly exceeds that of other diffusion models like LLaDA2.2 flash and same-sized autoregressive models such as Qwen3.6-35B-A3B. Specifically, it achieved a score of 73.6 on SWE-bench Verified.

In terms of inference efficiency, internal tests showed the model reaching a generation speed of 680 tokens/s on a single NVIDIA GB300 GPU at BF16 precision. Across 49 agent tasks, the median execution time was reported as 1.52 seconds. Additionally, DiffuSpace previewed a decision-making model for structured judgment tasks, claiming a 95.0% accuracy rate for its 'joint mode' in internal validation.

All performance metrics are self-reported by the vendor and have not yet been independently reproduced. The model weights are available on Hugging Face and GitHub.

Fontes:zhidx.com

Pesquisa
Próxima →
Entrar / Cadastrar
LeituraPesquisaRadarFramework de investimento

It should be noted that this article is part of an editorial partnership between Tech Funding News and Mind Foundry investor Parkwalk. Parkwalk has invested since the company’s founding through the University of Oxford Innovation Fund, returned investments via its Opportunities Fund in 2017 and 2019, and participated in the $13.6 million Series A round in 2020. Consequently, Chadwick’s statements serve as both a strategic explanation and a market positioning statement. The article does not disclose Mind Foundry’s revenue, contract values, or customer count, nor does it address company valuation or the specific terms of the latest funding round.