LectureRechercheRadarCadre d'investissement
Connexion / Inscription
ToutTechnologieProduitsInvestissementBourseAutres

Lecture

2026-10-27149 publications

Actualiser

Supply Chain: TSMC SoIC Monthly Capacity to Top 30k Wafers in 2027, Chiayi AP7 Leads Expansion

Sujet · 台积电先进封装Matériel
2026-10-27 09:25 GMT+8

Driven by strong demand for AI and High-Performance Computing (HPC), TSMC is accelerating the expansion of its SoIC advanced packaging capacity, according to a report by MoneyDJ citing supply chain sources. The company's SoIC monthly capacity is estimated at around 10,000 wafers in 2026 and is projected to exceed 30,000 wafers in 2027, representing multiple-fold growth.

SoIC is a key technology for 3D IC vertical stacking, utilizing Hybrid Bonding to increase interconnect density and reduce power consumption. As AI chip complexity rises, SoIC has become another critical advanced packaging capability that TSMC is expanding, following CoWoS.

The new capacity will primarily be added at the Chiayi AP7 P2 facility in Taiwan, while existing capacity is mainly located at Zhunan AP6. Additionally, TSMC plans to prioritize SoIC deployment in its first planned advanced packaging plant in the United States to address US customers' needs for local supply chains.

Sources :MoneyDJ理財網

Recherche

Infineon and Enphase Announce World's First 900V GaN Bidirectional Switches for AI Data Center Solid-State Transformers

Matériel
2026-10-27 01:51 GMT+8

German chipmaker Infineon Technologies and U.S. energy technology company Enphase Energy announced on October 26, 2026, that they are deploying the world's first 900V gallium nitride (GaN) monolithic bidirectional switches (BDS) into Enphase's IQ Solid-State Transformer (SST) power modules designed for AI data centers. This move marks a significant step for GaN technology challenging the traditional dominance of silicon carbide (SiC) in high-power conversion applications.

Solid-state transformers are emerging as key components in next-generation power distribution architectures, offering advantages in efficiency, power density, and reliability. However, handling medium-voltage inputs typically requires series-connected multiple power conversion units to share the voltage load, increasing system losses and complexity. By raising the blocking voltage of GaN devices from the common 650V to 900V while maintaining high-frequency switching characteristics, Infineon enables Enphase to process medium-voltage input with fewer series inverters, keeping component ratings below 1kV.

According to Adam White, Division President Power Systems at Infineon, the new 900V CoolGaN BDS expands the range of applications where customers can benefit from GaN's efficiency and power density advantages, opening new possibilities for high-performance power conversion systems. Hans van Antwerpen, Senior Vice President and CTO at Enphase Energy, noted that the higher voltage headroom allows for the removal of converter stages and reduction of components in the power path, translating to higher efficiency, power density, and improved system reliability.

The 900V GaN BDS devices have been successfully tested and are currently being integrated into Enphase's IQ SST product development, with production qualification expected to follow. Beyond AI data centers, this technology is also anticipated to benefit commercial solar inverters by enabling three-phase designs for 480V applications in the United States and 400V in other regions. For investors tracking AI infrastructure supply chains, this progress highlights how power semiconductor manufacturers are addressing energy and thermal bottlenecks through material innovation.

Sources :Semiconductor for You

Recherche

South Korea's JB Financial Group Posts Record Q3 Profit, Declares Dividend

Prise rapide
2026-10-27 16:46 GMT+8

South Korea's JB Financial Group announced on October 27 that its net profit attributable to controlling interests reached 210 billion KRW in the third quarter, up 0.8% year-on-year. Cumulative net profit for the first three quarters stood at 595.7 billion KRW, a 2.9% increase from the same period last year. Both figures represent all-time highs for the company.

Key operational metrics remained robust: the return on equity (ROE) was 13.3%, return on assets (ROA) was 1.08%, cost-to-income ratio (CIR) was 35.8%, and the provisional common equity tier 1 (CET1) ratio was 12.39%. The company noted that its asset rebalancing strategy kept the growth rate of risk-weighted assets (RWA) at 5.1%, significantly lower than the 10.0% growth in won-denominated loans, thereby enhancing capital efficiency.

Major subsidiaries contributed steady earnings: Jeonbuk Bank posted 72.5 billion KRW, Gwangju Bank 82.4 billion KRW, and JB Woori Capital 66.3 billion KRW. Overseas subsidiary Phnom Penh Commercial Bank (PPCBank) in Cambodia generated 14.3 billion KRW in net profit.

As part of its shareholder return policy, JB Financial Group's board of directors resolved to pay a quarterly dividend of 317 KRW per share.

Sources :zdnet.co.kr

Recherche

Former Intel CEO Pat Gelsinger Backs Nagare Labs' $5M Seed Round to Break AI 'Memory Wall' with Z2RAM

Sujet · Nagare LabsPrise rapide
2026-10-27 12:10 GMT+8

According to TechNews, former Intel CEO Pat Gelsinger is backing a startup called Nagare Labs, which is developing a new memory technology named Z2RAM aimed at breaking through the semiconductor industry's long-standing "memory wall." The company has completed its first funding round, raising $5 million.

Founded by Georgia Institute of Technology professors Suman Datta and Shimeng Yu, Nagare Labs states that Z2RAM aims to combine the advantages of existing mainstream memory types: "speed close to SRAM, density close to DRAM, and scalability similar to Flash." Unlike traditional DRAM cells that use a transistor and capacitor structure, Z2RAM cells utilize two transistors. The gate capacitance of the second transistor stores charge, theoretically preventing charge depletion during reads and improving latency.

Pat Gelsinger noted that AI workloads are inherently memory-bound, and there is unprecedented interest in breakthrough technologies that can simultaneously address bandwidth and capacity issues. Nagare Labs claims that the bit density of its first-generation Z2RAM could be more than five times that of SRAM; for example, where a 64MB SRAM block fits, Z2RAM could hold approximately 320MB in the same space, with comparable bandwidth.

However, the technology is currently only a "lab prototype." Datta stated that the company's goal is to provide working memory chip samples within four years and begin mass production shipments within ten years. Additionally, Shimeng Yu pointed out that there are currently no standards for integrating such new memory into accelerators, and leveraging high capacity and speed requires 3D hybrid bonding, for which standardization has not yet been established.

Sources :TechNews 科技新報

Recherche

CAS Investment Partners Founder Cliff Sosin Discusses AI Prediction Limits and Avoiding AI Infrastructure Investments in Podcast

Matériel
2026-10-27 16:00 GMT+8

The US-based investment podcast 'Invest Like the Best' released Episode 497, titled 'The Limits of Intelligence', on October 27, 2026. The guest is Cliff Sosin, founder of CAS Investment Partners. The firm was started in 2012, and Sosin previously spent five years in the Fundamental Investment Group at UBS, an equity long-short business within the investment bank. According to the episode description, Sosin explains why he thinks there are hard limits to what even the smartest AI can predict and which businesses benefit as intelligence gets cheap.

The show notes indicate discussions on frontier labs as businesses, the Jevons Paradox and limits of demand, and automated labs with speed limits on experiments. Notably, Sosin covers why he stays away from investing in the AI build-out and how AI agents will push customers to the lowest-cost producer. He also addresses why he is comfortable holding roughly 80% of his portfolio in Carvana and discusses educating kids in the age of AI.

For readers interested in AI industry investment opportunities, this episode offers a cautious perspective from an established value investor. Amid market narratives focused on AI compute expansion and model capability breakthroughs, Sosin's argument that intelligence is not an unbounded scalar but a ratio, along with his emphasis on the hard boundaries of AI prediction, provides a different framework for assessing long-term risks in AI-related assets. While the full transcript requires login on the Colossus platform, listeners can locate specific discussions on agent economics, capital cycle lessons, and TSMC's physical moats using the detailed timestamps provided by the publisher.

Sources :colossus.com

Recherche

Nick Maggiulli, Personal Finance Data Blogger: Social Security Benefits Will Not Drop to Zero After Trust Fund Depletion; Advises Planning with 70% of Estimated Amount

Matériel
2026-10-27 19:45 GMT+8

If you are in your 20s, 30s, or 40s, there is a high probability you have heard since childhood that the U.S. Social Security system is going bankrupt. The 2026 U.S. Social Security Trustees Report indeed predicts that the Old-Age and Survivors Insurance (OASI) trust fund, which pays retirement and survivor benefits, will be depleted by 2032. However, Nick Maggiulli, author of Of Dollars and Data, points out that trust fund depletion does not mean benefits drop to zero. Social Security payments come from two sources: payroll taxes from current workers and the trust fund itself. Once the fund is exhausted, the system cannot borrow money and can only pay what payroll taxes cover. According to the Trustees Report estimates, in the first year after 2032, benefits will drop to 78% of scheduled amounts. The average monthly Social Security benefit in the United States is currently about $2,000; 78% means about $1,560—indeed a reduction, but far from nothing.

Maggiulli further explains why this number will continue to decline over time. As the ratio of the working population to the retired population continues to worsen, the proportion of benefits covered by payroll taxes will also decrease year by year. By 2100, the Trustees Report estimates the payment ratio will drop to 62% of scheduled benefits. He admits no one knows exactly how things will develop in reality, but this is the direction indicated by the mathematics.

With the premise understood that benefits will decrease but not disappear, Maggiulli shifts the question from whether Social Security will exist to how much you can actually count on. His core recommendation is a simple planning method: log into the U.S. Social Security Administration’s my Social Security account, view the system’s estimated monthly retirement benefit, and then multiply it by 0.7. If the system shows $3,000 per month, your planning assumption would be $2,100.

Why 0.7 instead of 0.78? Maggiulli believes Congress is likely to take action—either raising taxes or cutting benefits—to mitigate the worst-case scenario, but the likelihood of completely solving the deficit is not high. 70% is his compromise estimate combining both scenarios. He cites a 2025 survey by the National Academy of Social Insurance: 85% of respondents believed benefits should be maintained or increased, even if it meant raising taxes on some or all Americans. Among the repair options listed by the Social Security Administration, the only option that could eliminate the entire deficit on its own was raising the payroll tax rate from 12.4% to 16.4%. Based on this public opinion foundation and the structure of the proposals, Maggiulli leans toward believing the final outcome will involve tax increases rather than significant benefit cuts, making 70% more conservative than 78% but more realistic than complete disappearance.

Maggiulli also explained the specific calculation method for Social Security benefits, as many people mistakenly believe it works like a pension where working longer yields more. In fact, the calculation involves three steps: First, adjust your annual income to an indexed level based on national wage growth; second, take the highest 35 years of indexed income and divide by 420 months to get the Average Indexed Monthly Earnings (AIME); third, use the Social Security Administration’s bend point formula to calculate benefits. For someone turning 62 in 2026, the bend points are: 90% of the first $1,286 of AIME, 32% of the amount between $1,286 and $7,749, and 15% of the amount exceeding $7,749. This means lower-income earners receive a higher replacement rate, while higher-income earners receive a lower proportion.

Sources :Of Dollars And Data

Recherche

AI Audio Infrastructure Firm Edge Sound Raises $2M to Enhance Physical AI Perception

Prise rapide
2026-10-27 21:00 GMT+8

US-based AI startup Edge Sound Research Inc. announced today that it has raised $2 million in new funding, led by Stormbreaker Ventures. This brings the company's total funding to $9 million. The capital will be used to build out its audio infrastructure platform, designed to help physical AI systems like robots and autonomous vehicles better understand their real-time environments.

Edge Sound's core technology addresses the challenge of audio being "channel-oriented" rather than "object-oriented." While microphones capture mixed sound signals, Edge Sound uses digital twins of sensor environments combined with visual tracking data to associate specific sounds (like a basketball hitting the floor or a siren) with particular objects. Co-founder and CEO Valtteri Salomaki noted that while humans naturally distinguish sound sources in a multimodal world, current physical AI relies heavily on vision, often treating audio as a secondary input.

The company has already validated its technology in live sports broadcasting, with clients including the NBA and Fox Sports. The new funding will support expansion into broader industries, such as providing autonomous vehicles with "digital ears" capable of identifying emergency vehicles from blind corners using acoustic features like the Doppler effect. Edge Sound plans to use the funds to complete integrations with existing enterprise customers and prepare for a Series A round.

Sources :SiliconANGLE

Recherche

Hanwha Systems' Q3 Operating Profit Jumps 60%, Net Income Plummets 97% on US Shipyard Sale Losses

Sujet · 韩华系统Q3财报Matériel
2026-10-27 17:56 GMT+8

South Korean defense and IT firm Hanwha Systems released its Q3 2026 earnings report. The company posted consolidated revenue of 831.6 billion KRW, up 30% year-on-year, and operating profit of 99.1 billion KRW, a 60% increase. However, net income fell sharply to 3.8 billion KRW, down 97% year-on-year, primarily due to losses classified under discontinued operations resulting from the sale of its stake in the US-based Philadelphia Shipyard.

The defense sector drove this growth, with revenue reaching 617.4 billion KRW (+28% YoY) and operating profit at 76.0 billion KRW (+53% YoY). Key contributors included exports of Cheongung-II (M-SAM) multi-function radars to the UAE, Saudi Arabia, and Iraq, alongside domestic production projects for KF-21 avionics and K2 tanks. Exports accounted for 22% of defense revenue in Q3, with Iraq being the largest contributor among Middle East markets.

The Information and Communication Technology (ICT) division also showed robust performance, generating 212.6 billion KRW in revenue (+37% YoY) and doubling its operating profit to 24.0 billion KRW. The operating margin improved from 7.7% to 11.3%, aided by high-value system integration projects for affiliates such as Hanwha Life Insurance.

Looking ahead, Hanwha Systems expects full-scale revenue recognition from existing Middle East contracts to begin in 2027. The company has signed a term sheet with UAE state-owned defense firm EDGE Group for an integrated air defense network, signaling a shift from single-system sales to comprehensive solutions where AI-driven system interoperability becomes a critical competitive factor.

Despite positive top-line outlooks, profitability growth faces headwinds from increased investment. The company guided that self-funded R&D expenses would triple or quadruple compared to last year (reaching approx. 75-100 billion KRW), with significant concentration in Q4. Continued high R&D spending next year for laser anti-aircraft weapons and mobile equipment suggests that defense operating profits may remain flat year-on-year even as revenues grow.

Sources :zdnet.co.kr

Recherche

Siemens: Enterprise AI Adoption Stalls on Infrastructure and Governance, Not Model Capability

Sujet · 西门子企业AI落地瓶颈Matériel
2026-10-27 09:34 GMT+8

Siemens recently released a diagnosis regarding enterprise AI adoption through ZDNet Korea, stating that the current bottleneck lies not in model performance, but in infrastructure and governance. According to data cited by Deloitte Consulting Korea, global enterprise AI adoption rates have reached 88%, yet only 39% of companies have realized actual financial returns, and merely 7% have achieved full-scale expansion across the organization. Siemens attributes this gap to the lack of a foundational connection between AI and enterprises' actual data, business workflows, and frontline personnel.

Siemens identified five common operational gaps present at the enterprise level: First, data is trapped in independent systems such as ERP and MES, preventing AI from understanding the enterprise's specific business context; second, machine learning outputs fail to translate into concrete business actions; third, AI functionalities are not integrated into the business interfaces used daily by frontline employees; fourth, workflow handovers between humans and AI agents are not smooth; fifth, there is a lack of permission management and audit trail systems. The common characteristic of these five gaps is that they are not issues of model capability, but rather problems of system connectivity, process design, and governance architecture.

The regulatory environment provides additional context for this diagnosis. The South Korean 'AI Basic Act,' which took effect on January 22, 2026, classifies employment, healthcare, finance, public safety, and education as high-impact areas, requiring companies to fulfill obligations for risk management, impact assessment, and compliance reporting. Based on this, Siemens emphasized that auditability, traceability, and human-in-the-loop controls are not features to be added later, but governance elements that enterprises must build into their design phase to achieve faster scaling.

Sources :zdnet.co.kr

Recherche

Volkswagen Gets L3 Testing Permit in Hefei, China; XPeng Approved for City-Wide L4 Robotaxi Demo in Guangzhou

Matériel
2026-10-27 16:27 GMT+8

German automaker Volkswagen Group (China) and Chinese electric vehicle maker XPeng have recently announced they obtained higher-level autonomous driving road permits in China. Volkswagen stated it has been approved to conduct Level 3 (L3) conditional automated driving tests on public roads in Hefei, Anhui Province, with testing already underway. XPeng announced its robotaxi service has been approved for Level 4 (L4) demonstration operations across the entire city of Guangzhou, Guangdong Province. These developments mark a gradual regulatory thaw in China's autonomous driving sector following the Baidu Wuhan incident earlier this year.

Volkswagen's permit is for road testing only and does not yet allow the sale of L3 vehicles to consumers. The permission also covers large-scale driving data collection in Hefei, aiming for L3 technology readiness in China by 2027. Volkswagen noted the permit was issued under China's strictest L3 regulatory framework and complies with the latest mandatory safety design standards, but did not specify the issuing authority or fleet size. Volkswagen's autonomous driving software in China is developed by CARIZON, a joint venture between Volkswagen's software subsidiary CARIAD and Chinese chipmaker Horizon Robotics.

XPeng's qualification for intelligent connected vehicle demonstration application in Guangzhou allows city-wide L4 robotaxi demonstration operations, symbolizing a shift from technical verification to operational and business model validation. Initial access will be gradually opened via invitation codes. XPeng Chairman and CEO He Xiaopeng previously reaffirmed the goal of operating passenger services without safety drivers by 2027. Technically, XPeng's robotaxi is based on the GX model, equipped with four self-developed Turing AI chips delivering 3,000 TOPS of computing power, and uses the VLA 2.0 Vision-Language-Action model.

Electrek noted that China's Ministry of Industry and Information Technology issued the first batch of L3 product access permits in December 2025 to Changan and BAIC Arcfox models; GAC Aion Hyptec and XPeng have also obtained L3 road test permits, meaning Volkswagen is catching up on Chinese brands' home turf in the L3 domain. The Next Web, citing Bloomberg analysis, indicated these permits show Chinese regulators are gradually moving past the shadow of the late March incident where multiple Baidu Apollo Go robotaxis stalled on Wuhan roads. China subsequently suspended new robotaxi permits in April and launched an industry safety review, which concluded at the end of June, with some cities resuming permit issuance starting in July.

Sources :TechOrange 科技報橘

Recherche

Vertiv Completes $1.45 Billion Acquisition of UtilityInnovation Group to Expand Power Solutions for AI Data Centers

Sujet · VertivMatériel
2026-10-27 18:30 GMT+8

Vertiv Holdings Co. (NYSE: VRT), a U.S.-based global leader in critical digital infrastructure, announced on October 27, 2026, that its wholly-owned subsidiary, Vertiv Corporation, has completed the acquisition of Utility Innovation Holdings, Inc., operating as UtilityInnovation Group (UIG). The transaction involves approximately $1.45 billion in upfront cash, along with additional contingent consideration of up to $1.15 billion based on achieving certain EBITDA targets over 12- and 24-month periods.

Founded in 2020 and headquartered in Raleigh, North Carolina, UIG specializes in microgrid solutions, advanced power controls, and behind-the-meter (BTM) power architecture design for data centers. This acquisition extends Vertiv's portfolio upstream from chip-level cooling and distribution to the grid interconnect, adding capabilities in microgrid control, onsite generation orchestration, and energy storage coordination. Giordano Albertazzi, CEO of Vertiv, stated that as AI data center operators accelerate deployment in an increasingly power-constrained environment, access to power has become a critical competitive advantage. The addition of UIG allows Vertiv to offer coordinated solutions from behind-the-meter generation to the chip, helping customers achieve faster time-to-power without being tied to a single generation technology or supplier.

At the approximately $1.45 billion purchase price, the acquisition represents roughly 13 times UIG's expected 2027 EBITDA. Vertiv anticipates the multiple will be significantly lower if the full earnout is paid. The company expects the acquisition to be accretive to adjusted earnings per share in the first year following completion. J.P. Morgan Securities LLC acted as financial advisor to Vertiv, while Morgan Stanley & Co. LLC advised UIG.

Sources :prnewswire.com

Recherche

South Korean voice AI firm Hudson AI launches localization agent, claims over 90% automated dubbing QC

Sujet · Hudson AIPrise rapide
2026-10-26 14:27 GMT+8

South Korean voice AI company Hudson AI launched its "Localization AI Agent" at MIPCOM 2026, the television content market held in Cannes, France, from October 12 to 15. The service builds on the company's existing AI dubbing platform, Hudson Studio, aiming to automate the entire workflow from translation and dubbing to quality inspection.

Traditional AI dubbing often requires significant human intervention when dealing with multiple characters or emotionally complex programs. Hudson AI's new agent reportedly distinguishes between more than 40 speakers and separates segments where multiple people speak simultaneously. Instead of merely numbering speakers, the system identifies character roles and relationships within the narrative context, assigning names accordingly. It then generates translations and emotional expressions suited to the scene, including non-verbal sounds like laughter and sighs.

Regarding quality control, the company states that it trained the agent using data from professional translators' reviews and corrections. This allows the agent to autonomously assess dubbing quality and flag only the sections requiring human attention. According to Hudson AI, this process automates over 90% of quality inspections, reducing the volume of manual review to one-tenth of previous levels.

Shin Hyun-jin, CEO of Hudson AI, stated that if AI-generated dubbing still requires item-by-item human checking and correction, it fails to solve the efficiency problem. The company plans to use this technology to help broadcasters and OTT platforms economically distribute their content libraries to overseas markets. Currently, Hudson AI is conducting proof-of-concept tests with foreign media companies and has completed dubbing quality tests in Japanese, English, Spanish, Portuguese, Arabic, and French.

Sources :스타트업 생태계의 모든 것 \'플래텀(Platum)\'

Recherche
Suivant →
Connexion / Inscription
LectureRechercheRadarCadre d'investissement

He gave a specific example: If the AIME is $5,000, the monthly benefit at full retirement age (67) would be $2,345, which is 46.9% of the AIME. If claimed early at age 62, it would be reduced by about 30%; if delayed until age 70, it would increase by about 24%. The maximum monthly benefit in 2026 is $5,181, applicable to those who earned at or above the taxable limit for more than 35 consecutive years and waited until age 70 to retire.

Maggiulli clearly marked the boundaries of these assumptions. He said, 'I certainly made many assumptions. In reality, nothing is guaranteed. Congress could pass legislation tomorrow to completely eliminate Social Security, or next year they could raid the trust fund. They could do many extreme things, but I don’t think they will.' He simultaneously disclosed that he is the Chief Data Officer at Ritholtz Wealth Management, a website that provides financial advice services.

His final judgment is: Do not base your retirement plan entirely on Social Security checks, but assuming they won't be there is a bigger mistake. For the average person retiring in 2032, 70% means about $1,400/month; if you retire later, because wages typically grow faster than prices, the AIME will exceed current levels, and the actual amount may be higher.