LectureRechercheRadarCadre d'investissement
Connexion / Inscription
ToutTechnologieProduitsInvestissementBourseAutres

Lecture

2026-10-29198 publications

Actualiser

Cowboy Space Unveils Purpose-Built Orbital Data Center Rocket, Plans 'Big Iron Express' Launch Service to Challenge SpaceX

Sujet · Cowboy SpaceMatériel
2026-10-29 02:24 GMT+8

US startup Cowboy Space this week unveiled plans for its own purpose-built data center rocket, designed to carry five racks of NVIDIA Vera Rubin GPUs into orbit. This move marks a shift from experimental satellite payloads to dedicated hardware for space-based computing. Led by Robinhood co-founder Baiju Bhatt and backed by investors including Index Ventures, the company aims to address the growing demand for AI compute capacity.

Beyond deploying its own infrastructure, Cowboy Space plans to use the rocket to offer commercial launch services under the name “Big Iron Express,” directly competing with SpaceX. In an interview with Newcomer, Bhatt stated that the company’s “north star” is to make space compute cost-competitive with terrestrial compute, achieving this through extreme vertical integration.

The announcement follows recent developments in orbital AI, such as Google’s experimental satellite testing Gemini models and Satlyt’s push for onboard inference. Another key player, Starcloud, currently valued at $2.3 billion, is also working to mitigate the power and cooling constraints of ground-based data centers.

While ambitious, the economic viability of orbital data centers remains unproven. Current evidence relies on vendor-reported roadmaps and funding status, lacking independent verification of large-scale commercial operations or confirmed cost advantages over traditional infrastructure.

Sources :newcomer.co

Recherche

Dave Chilton, Author of 'The Wealthy Barber', on Canadian Personal Finance: Retirement Planning Gaps and the Reality of AI Advisors

Matériel
2026-10-29 17:30 GMT+8

Dave Chilton, a well-known Canadian personal finance author, recently appeared on episode 433 of the Rational Reminder podcast. Chilton is the author of the bestselling book "The Wealthy Barber," which has sold over two million copies and helped millions of Canadians manage their money. He previously served as an investor on CBC's entrepreneurship show "Dragons' Den" and launched his own namesake personal finance podcast in 2024. As he approaches the end of his career, Chilton used this episode to reflect on the evolution of the personal finance advisory landscape in Canada.

According to the official topic list published for the episode, the conversation explored several core issues. These included why retirement income planning has remained an underserved area of financial advice, how to evaluate different advisor fee models, and why strategies like the RRSP (Registered Retirement Savings Plan) meltdown can be valuable for some people but inappropriate for others.

Beyond traditional financial planning topics, Chilton also expressed concerns about rising living costs, high investment fees, and the potential economic disruption from artificial intelligence, while explaining why he remains optimistic about the future of financial planning. For readers interested in wealth management industry trends and consumer behavior shifts, this episode offers a window into how a veteran practitioner assesses current industry shortcomings and technological impacts.

Sources :rationalreminder.ca

Recherche

AI Models Find Critical PHP Flaws in Hours, Triggering Global Website Shutdowns and German BSI Orange Warning

Sujet · PHP2Shell漏洞Structurel
2026-10-29 12:55 GMT+8

On Wednesday afternoon, numerous websites worldwide went offline as a precaution after two security flaws collectively named “PHP2Shell” became known. Identified as CVE-2026-109618 and CVE-2026-109619, these vulnerabilities each carry a risk score of 9.2 on the ten-level CVSS scale. The German Federal Office for Information Security (BSI) issued an official cyber security warning at the second-highest level, “Orange,” advising operators to temporarily disconnect affected web servers from the internet if necessary. According to Futurezone, government agencies in Finland and the Netherlands, along with parts of the UK’s BBC, also took similar measures.

The core of this incident lies in how artificial intelligence has changed the speed of vulnerability discovery. According to technical analysis by Wiz, these are four related memory safety flaws in the PHP runtime environment, dating back to version PHP 5.4 over a decade ago. Attackers can trigger a “Use-After-Free” mechanism via manipulated requests, allowing remote execution of malicious code or theft of passwords and API keys without authentication. While Quarkslab identified signs of one flaw in 2024, it remained unfixed due to the lack of a known attack path. On August 11, 2026, Calif and OpenAI first demonstrated its exploitability using the GPT-5.6 Sol model. Subsequently, the Wiz team independently reproduced the flaw family and constructed full attack chains using GPT-5.6 Sol and a pre-release version of Google DeepMind’s Gemini 4 Argon.

This process demonstrates that complex vulnerability chaining, which previously required months of manual analysis, can now be completed by large language models within hours. This capability compresses the reaction time window for both defenders and attackers. As PHP underpins approximately 70% of all websites globally (according to Wiz estimates), flaws in its foundation directly impact mainstream systems like WordPress and Drupal. Although OpenAI coordinated disclosure through its “Patch the Planet” initiative and pledged funding for open-source security, the premature leak of technical details forced the PHP development team to release emergency patches on Wednesday. Administrators must immediately upgrade to versions such as PHP 8.5.12 and 8.4.27, or adopt temporary mitigations like disabling upload progress indicators if updates are not possible.

Sources :t3n.de

Recherche

Valve Says AI Cloning Is Not a New Problem, Suggests Independent Developers Compete via Community and Iteration

Sujet · Valve AI克隆政策Matériel
2026-10-29 16:15 GMT+8

At the Steam Indie Connect developer event in Busan, South Korea, Valve's Steam business team addressed a concern among independent developers: what is the platform's plan regarding AI cloning? Tom Giardino responded that Valve does not wish to act as a gatekeeper and will not dictate which tools developers may use; whether or not to use AI is up to each studio. However, this does not mean the platform is entirely hands-off. Since early 2024, Steam has required developers to disclose whether and how they have used AI, and Valve has stated it will follow up on reported extortion cases.

Giardino offered two lines of defense: building player trust in the studio and continuously updating the game. He called the former the first line of defense but immediately added that it takes time and is insufficient on its own. The latter is equally important. He drew an analogy with Valve's own history of handling piracy: rather than installing aggressive anti-piracy measures, Valve has always tried to make the legitimate version the best one. Giardino's logic is that a single game mechanic can be easily copied, but clones cannot catch up if features and creativity are constantly added through updates.

The premise of this logic is that the original creator's speed advantage comes from continuous creation, not from a one-time release. Giardino believes AI has merely accelerated existing plagiarism phenomena rather than creating new problems. Good demo concepts were being imitated even before AI appeared; AI simply made the process faster.

Regarding complaints about the large number of low-quality AI games at Next Fest, another member of Valve's Steam business team, Marc Wilkinson, explained the mechanism: Steam distributes demos almost randomly during the first two days of the event, with personalized recommendations only taking effect afterward. Wilkinson noted that many complaints about the flood of AI slop games targeted only these first two days. The purpose of random distribution is to prevent developers from gaining visibility advantages by registering early, giving every game a fair chance. Valve accepts the cost that low-quality games will therefore also be randomly showcased.

Notably, Giardino's response to extortion emails was that he had heard of such cases for the first time, viewing them more as a law enforcement issue than a game platform issue, though Valve would follow up on reported cases. The Korean two-person studio YNOT withdrew the demo for its cooking game Spookitchen after discovering an AI clone, and developers of another strategy RPG reportedly received an extortion email demanding $200. These cases illustrate that for small teams, the threat of AI cloning is not an abstract acceleration of competition, but a concrete attack under time pressure.

Valve's position can be summarized as follows: the platform provides distribution infrastructure and transparency requirements, but no active protection. The burden of defense lies with the developers. For teams with an established community base and continuous development capabilities, this may be reasonable advice; however, for small studios just releasing their first demo with no accumulated player trust, the time window for this path may be wider than the speed at which clones are released.

Sources :heise online

Recherche

French University Study: Mainstream AIs Fail to Reliably Detect Other AIs in Reverse Turing Tests

Sujet · AI身份验证可靠性争议Matériel
2026-10-29 12:20 GMT+8

According to a report by TechNews, a new study from Université Paris Cité in France indicates that when seven mainstream chatbots were placed in the position of "interrogators" attempting to distinguish whether their conversation partner was a human or another AI, their performance was almost completely inaccurate.

The experiment involved models including ChatGPT 4.5, Claude 3.7 Sonnet, Gemini 2.5 Pro, and Grok 3 questioning hidden-identity subjects for up to ten rounds. The results showed that out of 98 tests against AI subjects, the AI judges successfully identified only 3; if the subject AI used more complete persona settings, the identification rate dropped to zero. In contrast, when facing human subjects, these models correctly judged 123 out of 140 tests.

The research team analyzed that the errors stemmed primarily from expression style rather than content itself. The few AIs that were detected often revealed themselves through overly polished and structurally complete answers; when AIs responded in a more natural, concise manner (with average word counts close to humans), these cues nearly disappeared. Furthermore, AI judges rated the "human-likeness" of AI subjects higher than that of actual humans, showing that machines not only struggle to catch machines but are even more likely to mistake them for humans.

This result highlights that using AI to detect AI is currently not a reliable tool for real identity verification. The study reflects limitations in mimicry capabilities within specific conversational contexts, rather than whether AI possesses true understanding.

Sources :TechNews 科技新報

Recherche

TrendForce: AI Inference Demand Extends High-Capacity MLCC Shortage to H1 2027; Murata Expansion Delayed Until 2028

Sujet · AI推理高容MLCC短缺Matériel
2026-10-29 13:55 GMT+8

Market research firm TrendForce reports that as AI applications shift from model training to inference and AI agents, demand for CPUs is rising. Since each CPU requires power modules and DDR5 memory, the usage of high-capacity Multi-Layer Ceramic Capacitors (MLCCs) around motherboards has doubled, leading to particularly tight supply.

With NVIDIA's Vera Rubin platform ramping up shipments in Q4 2026 and AMD's MI450 starting to pull high-capacity specialized components, ODMs expect Q1 and Q2 2027 to be the peak of the high-end MLCC shortage. TrendForce estimates this gap will persist into the first half of 2027.

Meanwhile, Japanese electronic component manufacturer Murata admitted that current capacity is "far from sufficient" and plans to accelerate expansion, but new capacity will not come online until after 2028. This indicates that short-term supply bottlenecks are difficult to alleviate, leaving pricing power with sellers.

In contrast, AI applications occupying production capacity have driven up prices for memory and other components, suppressing consumer electronics demand due to higher end-user prices. Surveys show that predicted order volumes for major laptop ODMs in Q4 2026 fell by an average of 10-15% year-over-year, while smartphone brands also reduced shipments of mid-to-low-end models.

Sources :TechNews 科技新報

Recherche

US Cloud Provider Akamai Launches AI Agent Sandbox and Inference Services to Enhance Code Execution Security

Sujet · AkamaiMatériel
2026-10-29 13:59 GMT+8

Akamai Technologies, a US-based cloud computing, content delivery, and cybersecurity provider, announced two new AI infrastructure services on October 27 (local time): "Akamai Inference," an AI model operations service, and "Akamai Agent Sandbox," an execution environment for AI agents. Both services are currently available in beta.

The key highlight of this launch is the "Agent Sandbox," designed to mitigate security risks when AI agents autonomously write and execute code. Traditionally, AI agents executing user-requested generated code pose a risk of accessing unauthorized systems or data. Akamai employs WebAssembly-based isolation technology to separate the AI agent's runtime environment from the execution environment of the generated code, strictly limiting the computational resources, memory, external network connections, and system functions the code can use.

The other service, "Akamai Inference," allows enterprises to deploy and use AI models without managing the underlying infrastructure. The service offers an OpenAI-compatible API and supports 12 open-weight models, including NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b, and Google Gemma 4. Additionally, Akamai expanded its AI infrastructure by adding NVIDIA Blackwell Ultra GPU clusters and is preparing data centers for support of the Vera Rubin platform in 2027.

Adam Karon, Chief Operating Officer of Akamai Cloud, stated that customers want to build, run, and protect AI on a single platform. By adding inference and execution layers alongside its existing security portfolio, Akamai aims to provide enterprises with a full-stack protection solution from model hosting to agent execution.

Sources :미래를 보는 창 - 전자신문

Recherche

Taiwan Optical Firm Landmark Posts Record Q3 Profit, Plans Sixfold Capacity Expansion by 2028

Matériel
2026-10-28 18:08 GMT+8

Landmark Optoelectronics, a Taiwan-based optical communication component manufacturer, held its third-quarter earnings call on October 28, 2026, reporting strong results driven by high-speed transmission demand from AI data centers. Consolidated revenue for the quarter reached NT$1.562 billion, up 28% quarter-over-quarter and 183% year-over-year. Gross margin rose to 58.56%, an increase of approximately 1.1 percentage points from the second quarter. Net income after tax was NT$621 million, up 33% quarter-over-quarter and 5.28 times year-over-year, resulting in a single-quarter EPS of NT$6.12, a historical high. Cumulative EPS for the first three quarters stood at NT$13.88.

Product mix analysis shows that DataCom-related products accounted for approximately 80% to 85% of total revenue, with the majority derived from silicon photonics applications. As AI servers and data centers continue to expand, demand for high-speed optical components remains robust, boosting both shipment volumes and profitability. Looking ahead to the fourth quarter, Landmark expects revenue growth of 15% to 20% quarter-over-quarter, noting that key customers have already begun placing long-term orders exceeding six months, indicating sustained momentum.

To meet future growth demands, Landmark's board-approved capital expenditure plans over the past year have accumulated to more than NT$8 billion. The company provided explicit guidance that output volume in 2027 is expected to increase 2.5 times compared to 2026 (approximately 150% YoY), and total output in 2028 is projected to increase another 2.5 times compared to 2027. This aggressive expansion plan reflects management's confidence in the long-term AI optical interconnect market but also implies that equipment depreciation and personnel costs will gradually impact financial statements. CFO Yang Ji-yu stated that the company aims to keep depreciation expenses within 8% to 10% of revenue and does not expect profit margins to rise indefinitely without limit.

Regarding technology roadmaps, Landmark indicated that 1.6T silicon photonics products have entered mass production, while higher-power silicon photonics products for 3.2T, NPO (Near-Packaged Optics), and CPO (Co-Packaged Optics) applications are under development and validation with customers. Next-generation CPO and LDTC products for mobile applications are also progressing according to plan. These developments suggest Landmark is positioning itself to capture opportunities in higher-speed and more advanced packaging forms of optical communications.

Sources :TechNews 科技新報

Recherche

Mercedes-Benz Announces Shift to Local Sourcing in China, Starting with Electric E-Class

Sujet · 奔驰中国业务转型Matériel
2026-10-29 07:51 GMT+8

German automaker Mercedes-Benz Group CEO Ola Kaellenius announced on October 28 that the company will fundamentally change its business model in China, which has relied on producing vehicles locally while supplying components from Europe for the past two decades. The core initiative involves significantly expanding local component procurement to lower production costs and enhance pricing flexibility, aiming to address intensifying competition and declining sales in the Chinese market.

According to the newly released third-quarter results, the automotive division's adjusted operating profit fell to €560 million, a decrease of approximately 51% from €1.139 billion in the same period last year. The adjusted return on sales, a key metric for automotive profitability, dropped to 2.5% from 4.8% a year earlier. Citing fierce competition in China and expected increases in raw material costs in the fourth quarter, Mercedes-Benz lowered its full-year adjusted return on sales forecast for the automotive division from 3% to 5% down to 2% to 3%.

Kaellenius stated that future vehicles manufactured in China will source almost all components locally. The first model to implement this new strategy is expected to be the electric E-Class, scheduled for launch in the second half of 2027. Facing price wars, the CEO emphasized that Mercedes-Benz would not drastically cut prices to gain market share, but instead seek a balance between volume and profit, adjusting capacity at Chinese factories based on market demand.

Data indicates that Mercedes-Benz delivered 297,000 vehicles in China during the first nine months of the year, a 29% year-on-year decline, representing only about half the volume of the same period in 2023. China's share of Mercedes-Benz's global passenger car sales has fallen from roughly one-third several years ago to about one-fifth currently. Despite the weakness in the Chinese market, the company stressed it will not exit China and plans to offset losses by expanding its market share in the United States and other Asian markets. As of the end of September, net liquidity for the automotive and van business stood at €31 billion, indicating a relatively robust overall financial position.

Sources :中央社 CNA

Recherche

Activist Investor Nelson Peltz's Trian Proposes $1.4 Billion Take-Private Bid for Wendy's, Shares Jump 15%

Sujet · 温蒂汉堡私有化案Matériel
2026-10-29 13:40 GMT+8

Trian Fund Management, led by prominent activist investor Nelson Peltz, has joined forces with BlueFive Capital, Sixth Street Partners, and Wendy's franchisee Flynn Group to propose a total acquisition value of approximately $1.4 billion to take US fast-food chain Wendy's private. According to documents filed with the U.S. Securities and Exchange Commission (SEC), the investment team proposes acquiring Wendy's shares at $7.35 per share, representing a premium of about 20% over the previous closing price. If the transaction is completed, Wendy's will initiate delisting procedures.

Trian argues that Wendy's has faced operational difficulties in recent years and urgently needs long-term capital support to drive transformation. As the company's largest shareholder, Trian and its affiliates currently hold approximately 16.25% of Wendy's equity. Taking the company private would help alleviate short-term earnings pressure, allowing management to focus more on brand reform and strategic adjustments. Affected by reduced consumer spending on dining out and intense competition in the fast-food market, Wendy's same-store sales have declined for six consecutive quarters, and its stock price has fallen by approximately 65% over the past two years.

Boosted by this acquisition news, Wendy's shares surged 15.15% on October 28, closing at $7.07. This market reaction indicates investor optimism about a potential high-premium exit, while also reflecting a lack of confidence in the company's ability to turn around independently.

Sources :TechNews 科技新報

Recherche

South Korean Court Dismisses Tesla FSD Refund Lawsuit, Ruling No Promise of Full Autonomy

Sujet · 特斯拉紧急驶离功能Prise rapide
2026-10-29 16:14 GMT+8

According to Invezz, the Seoul Central District Court dismissed a lawsuit seeking refunds for Tesla's Full Self-Driving (FSD) software on Thursday. The court ruled that Tesla had not promised fully autonomous vehicles, thereby negating the claim of breach of contract. This verdict removes an immediate refund threat in the South Korean market but does not establish that Tesla has achieved Level 5 autonomy, which requires no human intervention, nor does it resolve similar disputes in other regions.

Plaintiff owners argued they paid for self-driving capabilities that Tesla failed to deliver. Judges stated that Tesla supplied some contracted features, including Autopilot functions, and that additional capabilities depended on technological advances and regulatory approvals. For investors, this legal victory alleviates specific compliance pressure in one market, but Tesla's financial fundamentals remain strained: operating margins fell to approximately 2.6% in the third quarter, with operating profit declining sharply due to increased spending on AI and robotics development.

Sources :invezz.com

Recherche

Amazon Launches 16-inch Echo Show with Fire TV Integration, Diverging from Apple's HomeView Strategy

Prise rapide
2026-10-29 13:11 GMT+8

US technology giant Amazon launched five new smart devices on October 28, 2026, including the redesigned Echo Show 16. The new model features a 16-inch Full HD screen and integrates the full Fire TV experience, supporting streaming services like Netflix and Disney+ with an included remote control. This design signals Amazon's push to position smart displays as "second TVs" rather than just voice assistant terminals.

In contrast, Apple's first smart display, HomeView, launched earlier this month with a 7-inch screen priced at $299, focusing on lightweight interaction via Siri AI and Apple Intelligence. Analysis by CNET notes that the two companies have taken opposite hardware paths: Amazon prioritizes larger visual real estate and deeper video content integration, while Apple favors a compact desktop companion form factor.

Pricing reflects this divergence, with the standard Echo Show 16 starting at $499.99 and the anti-glare version at $569.99, significantly higher than HomeView. Amazon stated that the Echo Show line is its fastest-growing product family, with annual growth exceeding 25%. The launch also includes the camera-equipped Echo Studio, the updated Echo Dot, and the Echo Time smart alarm clock, all deeply integrated with Alexa+ generative AI features such as visual recognition and sleep monitoring.

Sources :TechOrange 科技報橘

Recherche
Suivant →
Connexion / Inscription
LectureRechercheRadarCadre d'investissement