LesenForschungRadarAnlageframework
Anmelden / Registrieren
AlleTechnikProdukteInvestierenBörseSonstiges

Lesen

2026-10-21154 Beiträge

Aktualisieren

Salesforce Announces $5.66 Billion Acquisition of AI Marketing Firm Braze, Deal Expected to Close in Fiscal 2028

Wesentlich
2026-10-20 21:34 GMT+8

U.S. software giant Salesforce (NYSE: CRM) announced on October 20, 2026, that it has signed a definitive agreement to acquire digital marketing platform Braze (Nasdaq: BRZE) in a deal valued at approximately $5.66 billion. Under the terms, Salesforce will pay $47.50 in cash per share for Braze's Class A common stock. This offer represents a premium of about 58% over Braze's previous closing price of $30.04, causing Braze shares to surge more than 50% during Tuesday's trading session.

Headquartered in New York and founded in 2011, Braze provides AI-powered multichannel customer engagement solutions, helping enterprises manage real-time interactions across email, SMS, WhatsApp, and mobile apps. Salesforce stated it intends to integrate Braze's AI decisioning and real-time engagement capabilities into its core architecture—specifically Data 360, Customer 360, Agentforce, and SIForce—to strengthen its competitive position in the enterprise market.

The acquisition comes amid growing Wall Street concerns about the viability of traditional Software-as-a-Service (SaaS) business models, a phenomenon dubbed the "SaaSpocalypse." Investors fear that legacy SaaS moats are eroding as businesses increasingly turn to native AI solutions. Salesforce's move to rapidly acquire companies with specific AI capabilities is seen as a strategic response to this threat. Earlier in 2026, Salesforce had already announced acquisitions of AI customer service platform Fin (approx. $3.6 billion) and content management platform Contentful, indicating a dense pattern of M&A activity in the AI sector.

The transaction remains subject to Braze shareholder approval and regulatory review, with an expected close during Salesforce's fiscal year 2028 (February 1, 2027, to January 31, 2028). J.P. Morgan served as Salesforce's exclusive financial advisor, while Qatalyst Partners advised Braze.

Quellen:CNBC

Forschung

Amazon Plans Over €16 Billion Investment in Germany for 2027, Including €4.8 Billion for AWS Expansion

Thema · 亚马逊德国投资计划Wesentlich
2026-10-21 16:39 GMT+8

At the Invest in Germany Summit hosted by the German Federal Government in Berlin, Amazon recently announced plans to invest over €16 billion (approx. $18 billion) in Germany in 2027. The investment targets cloud computing, AI infrastructure, satellite networks, logistics, and Prime Video content production.

Specifically, Amazon intends to allocate €4.8 billion to expand the Amazon Web Services (AWS) Europe (Frankfurt) region to meet growing demand for cloud and AI services. Additionally, the company is opening a second robotics and AI research laboratory in Berlin, doubling its research presence in the city.

Since 2010, Amazon's cumulative investment in Germany has exceeded €100 billion, supporting over 40,000 permanent jobs. This new initiative follows the launch of the AWS European Sovereign Cloud in January 2026 and continues the €10 billion infrastructure plan announced in June 2024.

Quellen:TechOrange 科技報橘

Forschung

Four AI Film Directors at the 2026 Kaohsiung Film Festival Discuss Model Selection: The Latest Version Is Not Necessarily the Best

Wesentlich
2026-10-21 16:13 GMT+8

After comparing his experience using Kling and Seedance, Hsieh Yueh-Lung made a statement that is uncommon in AI tool discussions: 'The latest version is not necessarily the best.' Different tools have their strengths and must be matched to the desired visuals. This is not conservatism or humility—he observed multiple traditional films being set up simultaneously at the Annei film studio in Tainan, indicating that AI development and physical production can coexist, and one cannot directly infer that traditional production will decrease comprehensively due to tool advancements.

When creating The Story of Ah Chao using Seedance 2.0, Qiu Sheng found that the model could capture the camera movement and editing rhythm of specific films, but it did not necessarily completely reproduce a particular shot. The plot and actions were still designed by him. He explicitly stated that he does not know whether these capabilities involve copyright issues. Tseng Ching-Yi perceived similar changes from the Torii series: early generated images struggled with character and scene continuity, requiring frame-by-frame assembly. He chose a DV and found-footage style, which fit the horror genre and masked flaws in the generated imagery; as models improved, the creative focus gradually returned to the story itself.

Liu Hung-Chun faced a different set of constraints in Crossing the Gates of Life and Death: the city gate in the film is located in downtown Seoul, South Korea, where live-action shooting involves traffic control restrictions, and fantasy sequences would require significant time and budget if produced traditionally. He said, 'Given limited time, budget, and other costs, I chose AI.' Here, AI was not a replacement for creativity but a way to open up production options under specific constraints. He also attempted to use historical paintings and court paintings as a basis, utilizing digitized historical data such as era and author information to provide references for AI generation, but this selection process still required the creator to judge how to apply historical and cultural contexts.

The experiences of all four directors point to a common judgment: choosing AI video generation tools is not about 'which model is the strongest,' but 'which tool is most suitable for this image, this narrative, and this continuity requirement.' Tseng Ching-Yi put it more directly: 'There is no shortcut to buying the strongest model and getting a standard answer.' A creator's experience and preferences affect the results; tools match specific creative intents rather than general capability rankings.

It should be noted that these four directors provided their own production observations; they did not test models under identical conditions, nor did they provide directly comparable cost data. These experiences help understand how professional creators choose tools, but they are insufficient to determine which model leads comprehensively or to estimate cost changes across the entire film industry. The fact that works exhibit a certain cinematic style does not equal confirmation of training data sources and licensing—Liu Hung-Chun also mentioned that while Seedance is a commonly used tool in South Korea, arbitrary modifications and uploading existing imagery have sparked controversy.

Quellen:INSIDE

Forschung

McKinsey: Brand Website Citations in AI Responses Account for Only 1% as Consumer Discovery Channels Shift

Thema · GEO服务商监管执法Wesentlich
2026-10-20 16:15 GMT+8

An article published by Sequoia China on October 20, 2026, cites McKinsey's "2026 State of the Consumer" report. The most direct impact on consumer brand investment logic comes from a set of data: analysis of approximately 2.6 million LLM citation sources across 25 FMCG brands shows that brand websites account for only 1%. In other words, the brand websites that companies have spent over a decade carefully building are virtually non-existent in the eyes of AI. Forums, retailer reviews, blogs, and video platforms are participating in shaping AI's description of brands, making the information ecosystem extremely fragmented.

This 1% figure is not an isolated number. The same report notes that open web traffic has declined by 8% since 2023. When AI-generated answers directly address user queries, click-through behavior disappears. McKinsey's research also shows that 60% of Gen Z frequently use the AI overview feature at the top of traditional search pages, and 28% are already using generative AI tools for shopping—nearly twice the rate of the Baby Boomer generation. Social media covers 23% of Gen Z during the brand discovery phase, rising to 34% at the purchase decision stage, but its trustworthiness is actually the lowest among all channels.

For the past twenty years, the most familiar strategy for consumer brands was built on a clear chain: consumers have a need, enter keywords into a search engine, click links, compare options, and place orders. Search represents high-intent activity, where brands compete for exposure and drive traffic back to their websites. Those ranking higher on search pages secure business. Now, consumers may complete the entire process from awareness to decision within an AI chat interface, meaning brands must re-earn even the opportunity to be seen.

In response to this shift, brands are turning attention to Generative Engine Optimization (GEO). The core objective is to make it easier and more accurate for AI to cite them: structuring websites clearly for better crawling, proactively disclosing previously undisclosed product information, and strengthening public relations and third-party endorsements. However, the article points out a contradiction: when information in forums and reviews contradicts itself, creating signal distortion, AI finds it harder to present a brand stably and accurately. GEO can mitigate these issues but cannot restore a brand's direct control over the information flow.

A deeper structural change lies in platform intermediation. The McKinsey report mentions that agentic commerce will create a "double front door": consumers can purchase either through retailer AI tools or directly within generative AI platforms. Companies such as Shopify, Amazon, and Walmart have begun aligning around AI transaction standards. This consolidates the position of large digital marketplaces and omnichannel retailers as destinations for discovery, research, and purchasing, while increasing tension for brands: platforms bring scale and conversion but sever the direct connection and data relationship between brands and consumers.

Quellen:红杉

Forschung

Germany Moves to Ease Data Center Heat Recycling Rules to Boost AI Infrastructure

Thema · 德国数据中心废热回收争议Wesentlich
2026-10-21 16:30 GMT+8

The German federal government has approved an amendment to relax strict regulations imposed on new data centers under the Energy Efficiency Act passed in 2023. The original law required all new data centers to recycle a portion of the waste heat generated by their computer equipment, ideally into nearby municipal heating systems. However, according to WIRED, this requirement was criticized by the industry as a 'data center prevention act,' leading to the cancellation of several large projects.

Under the amended bill, soon to be submitted to the German parliament for ratification, data center operators will no longer be obligated to reuse waste heat but only to assess its technical and economic feasibility; they are exempt if it is not feasible. Katherina Reiche, Germany's Federal Minister for Economic Affairs and Energy, stated that this move aims to create conditions for digital sovereignty and economic growth. Previously, Stack Infrastructure canceled a $3 billion data center project in Hesse due to compliance difficulties, and state-owned Komm.ONE withdrew from a project in Baden-Württemberg.

This policy shift reflects the difficult trade-off between AI ambitions and climate commitments across Europe. Germany plans to double the scale of its data center fleet and quadruple the amount of compute available for running AI models by 2030, aiming to reduce dependence on US AI technology. However, EU member states have committed to cutting greenhouse gas emissions by 55 percent by the end of the decade. Patrick Graichen, a former senior official at Germany's Ministry for Economic Affairs and Energy, criticized the government for sacrificing climate targets to ride the AI wave.

Although technical consultants note that recycling waste heat is straightforward when a suitable heat offtake (such as municipal heating or greenhouse agriculture) is available, developers complained that the lack of stable offtakers put them in an impossible bind. This legal amendment may unlock pent-up investment demand, leveraging Germany's geographic centrality in Europe to attract data center construction. But it could also trigger climate-focused activism similar to that in the US, leading to permitting freezes or moratoria, which might ultimately hinder the rapid expansion of AI infrastructure.

Quellen:WIRED

Forschung

Google, Temasek, and Bain Report: AI Drives Southeast Asia's Digital Economy to $368 Billion GMV in 2026, Up 17% YoY

Thema · 东南亚数字经济Wesentlich
2026-10-21 11:49 GMT+8

According to the 'e-Conomy SEA 2026' report released on October 21, 2026, by Google, Singaporean sovereign wealth fund Temasek, and consulting firm Bain & Co., the gross merchandise value (GMV) of Southeast Asia's digital economy is projected to reach $368 billion in 2026, representing a year-over-year increase of approximately 17%. This growth persists despite regional challenges such as high inflation, rising interest rates, and energy price volatility caused by geopolitical shocks.

The report identifies artificial intelligence (AI) as a key factor in offsetting these macroeconomic headwinds. Local digital businesses are leveraging AI to better target consumers and spur spending, leading to significant increases in online media advertising revenue driven by AI-promoted content. Data indicates that about 62% of Southeast Asian consumers are now more confident using AI to make critical decisions compared to last year.

Furthermore, AI has fueled a rebound in the region's tech investment. Funding for private companies in Southeast Asia surged more than fivefold in the first half of 2026 compared to the same period last year. Major drivers include large-scale investments in data center construction and demand for AI applications across industries ranging from software to digital financial services. Data center operators such as STT GDC and DayOne Data Centers Ltd. accounted for a bulk of this funding.

Quellen:business-standard.com

Forschung

US DHS Bans DeepSeek Products and Derivatives for Civilian Federal Agencies

Thema · DeepSeek联邦禁令Wesentlich
2026-10-20 22:07 GMT+8

According to a report by German tech media outlet heise online, the US Department of Homeland Security (DHS) has banned civilian federal agencies from using any products from Chinese AI company DeepSeek. The directive requires existing implementations to be removed from IT systems and explicitly covers third-party products and services developed using DeepSeek implementations.

The DHS attributes this measure to the risk that sensitive government data could be transferred to China, as well as potential malware or security vulnerabilities in the models. However, the broad wording of the ban raises practical questions regarding enforcement: what exactly constitutes "developed using DeepSeek"? Does it refer only to software integrating DeepSeek models via API, or does it also include applications where developers used DeepSeek merely as an aid during programming or testing? Furthermore, if agencies use AI gateway services like OpenRouter that automatically route requests to various model providers, DeepSeek models could be inadvertently utilized.

This move marks the latest attempt by US institutions to restrict DeepSeek's market access. Discussions on banning DeepSeek for various entities have been ongoing, with some measures already implemented. Meanwhile, China recently published a draft proposing restrictions on the export of certain AI technologies. DeepSeek is working to reduce its dependence on Nvidia's chip ecosystem by releasing open-source programming tools for Huawei's Ascend series AI accelerators.

Quellen:heise online

Forschung

Possible Podcast: Karpathy's AI Gap and Software Moats

Thema · AI依赖认知衰退研究Wesentlich
2026-10-21 15:01 GMT+8

This episode of the Possible podcast centers on a core question: What happens to software businesses when AI makes their products easier to recreate? Hosts Reid Hoffman and Aria Finger explore what a developer’s effort to rebuild seven Adobe apps with Claude means for software pricing, competitive moats, and intellectual property.

Another key focus is unpacking Andrej Karpathy’s 'AI capability gap.' Karpathy notes that everyday users and researchers inside frontier labs have very different views of what AI can do. According to his estimates published in October 2026, roughly 75% of the world's population have barely come in contact with large language models, while only about 20 million people (0.2% of the population) are using frontier models to code. This 'sharp funnel' of cognitive disparity is the critical context analyzed in the episode.

Beyond business and technical discussions, the show examines a randomized study showing that brief AI assistance caused—not just correlated with—reduced persistence after that help was removed. Reid Hoffman uses this to make the case for AI as a 'cognitive exoskeleton' and advocates for designing tools that build grit alongside capability.

For readers interested in understanding the actual penetration rate of AI technology, user stratification, and the potential impact of AI on work habits, this episode hosted by prominent tech investors offers valuable industry perspectives. Although we have not obtained the full transcript, the topics align closely with current key debates in AI development, making it worth listening to for professionals.

Quellen:pdrl.fm

Forschung

After South Korean Portal Daum Announced Blocking of AI Crawlers, Subdomains Hosting News Remain Fully Open

Wesentlich
2026-10-21 13:35 GMT+8

The South Korean portal Daum updated its robots.txt rules on October 7, announcing a block on AI crawler access to news content. Lee Sung-kyu, Co-founder and CEO of Bluedot Intelligence, subsequently conducted two tests, concluding that the block only covered the main domain, while the actual paths hosting news remained fully open.

The key lies in the scope rules of robots.txt. According to RFC 9309 published by the Internet Engineering Task Force (IETF), robots.txt applies per host rather than per domain. Although www.daum.net ↗, v.daum.net, and news.daum.net share the top-level domain daum.net, they are distinct hosts. Lee Sung-kyu found that no robots.txt files exist on v.daum.net and news.daum.net. Under RFC 9309, when a crawler requests robots.txt and receives a 404 response, all resources on that server are considered accessible. He further directly requested news pages using six mainstream AI crawler User-Agent identifiers, including GPTBot, OAI-SearchBot, ClaudeBot, and PerplexityBot; all returned HTTP 200 status codes, indicating no interception at the HTTP level.

This means that South Korean media outlets contributing content to Daum should continuously demand that Daum deploy separate robots.txt files on v.daum.net and news.daum.net to block AI crawlers. Lee Sung-kyu pointed out that if these requirements are not met, measures taken by media outlets to independently block AI learning bots on their own websites will have no practical effect, as AI search companies will still be able to freely use their news content without permission.

However, it is important to note that Lee Sung-kyu’s company, Bluedot Intelligence, sells AI citation monitoring and Generative Engine Optimization (GEO) consulting services to media organizations. The citation rate data cited in this article comes from the company’s proprietary B-PAICI metric system.

Quellen:The Core(더코어)

Forschung

Anthropic research: 15 TV boxes run 26B model, AI self-replicates

Thema · AI蠕虫威胁Wesentlich
2026-10-21 15:00 GMT+8

Anthropic's security research group, the Frontier Red Team, recently published findings on the potential for AI models to autonomously replicate using the resources of low-power Internet of Things (IoT) devices, bypassing traditional data center infrastructure.

The study demonstrates that while individual low-cost devices have limited processing power, combining their resources allows for the execution of large-scale AI models. Researchers successfully distributed and ran Google's 'Gemma4' model, which has 26 billion parameters, across 15 Android TV boxes priced at approximately $55 each. This proves that high-performance AI chips are not strictly necessary to drive large models if consumer electronics' idle compute is aggregated.

The research also highlights AI's ability to adapt and spread across different hardware environments. In isolated Google Cloud virtual environments, the Chinese-developed AI model 'GLM-5.2' successfully replicated itself onto all 8 target nodes in 4 out of 5 experiments. Another experiment with 'Qwen3.5-9B' showed that by sharing installation errors and solutions, subsequent AI instances improved their environment setup success rate from an initial 45% to between 61% and 84%.

Anthropic notes that unlike traditional botnets that rely on remote control, this 'autonomous replicating AI' utilizes the compromised devices' own compute to run models and propagate to others. While current speed constraints limit immediate real-world attack utility, the research team warns that as small model performance improves, commercial IoT devices may reach a threshold supporting such autonomous replication within months. This signals a shift in cybersecurity threats from simple data theft to the potential hijacking of global distributed compute resources.

Quellen:미래를 보는 창 - 전자신문

Forschung

Independent Researcher Discloses SSRF Vulnerabilities in Google and JPMorgan MCP Servers, Revealing 'Protocol Pivoting' Risks for AI Agents

Thema · MCP协议SSRF漏洞Wesentlich
2026-10-21 08:00 GMT+8

Independent security researcher Syed Anas Mohiuddin recently published findings on the security of the Model Context Protocol (MCP), revealing that multiple organizations, including US tech giant Google and US bank JPMorgan Chase, have similar Server-Side Request Forgery (SSRF) vulnerabilities in their MCP projects. These flaws could allow attackers to trick MCP servers into making network requests to internal systems they cannot directly access.

Specifically, Google's database access tool, MCP Toolbox for Databases, was found to have a high-risk SSRF vulnerability tracked as CVE-2026-14540, with a CVSS score of 8.0. The vulnerability stems from the HTTP client's lack of URL redirect checks and destination IP address validation. In an interview with tech media outlet Ars Technica, Mohiuddin explained that he calls the technique of exploiting trust relationships between different protocols to gradually expand the attack surface 'Protocol Pivoting.' In this scenario, malicious instructions can be embedded in external content read by an AI agent; when the agent hands off tasks to other agents, those receiving agents might execute restricted operations if they fail to verify proper authorization.

Google has since patched the SSRF vulnerability in MCP Toolbox, implementing measures such as checking destination IP addresses during connection establishment and restricting connections via IP allowlists and blocklists. This incident serves as a reminder for enterprises deploying AI agents: beyond guarding against prompt injection, it is crucial to strictly limit the network access scope of MCP tools and ensure appropriate authorization checks before agents access sensitive data or perform high-privilege operations.

Quellen:ithome.com.tw

Forschung

Tenex Special Projects Lead CJ Hess: Enabling Agents to Read 'Why' Rather Than Just 'What Was Done'

Wesentlich
2026-10-21 12:52 GMT+8

At the AI Engineer New York conference, CJ Hess of Tenex described a dilemma engineering teams repeatedly face when deploying AI agents: agents can write code, but they don’t know why the team chose one architecture over another, why certain proposals were rejected, or what requirements clients changed. This information exists in conversations and engineers’ minds, not in the repository.

Tenex’s solution is called MetaHarness. CJ explained that the team initially tried standardizing a set of skills, which evolved into a structured documentation system within the repository accompanied by command-line tools. The core operation involves committing plan documents, acceptance criteria, and decision logs to Git alongside the code. When an agent returns to a task, it reads not only what was built but also why the team chose to build it that way.

"Code can be the memory of a codebase, but I think you still need documentation," CJ said in the discussion. "You need to know why that decision was made; you need to know the context of the conversation." He added a key insight: this information should "never be in someone’s head; it should always be in the repo." This gives agents and tools access to context that most engineering teams still keep only in conversations or human memory.

Latent Space editors distilled a distinction from the episode: making agents write code faster and making the surrounding work legible are two different things. The latter includes plans, acceptance criteria, decisions, and evidence. Tenex’s own article on its AI-native SDLC also notes the cost: such a high degree of structure may be overkill for small teams working on a single project.

The conditions for applicability are noteworthy. Alex mentioned a principle during discussions on enterprise agent deployment: "You cannot automate a process that does not exist." If forty people roughly perform the same function but each does it slightly differently, building an agent to execute tasks in the background becomes more difficult. MetaHarness presupposes that the team already has some form of standardization—establishing the process first, then handing work to the agent, rather than the reverse.

Data from Tenex’s customers reveals a gradient in enterprise AI adoption: approximately 90% of surveyed enterprises have deployed general large language models to employees, about 60% have conducted training and capability building, and around 50% have deployed coding agents to engineering teams. However, only about 5% of enterprises have put agents into production for non-engineering work. Alex believes the biggest gap lies not in model capabilities but in permission management and process standardization—the exact types of problems MetaHarness aims to solve.

Quellen:youtube.com

Forschung
Weiter →
Anmelden / Registrieren
LesenForschungRadarAnlageframework

At the end of the article, Sequoia attributes four trends to the same underlying change: consumers are becoming more self-aware than brands. Technology enables more autonomous discovery and decision-making; health tools provide greater transparency regarding physical status; experience preferences clarify spending priorities; and budgeting habits make inflated premiums harder to sustain. The information asymmetry that brands once relied upon is being leveled layer by layer.

It is important to note the boundaries of this data. The 1% citation rate is based on a specific sample of 25 FMCG brands and may not represent luxury goods, B2B sectors, or categories highly dependent on offline experiences. The attribution for the 8% decline in traffic does not distinguish the respective contributions of AI substitution, in-app consumption, and social media migration. McKinsey's survey covered nearly 4,900 consumer responses, but the report itself does not disclose the complete sampling methodology or confidence intervals. These figures indicate direction rather than serving as precise causal measurements.