LectureRechercheRadarCadre d'investissement
Connexion / Inscription
ToutTechnologieProduitsInvestissementBourseAutres

Lecture

2026-10-29198 publications

Actualiser

Voice AI Infrastructure Firm Deepgram Announces Acquisition of Pipecat Maker Daily, Pledges Open Source and Launches New Protocol

Sujet · DeepgramMatériel
2026-10-29 21:00 GMT+8

US-based voice AI infrastructure company Deepgram announced on October 29, 2026, that it is acquiring Daily, the developer and maintainer of the popular open-source voice AI framework Pipecat. The companies did not disclose the deal amount but confirmed that Pipecat will remain open-source.

Pipecat is an orchestration framework that manages voice and video streams along with components like speech-to-text (STT), large language models, and text-to-speech (TTS) to reduce latency. Daily itself provides hosted infrastructure for voice and video calls. This acquisition is expected to give Deepgram deeper expertise in real-time multimodal infrastructure and expand its capabilities beyond pure voice.

Both companies are alumni of the 2016 Y Combinator batch. According to PitchBook data, Daily had previously raised over $62 million from investors including Tiger Global and Highfield Capital; Deepgram reached unicorn status earlier in 2026 with a $130 million Series C round.

Kwindla Kramer, co-founder of Daily, stated that they chose Deepgram due to CEO Scott Stephenson's commitment to keeping Pipecat permanently open-source and vendor-neutral, as well as the matching engineering cultures. Stephenson noted that the move aims to open up the ecosystem so developers can use the best models for their specific use cases, not just Deepgram's own products.

Alongside the acquisition, Deepgram launched the Real-Time Context Standard (RCS), an open protocol co-developed with the Pipecat team. The protocol defines metadata such as whispers and instructions within audio, as well as response guidance, aiming to standardize how voice agents are built.

Sources :TechCrunch

Recherche

STMicroelectronics Reports Q3 2026 Results: Revenue Up 17.9% to $3.76 Billion Driven by AI Data Centers and Satellites

Sujet · 意法半导体业绩Matériel
Vérifié 2026-10-29 21:09 GMT+8

Global semiconductor manufacturer STMicroelectronics N.V. released its financial results for the third quarter ended September 26, 2026, on October 29, 2026. The report shows net revenues of $3.76 billion, a 17.9% increase year-over-year, exceeding the midpoint of its previous business outlook. Gross margin reached 37.3%, an improvement of 410 basis points from 33.2% in the same period last year. Operating income stood at $367 million ($429 million on a non-U.S. GAAP basis).

CEO Jean-Marc Chery attributed the outperformance primarily to higher revenues in Communication Equipment and Computer Peripherals (CECP) and Personal Electronics. Demand from AI data centers and low-earth-orbit (LEO) satellite communications was highlighted as a key driver. Management expects approximately $1.3 billion in AI data-center revenue for 2026 and projects this segment's revenue to more than double in 2027. For LEO satellite communications, 2027 revenue is expected to meet or exceed the record levels seen in 2026, with capacity rather than demand potentially limiting growth.

Looking ahead, STMicroelectronics forecasts fourth-quarter net revenues of $4.15 billion at the midpoint, representing sequential growth of 10.5% and year-over-year growth of 24.7%. Gross margin is expected to be around 39.0%. Based on this outlook, full-year 2026 revenue is projected to reach approximately $14.5 billion, up 22.8% year-over-year. Despite uncertainties such as global trade tariffs, the company remains focused on improving profitability through manufacturing footprint restructuring and cost optimization.

Sources :Morningstar, Inc.

Recherche

DoorDash Announces $17 Billion Acquisition of Instacart at $65 Per Share, a 26% Premium

Structurel
Vérifié 2026-10-29 20:56 GMT+8

US food delivery leader DoorDash (NASDAQ: DASH) announced on October 29, 2026, that it has entered into a definitive agreement to acquire grocery e-commerce platform Instacart (NASDAQ: CART). Under the terms, DoorDash will pay $65.00 per Instacart share, valuing the company's equity at approximately $17 billion. This price represents a 26% premium to Instacart's closing price on October 22, 2026, the last trading day before media reports of potential talks emerged.

The transaction consideration will be at least 75% cash, with Instacart stockholders able to elect cash, DoorDash stock, or a combination, subject to a cap ensuring stock does not exceed 25% of the aggregate consideration. DoorDash intends to fund the cash portion with existing cash and debt financing, having secured a 364-day senior unsecured bridge loan commitment from JPMorgan Chase Bank. The deal is expected to close in late 2027 or the first half of 2028, contingent upon shareholder approval and regulatory clearances, including the expiration of the Hart-Scott-Rodino Antitrust Improvements Act waiting period.

DoorDash CEO Tony Xu stated that the combined entity would leverage DoorDash's delivery network and Instacart's grocery technology to help local retailers compete online. Instacart CEO Chris Rogers emphasized that the merger would create new opportunities for retail partners and provide a more seamless experience for households. However, the American Economic Liberties Project has called on the US Federal Trade Commission (FTC) to block the merger, arguing it would harm businesses, workers, and consumers.

Sources :company.instacart.com

Recherche

Indian SaaS Firm Zoho Reports 54% Revenue Growth in Taiwan, Launches Self-Built LLM-Powered AI Agents

Sujet · Zoho AI代理平台Prise rapide
2026-10-29 20:33 GMT+8

Indian collaboration software veteran Zoho held its first media briefing in Taiwan on October 29, 2026, announcing plans to expand its product promotion scale locally. Gibu Mathew, Managing Director of Zoho Asia-Pacific, revealed that the company's Taiwan market revenue grew by 54% over the past year, with more than 1,000 Taiwanese small and medium-sized enterprises (SMEs) now adopting Zoho services.

The core of this launch includes three enterprise-grade AI products: Zia Chat, a conversational interface driven by Large Language Models (LLMs); Zia Agents, an AI agent platform capable of executing tasks within workflows; and Zoho One, a one-stop platform integrating over 50 applications. Mathew emphasized that all 60+ Zoho applications are developed and maintained in-house, offering lower security risks compared to enterprises switching between different software brands.

Technically, Zia Chat supports ChatGPT, Claude, Gemini, and three self-developed small parameter LLMs from Zoho (Zia SLM 1.3B, 2.6B, 7B). Due to insufficient Traditional Chinese support in Zoho's self-built LLMs, Taiwanese users can choose their preferred third-party models without being forced into a lock-in. The Zia Agents platform provides a no-code/low-code development environment and an Agent Marketplace, allowing enterprises to build or deploy pre-built AI agents such as HR Agents and Sales Coach Agents.

Sources :iThome

Recherche

US Nominal Consumer Spending Rises 0.5% in September While Real Growth Stalls

Matériel
Vérifié 2026-10-29 20:55 GMT+8

The U.S. Bureau of Economic Analysis (BEA) released data on October 29, 2026, showing that personal consumption expenditures (PCE) increased by $113.1 billion, or 0.5 percent, in September 2026. However, after adjusting for price changes, real PCE rose only $2.6 billion, an increase of less than 0.1 percent at a monthly rate. This significant divergence indicates that the apparent strength in consumer spending was driven almost entirely by higher prices rather than an actual expansion in the volume of goods and services purchased.

During the same period, personal income increased by $54.5 billion (0.2 percent), and disposable personal income rose by $52.7 billion (0.2 percent). The personal saving rate stood at 3.8 percent. On the price front, the PCE price index rose 0.5 percent from the preceding month, while the core PCE price index excluding food and energy increased 0.3 percent; core PCE rose 3.1 percent year-over-year.

For investors, the gap between nominal and real consumption is a critical metric for assessing revenue quality. If retail or service companies' top-line growth relies primarily on pricing power rather than volume gains, its sustainability faces challenges from demand elasticity. This data suggests underlying consumer demand may be softer than headline nominal figures imply, requiring further observation alongside upcoming employment and income data to determine if consumption momentum is genuinely weakening.

Sources :bea.gov

Recherche

US Computer Science Enrollment Plummets as AI Displaces Entry-Level Coding Jobs

Sujet · 美国CS入学人数下滑Matériel
2026-10-29 13:33 GMT+8

According to preliminary data released Thursday by the National Student Clearinghouse, US colleges saw a significant exodus of students from computer and information science departments this fall. Undergraduate computer science enrollment declined by more than 12%, while graduate program enrollment dropped by more than 10%. This trend builds on last year's drop, reversing years of growth for computer science programs.

The report notes that the rise of artificial intelligence has made it increasingly difficult for computer science graduates to find jobs in recent years, as companies eliminate many entry-level coding roles. A Bureau of Labor Statistics report from August found that computer and mathematics jobs would feel the most impact from AI. In just two years, computer science undergraduate programs have lost 18.4% of their students, while graduate programs have lost 27.1%.

"Those are just remarkable declines for a field that until recently was one of the fastest growing areas that we'd seen in higher education," said Matthew Holsapple, the National Student Clearinghouse's senior director of research. He added that students are identifying pathways beyond the traditional route, with affordability likely being an important factor.

More broadly, undergraduate enrollment at US universities flattened this fall, rising by only 0.6%, marking a fourth year of growth but a slowdown from the 2.4% increase in fall 2025. Graduate program enrollment fell by 1.6%, driven by a 3.7% loss in master's programs. Community colleges were the only bright spot, seeing a 1.2% bump in undergraduate enrollment.

Sources :business-standard.com

Recherche

Taiwan's Ocean Affairs Council and National Cheng Kung University Complete 20-Drone Swarm Simulation Using MARL for Maritime Surveillance

Sujet · 台湾海委会无人机群计划Prise rapide
2026-10-29 11:14 GMT+8

According to TechNews, the Taiwan Ocean Affairs Council is collaborating with National Cheng Kung University (NCKU) to develop an autonomous drone swarm system in response to China's grey-zone tactics around Taiwan. The project has successfully completed simulations involving two swarms totaling 20 drones, aiming to use AI for autonomous task allocation, route planning, and flight coordination to expand the Coast Guard Administration's surveillance reach.

Executed by Associate Professor 闕志哲's team from NCKU's Department of Aeronautics and Astronautics, the research integrates artificial intelligence, flight mechanics, and cooperative control technologies. The team implemented Multi-Agent Reinforcement Learning (MARL) to enable the drones to self-organize and reduce the need for individual manual control. To ensure formation safety, the control algorithms incorporate three-dimensional distance constraints with a minimum vertical separation of 100 feet (approx. 30 meters), along with dynamic obstacle avoidance capabilities.

The simulations focus on three VTOL fixed-wing drone models: JUMP 20, V-BAT, and Flexrotor. Notably, the V-BAT, developed by US defense tech firm Shield AI, was recently delivered to the Coast Guard Administration; it features over 12 hours of endurance and is suitable for long-duration maritime patrols. While the simulation phase is complete, the system remains in the verification stage. Future steps include Hardware-in-the-Loop (HIL) testing, prototype integration, and actual maritime flight trials to confirm reliability under various sea states and weather conditions.

Sources :TechNews 科技新報

Recherche

GitHub says AI code surge drove monthly commits to 7.4 billion, requiring 4M extra CPU cores and 160PB storage

Sujet · AI规模CI重构Matériel
2026-10-29 18:30 GMT+8

GitHub, the world's largest code repository, announced that it added 4 million CPU cores, 160 petabytes of storage, and 50 terabits per second of networking capacity over the course of this year to manage the influx of AI-generated code, according to a Tom's Hardware report on GitHub's technical report. Monthly code commits on the platform grew sixfold year-over-year.

GitHub's technical report provides specific monthly figures: commits stood at 1.4 billion in April 2026, climbed to 2.9 billion in August, and reached 7.4 billion in September. The total number of code pushes over the past 12 months hit 17.9 billion, exceeding the combined totals of 2024 and 2025. Merged pull requests also increased fourfold.

The company attributes this surge to the adoption of agentic AI. By September 2026, more than 40% of pull requests were initiated by an agent, up from 17% in March 2026. Repositories that start using AI generally see eight times the activity compared to handcrafted operations.

This explosion in code activity places disproportionate strain on CI/CD pipelines. Every commit or merged pull request triggers automated checks such as linting, unit tests, and integration tests. GitHub has had to reengineer its backend to decouple storage from compute as much as possible, but given the rapid demand growth, further hardware upgrades are likely required soon.

Sources :Tom's Hardware

Recherche

Video Intelligence Firm Solink Raises Over $100M to Expand AI Vision Applications

Sujet · Solink融资Prise rapide
2026-10-29 20:00 GMT+8

Video intelligence company Solink Corp. announced on October 29, 2026, that it has raised more than $100 million in new funding. Headquartered in Ottawa, Canada, and founded in 2009, the company initially sold cloud video surveillance services. Its software now uses artificial intelligence to extract business insights from security cameras already installed by retailers, restaurants, and warehouse operators. According to the company, approximately 500,000 cameras across more than 35,000 customer locations in 65 countries currently feed into its platform.

The round was led by Toronto-based private equity firm Vertu Capital. Goldman Sachs Group Inc. contributed a "material amount" through its Growth Equity at Goldman Sachs Alternatives division. Golub Growth and Comcast Ventures joined as new investors. This marks Goldman's second backing of Solink, having previously led the company's $60 million Series C round in July 2023. Including this raise, total investor funding into Solink now exceeds $198 million.

Solink stated the new capital will fund its roadmap for what it calls "AI vision intelligence." In February, the company introduced AI agents for perimeter security and loss prevention, which run on edge devices without requiring proprietary cameras. In March, its VerifEye command center, sold to security operations centers, integrated vision language models aimed at filtering false alarms before human operators view them. Co-founder and CEO Mike Matta described the current moment as a "watershed" for physical businesses, stating the goal is to help customers reduce losses and uncover revenue using existing camera infrastructure.

Solink did not disclose a valuation or provide details on the strategic opportunities part of the funds are intended to support.

Sources :SiliconANGLE

Recherche

US Data Center Equipment Maker Vertiv Reports 61% Q3 Revenue Growth, Doubles Organic Orders, Raises FY2026 Guidance

Matériel
2026-10-29 11:30 GMT+8

Vertiv Holdings Co. (NYSE: VRT), a US-based provider of critical digital infrastructure, reported its third-quarter financial results on October 28, 2026, highlighting significant growth driven by the artificial intelligence infrastructure boom. The company's revenue reached $4.31 billion, a 61% year-over-year increase, surpassing Wall Street analyst estimates of $3.76 billion. Adjusted earnings per share came in at $2.21, up 78% year-over-year, beating the consensus estimate of $1.82.

A key indicator of demand strength was the more than 100% year-over-year growth in organic orders, which exclude the impact of acquisitions and currency fluctuations. This reflects expanding global demand for data center power management and cooling systems. By region, revenue in the Americas grew 75% year-over-year, Asia-Pacific rose 50%, and Europe, Middle East, and Africa increased 23%.

Based on current performance, Vertiv raised its full-year 2026 revenue guidance from the previous range of $13.8–$14.2 billion to $14.45–$14.85 billion. Full-year adjusted EPS is now projected between $7.10 and $7.20, with a midpoint of $7.15 exceeding the prior Wall Street estimate of $6.73. Giordano Albertazzi, CEO of Vertiv, stated that the largest infrastructure boom in history continues to expand, with unprecedented speed in data center demand development.

Vertiv primarily provides power management, thermal management, and related services, and is a liquid cooling technology partner for NVIDIA. Following the earnings release, Vertiv's stock price surged 16.48% on October 28, closing at $242.94. Despite this rally, the share price remains below its historical high of $379 recorded in May.

Sources :TechNews 科技新報

Recherche

AI Evaluator Artificial Analysis Updates Index v5 with 60% Private Tasks to Curb Benchmark Gaming

Matériel
2026-10-29 19:28 GMT+8

Independent AI evaluation firm Artificial Analysis has updated its core metric, the Intelligence Index v5. The new version aggregates results from nine benchmark tests and, for the first time, sets 60% of the test tasks to a non-public status. This adjustment aims to counteract developers' tendency to "game" or overfit their models against public benchmarks, thereby providing a more accurate reflection of general model capabilities.

Index v5 introduces Terminal-Bench-Science (featuring 70 agentic scientific tasks) and FrontierCode 2.0 (a non-public programming benchmark by Cognition that evaluates whether code changes would be accepted by repository maintainers). It removes older tests such as GDPval-AA v2.1, SciCode, and AA-LCR v1.1. While reduced task visibility makes external verification harder, the move is designed to enhance the anti-manipulation robustness of the rankings.

Sources :heise online

Recherche

South Korean Tech Firm Kakao Releases Open-Source Kanana-2 Embedding Models for Commercial AI Agents

Sujet · Kanana-2Prise rapide
2026-10-29 09:33 GMT+8

South Korean technology company Kakao released two open-source embedding models named “Kanana-2-embedding” on the Hugging Face platform on October 29, designed to provide foundational support for personalized artificial intelligence agents (AI Agents). The models include “Kanana-2-3b-embedding” with 3 billion parameters and “Kanana-2-1.3b-embedding” with 1.3 billion parameters. Embedding models convert text semantics into numerical vectors, enabling AI systems to retrieve relevant information based on meaning similarity even when users do not input precise keywords, which is a key component for building AI agents with long-term memory and context understanding.

According to Kakao, the new models utilize its proprietary language models to improve search accuracy and operational efficiency. In internal comparative evaluations against existing Kanana series and similar public models (such as llama-nemotron-embed-1b-v2 and Qwen3-Embedding-4B), Kanana-2 demonstrated the best performance in Korean language tasks, comparable performance in English, and higher inference efficiency than Qwen3-Embedding. Additionally, the models employ MRL (Matryoshka Representation Learning) technology, allowing developers to dynamically adjust vector dimensions based on storage and processing needs to balance performance and resource consumption.

The released models follow the “Kanana Open License,” permitting commercial use under specific conditions. This continues Kakao’s strategy from July this year when it open-sourced lightweight language models, further enriching the Korean AI infrastructure ecosystem. No Byung-seok, Performance Leader at Kakao Unified Foundation Model, stated that the initiative aims to help researchers and developers apply strong Korean-capable embedding technologies to various services while continuously optimizing search performance and efficiency.

Sources :zdnet.co.kr

Recherche
Suivant →
Connexion / Inscription
LectureRechercheRadarCadre d'investissement