2025-10-22 Briefing - alobbs.com

Today’s tech highlights include AMD’s new mobile chip with a 32MB Infinity Cache preventing DRAM bottlenecks, Google’s quantum computer achieving verifiable quantum advantage, and Tesla recalling nearly 13,000 vehicles over battery connection issues. Additionally, Google ported over 30,000 applications to Arm, while Meta cuts 600 AI roles amidst restructuring efforts.

▶️ Internet Infrastructure

AMD Strix Halo Mobile Chip Uses 32MB Infinity Cache to Prevent DRAM Bottlenecks

AMD’s Strix Halo mobile chip with 32 MB Infinity Cache maintains effective bandwidth management, capturing ~73% of traffic and avoiding DRAM bottlenecks up to 335 GB/s across workloads.

Strix Halo is AMD’s high-end mobile chip with 16 Zen 5 cores, 20 RDNA 3.5 WGP GPU, 256-bit LPDDR5X-8000 memory, and 32 MB Infinity Cache
Performance monitoring at Infinity Fabric and memory controllers reveals Infinity Cache captures approximately 73% of memory traffic
Testing shows the 32 MB cache effectively prevents DRAM bandwidth from becoming a bottleneck across various workloads, with GPU demanding up to ~335 GB/s in some cases

Linux Capabilities Enable Privilege Escalation and Detection Methods

Linux capabilities split root privileges into 40 units, allowing privilege escalation via setcap, with detection through getcap, getpcaps, and monitoring security.capability attributes.

Linux capabilities divide superuser privileges into 40 distinct units, as indicated by /proc/sys/kernel/cap_last_cap value 40
Capabilities like cap_setuid can be added to binaries (e.g., Python 3.12) using setcap cap_setuid+ep /usr/bin/python3.12, enabling privilege escalation
Enumerating binaries with capabilities can be done with getcap -r /, and capabilities are stored in security.capability attribute, viewable via getfattr

Rust Merkle Tree with Fixed Depth and Fast Proof Generation

Rust-based Merkle tree with fixed depth, incremental updates, configurable storage (sled, rocksdb, sqlite), and hash functions (keccak256, Poseidon), optimized for rapid proof generation.

Rust implementation of Merkle tree with fixed depth and incremental-only updates
Supports configurable storage backends: sled, rocksdb, sqlite
Supports hash functions: keccak256, Poseidon BN254 Circom T3
Optimized for fast proof generation and Merkle proof retrieval
Example usage includes adding leaves, retrieving root, proof, and customizing store, hash, and depth

MinIO Switches to Source-Only Distribution, Sparking Community Security Concerns

MinIO announced it is now a source-only distribution, no longer providing official Docker images for security releases, prompting community concerns over updateability and security implications.

MinIO did not release a new Docker image for the security CVE RELEASE.2025-10-15T17-29-55Z on DockerHub or quay.io as of October 18, 2025.
The project shifted to a source-only distribution model, requiring users to build their own containers.
Community members expressed concern over the lack of automatic updates and the impact on security and maintenance, especially for small deployments.

Google Flags Immich.cloud Subdomains as Dangerous in False Positive Alert

Google flagged immich.cloud subdomains as dangerous due to false positives from Safe Browsing, affecting internal and preview environments; resolution required Google Search Console review.

Google flagged all *.immich.cloud sites, including internal and preview environments, as dangerous in October 2025
The warning was issued via Google Safe Browsing, citing harmful content and deceptive practices on affected URLs
The issue impacted internal services (Zitadel, Grafana, Victoria Metrics) and the production tile server at tiles.immich.cloud; resolution involved Google Search Console review

Stalwart released a full implementation of JMAP for Calendars, Contacts, Files, and Sharing, establishing a unified, JSON-based protocol ecosystem to replace WebDAV-based technologies.

Stalwart announced full support for JMAP protocols for Calendars, Contacts, Files, and Sharing on October 22, 2025
Implements JMAP extensions: Calendars, Contacts, File Storage, Sharing
Marks Stalwart as the first JMAP server fully supporting the entire collaboration protocol family, enabling modern, JSON-based groupware

Tesla Recalls 13,000 Vehicles Over Faulty Battery Connection risking Power Loss

Tesla recalls almost 13,000 EVs over a faulty battery connection causing sudden power loss, affecting Model Y and Model 3 vehicles produced between March and August 2025.

Tesla recalls nearly 13,000 vehicles—8,000 Model Y SUVs and 5,000 Model 3 sedans—produced from March to August 2025 due to faulty battery pack connection parts.
The defect can cause sudden loss of propulsion, increasing crash risk, as reported to the NHTSA.
The issue stems from a faulty battery pack connection, potentially leading to unanticipated power loss without warning.

AWS Faces Market Shift as AI Startups Cut Traditional Cloud Spending

Internal documents indicate AWS faces a “fundamental” shift as AI startups prioritize models and inference over traditional cloud spend, risking market share amid slower growth and rising competition.

Internal documents from March and July 2025 reveal AWS is experiencing a “fundamental” shift in startup cloud spending, with AI models and tools prioritized over traditional cloud services.
Startups are delaying AWS cloud adoption, allocating more budget to AI inference, training, and developer platforms like GPU services, with Cursor’s AI spend on newer categories exceeding 90% of total cloud expenditure.
AWS’s revenue growth from AI startups lagged behind competitors in Q2 2025, with Google Cloud and Microsoft Azure growing over 30%, while AWS grew 18%; market share among AI startups declined from 33% to 30% between 2022 and 2025.

Greenland’s Tusass Partners with Eutelsat to Boost Remote Internet Access

Greenland’s Tusass partners with Eutelsat to deploy OneWeb satellites, enhancing internet access in remote areas, with Tasiilaq first to benefit; the country aims to maintain telecom control.

Greenland’s telecom company Tusass signed an agreement with French satellite provider Eutelsat to deliver internet via OneWeb satellites.
The satellites will provide faster, more stable coverage to remote towns lacking underwater cables or radio links, starting with Tasiilaq before expanding.
Tusass considered Starlink but chose Eutelsat due to trust and existing cooperation; the partnership aims to improve connectivity and support critical infrastructure.

Reddit Sues Data Scrapers Over Unauthorized Search Data Reselling

Reddit sues data scraper companies for unauthorized scraping of its site and reselling data to AI firms, amid increased demand for search data to train AI chatbots like ChatGPT.

Reddit filed a lawsuit accusing companies of scraping its website data and reselling it to AI firms, violating its ban on scraping and data access charges.
Start-up SerpApi, founded in Austin, Texas, began scraping Google search results in 2017 to assist clients with search rankings, later selling this data to AI companies post-ChatGPT emergence.
Data scrapers like SerpApi have become a significant source of training data for AI chatbots, leveraging scraped Google search information to meet the demand of AI training datasets.

China’s CR450 Bullet Train Sets New Speed Record at 453 km/h

China’s CR450 train achieved 453 km/h in pre-service testing, making it the fastest wheeled EMU globally, with 400 km/h operational speed and 600,000 km of testing planned before passenger service.

China’s CR450 bullet train reached 453 km/h during pre-service tests, surpassing CR400’s 420 km/h and Deutsche Bahn’s 405 km/h record
Operational speed will be 400 km/h; 600,000 km of testing remains before passenger service begins
Weighs 10% less than CR400 (which operates at 350 km/h); improved brakes maintain comparable stopping distances
Compared to European records: France’s TGV at 574.8 km/h (2007), UK’s High Speed 1 at 334.7 km/h, and US’s NextGen Acela at 257 km/h
Japan’s maglev train reached 650 km/h in testing, with operational max of 505 km/h; maglev lifts off at ~150 km/h for magnetic levitation
The CR450 is labeled “the world’s fastest” EMU by Science and Technology Daily, with a weight reduction and advanced braking system
China and Japan are advancing beyond wheel-based trains, with Japan’s maglev reducing Tokyo-Osaka travel time from 150 to 67 minutes

AI-driven Data Center Power Surge Sparks Aircraft Engine Repurposing and US Turbine Delays

AI power demands increase datacenter energy needs, causing gas turbine shortages; operators now repurpose aircraft engines, with US manufacturers experiencing three-year delays.

AI-driven datacenter power demand causes gas turbine shortages, prompting operators to repurpose used aircraft engines as generators.
US turbine manufacturers, including MHI, Siemens, and GE Vernova, face a three-year delivery wait due to high demand, especially in North America.
ProEnergy overhauls GE CF6 commercial aircraft engines into gas turbines for on-site power, with some turbines having up to five-year wait times and non-refundable reservation fees.

Google Ports Over 30,000 Applications to Arm Architecture to Boost Efficiency

Google has migrated 30,000+ internal applications to Arm architecture, using AI automation, to improve efficiency and reduce costs, with plans to port all 100,000+ workloads for optimized resource utilization.

Google has ported approximately 30,000 production packages to the Arm architecture, including YouTube, Gmail, and BigQuery, with plans to convert all 100,000+ applications.
The migration process involved initial efforts on top-tier workloads, utilizing AI tools like “CogniPort,” which automates build and test error fixes, achieving a 30% success rate.
Google claims Axion silicon-based machines offer up to 65% better price-performance and 60% higher energy efficiency than x86 instances, aiming to optimize workload distribution via Borg cluster manager.

Linux Kernel RFC Suggests Power Button to Cancel Hibernation Mid-Process

A Linux kernel RFC proposes a power-button interrupt to cancel hibernation mid-process, addressing hibernation’s complexity, reliability issues, and the need for better control and debugging tools.

Linux kernel RFC introduces a mechanism to cancel hibernation mid-process via power button press
Hibernation duration on Anjum’s machine is approximately 20 seconds
Hibernation support is complex, with issues like data loss, firmware updates, and swap space requirements affecting reliability

NextSilicon Maverick-2 Boosts HPC Performance with 10x GPU Efficiency

NextSilicon’s Maverick-2 HPC accelerator delivers up to 10x GPU performance in benchmarks, with 600 gigaFLOPS at 750W, using a dataflow architecture and a custom RISC-V core, deployed at Sandia National Labs.

NextSilicon’s Maverick-2 chip targets HPC workloads, not AI, achieving up to 10x performance over modern GPUs in benchmarks
In HPCG benchmark, the chip reaches 600 gigaFLOPS at 750W, matching leading GPU performance with half the power consumption
In PageRank, Maverick-2 delivers 10x higher graph analytics performance and surpasses GPU capabilities for graphs over 25 GB; achieves 32.6 GUPS, 22x faster than CPUs and 6x faster than GPUs

IBM Q3 2025 Revenue Rises 9% Driven by AI and Mainframe Growth

IBM’s Q3 2025 results show 9% revenue growth, driven by AI and mainframe sales, with the company renting GPUs from rivals instead of investing heavily in own cloud infrastructure.

IBM Q3 2025 revenue was $16.33 billion with a net income of $1.75 billion, up 9% YoY, reversing a $330 million loss in Q3 2024
IBM’s infrastructure revenue grew 15% YoY, driven by a 59% increase in mainframe revenue following the launch of Z17 machines
IBM’s AI-related projects, including $9.5 billion in generative AI work and a 45% productivity boost from “Project Bob,” contributed to profit margins and growth

▶️ Open Source

Decade-Long Dotfiles: Essential Shell Scripts for System and Media Management

Evan Hahn shares frequently used shell scripts for system, file, internet, media, and process management, enhancing productivity with tools like copy, mkcd, serveit, and notify, hosted at Codeberg.

Author has maintained dotfiles for over a decade, creating frequently used shell scripts for system, file management, internet, text processing, and media tasks
Scripts include clipboard wrappers (copy, pasta, pastas), directory management (mkcd, tempe), media controls (boop, sfx, tunes, pix), process management (each, murder, waitfor, bb), and system settings (theme 0/1, sleepybear)
Additional tools for web development, time management, text manipulation, and utility functions, with links to source code on Codeberg

Manu Criticizes AI-Enhanced Browsers as Superficial Chromium Overlays

Manu criticizes AI browsers like Atlas and Comet as superficial Chromium overlays, questioning the challenge of creating a new browser amid AI ambitions and emphasizing stagnation in browser innovation.

Manu critiques the proliferation of AI-enhanced Chromium browsers like OpenAI Atlas, Perplexity Comet, DIA, and Opera, noting they are essentially Chromium with AI overlays.
He highlights Sam Altman’s excitement about launching a basic Chromium-based AI product, questioning the difficulty of building a new browser from scratch despite ambitions for superintelligence.
The article emphasizes the lack of innovation in browser development, contrasting it with the company’s focus on AI and superintelligence.

Cloudflare Fixes Cryptographic Flaws in CIRCL FourQ Implementation After Vulnerability Report

Cryptographic flaws in Cloudflare’s CIRCL FourQ implementation allow invalid point and degenerate curve attacks due to inadequate validation, risking secret key exposure; issues fixed after March 2025 report.

Discovered cryptographic issues in Cloudflare’s CIRCL library’s FourQ elliptic curve implementation in early 2025
Reported vulnerabilities via HackerOne in March 2025, later confirmed and fixed by Cloudflare
Main issues include missing point validation, faulty comparison, and lack of cofactor and scalar multiplication validation, enabling invalid point and degenerate curve attacks

Erowid Fund Drive 2023 Raises Over 600 Donations for Drug Info and Harm Reduction

Erowid’s 2023 fund drive raised over 600 donations to support open drug information, harm reduction, and archiving, with donation options including PayPal and Bitcoin.

Erowid’s annual fund drive succeeded with over 600 donors, reaching its matching goal to support drug information, lab drug checking, harm reduction, and archiving
Donations options include PayPal links with amounts from $10 to $1,000 and cryptocurrency via Bitcoin
The website features extensive resources on psychoactive chemicals, plants, pharmaceuticals, and related cultural and legal topics

AlmaLinux 10.1 Adds Btrfs Support for Enhanced Storage and Reliability

AlmaLinux 10.1 introduces Btrfs support in the installer, enabling installation on Btrfs volumes, which provides advanced features like snapshotting and checksumming, contrasting RHEL’s deprecation.

AlmaLinux 10.1 supports Btrfs in its installer, enabling installation on Btrfs volumes instead of default XFS+LVM
Btrfs offers features like snapshotting, built-in volume management, checksumming, transparent compression, and reflinks, improving fault tolerance and performance
Red Hat deprecated Btrfs in RHEL in 2017; AlmaLinux’s support marks a strategic divergence, highlighting fragmentation in enterprise Linux distributions

Vulnerability in Rust async-tar Crate Puts Python Package Manager at Risk

A flaw in the Rust crate async-tar enables tar header parsing bypass, impacting uv’s Python package manager; most users run unpatched tokio-tar, which remains vulnerable.

Vulnerability in Rust crate async-tar affects uv Python package manager, with patched fork available but most downloads remain unfixed
The flaw exploits header parsing error in tar format, allowing hidden files, overwriting attacks, supply chain exploits, and BOM scan bypass
Affected crates include astral-tokio-tar (patched), but tokio-tar (most popular, 7 million+ downloads) remains unpatched; maintainers are unresponsive

Rivian Unveils Modular TM-B Electric Bike with 100-Mile Range and Advanced Features

Rivian’s TM-B e-bike, with a modular design, 28mph top speed, 180Nm torque, and 100-mile range, features a proprietary drivetrain, removable batteries, and integrated safety and connectivity systems, with preorders opening for spring 2026.

Rivian unveiled its TM-B electric bike, featuring 24 x 2.6-inch wheels, integrated lighting, and a modular frame capable of transforming into cargo or passenger configurations.
The bike uses a proprietary “DreamRide” pedal-by-wire drivetrain with a generator that replenishes the battery, which is available in 538Wh or 808Wh packs, offering up to 100 miles range.
The removable battery charges via USB-C at 240W, reaching full capacity in 2 hours 20 minutes (538Wh) or 3 hours 45 minutes (808Wh), and can serve as a portable power bank; it includes a battery charge display.

Linux Boosts FPS and Stability on ROG Xbox Ally, Outperforming Windows

A benchmark shows Linux (Bazzite) improves FPS by up to 32%, stabilizes framerates, and reduces sleep resume times on the ROG Xbox Ally, surpassing Windows performance.

Linux (Bazzite) yields up to 32% higher FPS than Windows on the ROG Xbox Ally, with more stable framerates and quicker sleep resume times, based on a YouTuber’s benchmark.
The test involved running games like Kingdom Come: Deliverance 2 and Hogwarts Legacy, showing FPS gains of +15 and +12 respectively, averaging a +6.6 FPS (13.47%) improvement.
Linux’s performance benefits are attributed to reduced OS overhead, optimized scheduling, and less resource consumption, with wake-from-sleep times nearly instantaneous on Linux versus up to 55 seconds on Windows.

▶️ Software Development

Carnegie Mellon’s Proto-X Boosts PostgreSQL Performance 10x with AI Tuning

Carnegie Mellon’s Proto-X employs vector embeddings and reinforcement learning to automatically optimize PostgreSQL settings, achieving up to 10x performance boosts and reducing tuning from 12 hours to 50 minutes.

Carnegie Mellon researchers developed Proto-X, a vector-based holistic tuning agent for PostgreSQL, claiming up to 10x performance improvements.
Proto-X uses a Wolpertinger architecture with vector embeddings, encoder-decoder models, and reinforcement learning to optimize database configurations automatically.
Training with LLM boosting reduces tuning time from 12 hours to approximately 50 minutes, enabling rapid, adaptive database performance adjustments.

▶️ Management and Leadership

Fermento: iOS App Automates Fermentation Monitoring and HACCP Compliance

Vadim Drobinin’s “Fermento” app automates fermentation monitoring and HACCP compliance on iOS, using phase-specific constraints and decision trees to manage long-term, variable processes.

Developed “Fermento,” an iOS app for tracking fermentation parameters, hazards, and HACCP compliance, with features like decision trees and profile management.
Implements phase-based control with constraints (e.g., pH, humidity, temperature) validated against cached, phase-specific constraints to optimize monitoring.
Generates HACCP documentation automatically, including hazard enumeration, control points, limits, and corrective actions, exportable as PDF.

The article suggests replacing website-based cookie consent with browser-enforced settings, enabling one-time user preferences to improve privacy control, reduce web clutter, and simplify legal enforcement.

The article advocates for shifting cookie consent management from websites to browsers.
Current laws like GDPR and CCPA burden individual websites with compliance, causing user annoyance and legal complexity.
Proposes a one-time browser setup where users specify privacy preferences, which browsers enforce universally, simplifying compliance and enhancing user control.

Steve Klabnik Praises jj Rust VCS as a Promising Git Alternative

Steve Klabnik sees jj, a Rust-based VCS with growing industry adoption, strong team, and community, as a promising alternative to Git, mirroring Rust’s successful trajectory.

Steve Klabnik discusses his positive outlook on jj, a new version control system (VCS) written in Rust, highlighting its market fit, team, and user adoption.
jj is being adopted incrementally by companies like Google and Oxide, with internal use in large monorepos and positive community feedback.
The jj team is experienced, with recent governance formalization, and the project shows strong growth potential similar to early Rust community development.

Apple Removes Tea Dating Apps Over Privacy and Content Violations

Apple removed Tea and TeaOnHer from the App Store over content moderation and privacy violations after user complaints and data breaches; the apps remain on Google Play.

Apple confirmed the removal of the dating safety apps Tea and TeaOnHer from the App Store due to violations of content moderation and privacy rules.
The apps were removed following numerous user complaints, negative reviews, and reports of minors’ data exposure; they remain available on Google Play.
Violations cited include failure to provide reporting/blocking features, unauthorized sharing of personal information, and excessive negative feedback, breaching App Review Guidelines 1.2, 5.1.2, and 5.6.

Meta Cuts 600 Roles in AI Superintelligence Lab to Boost Agility and Innovation

Meta is cutting roughly 600 roles from its AI superintelligence lab to streamline operations and foster agility, while actively recruiting for its TBD Lab, amid concerns over AI breakthroughs and a $15 billion investment in Scale AI.

Meta is reducing approximately 600 roles within its AI superintelligence lab, affecting FAIR AI research, product AI, and infrastructure units, while maintaining the newly formed TBD Lab.
The reorganization aims to improve agility by decreasing bureaucratic overhead, enabling faster decision-making and increased individual impact, according to Meta’s internal memo by AI chief Alexandr Wang.
Despite layoffs, Meta continues to hire for its TBD Lab, including recent recruitment of OpenAI scientist Ananya Kumar and co-founder of Thinking Machines Andrew Tulloch; CEO Mark Zuckerberg expressed concerns over AI progress, prompting the restructuring and a $15 billion investment in Scale AI.

Google Restructures US Ad Sales to Boost Agility and Speed

Google is restructuring its US ad sales unit by removing management layers, including “Managers of Managers,” to enhance agility and decision speed, with changes taking effect in January 2026.

Google is flattening management layers in its US ad sales division, specifically within Google Customer Solutions (GCS), effective January 2026.
The restructuring removes the “Managers of Managers” layer and converts managers into “Heads of business” reporting directly to directors, with no intermediate management.
No layoffs were announced; affected employees will transition into other roles; the change aims to increase agility and decision-making speed amid slowing growth and AI-driven competition.

Microsoft to Revamp GitHub as Central Hub for AI-Powered Development

Microsoft intends to overhaul GitHub to integrate AI tools across all developer workflows, expanding beyond code storage to become a central platform for AI-driven software development.

Microsoft plans to overhaul GitHub to counter AI coding tool competition from Cursor and Claude Code, aiming to make AI tools accessible across all developer environments.
The transformation seeks to unify development tools, removing boundaries between apps, documents, and websites, reflecting AI’s erasure of traditional distinctions.
Microsoft is investing in enhancing GitHub features like Actions, analytics, security, and compliance, and aims to position GitHub as a central hub for AI-powered software development.

Tesla Posts Record Q3 Revenue Amid Optimus V3 Tease and Profit Concerns

Tesla posted record revenue of $28.1 billion in Q3 2025, with Musk teasing Optimus V3 in Q1 and plans for 1 million robots annually, amid profit and stock concerns.

Tesla reported Q3 revenue of $28.1 billion, a record high, but adjusted EPS of $0.50 below the consensus estimate of $0.54
Musk highlighted the Optimus V3 prototype expected in Q1 2025, with plans for a million-unit production line by year’s end
Tesla’s operating income decreased 40% YoY due to higher vehicle costs and tariffs; stock fell 1.6% after hours

HP SitePrint: Faster, Precise Autonomous Construction Layout Robot

HP SitePrint is an autonomous construction layout robot offering up to 30% faster printing, high-precision accuracy, obstacle avoidance, and cloud management, compatible with major Robotic Total Stations.

HP SitePrint is an autonomous construction layout robot designed to improve site productivity, accuracy, and ease of use.
Supports high-precision layout with accuracy up to ±3/32 inch (±2 mm) and floor deviation marking up to ±1/32 inch (±0.8 mm) using the HP SitePrint SMR prism.
Features include faster print points (up to 30% improvement), advanced obstacle avoidance via depth camera, real-time route adjustments, and cloud-based project management.

US Considers Export Restrictions on China Over Software and Rare Earths

The U.S. is contemplating export restrictions on China for products made with U.S. software, aiming to escalate trade tensions and counter China’s rare earth export limits, risking global supply chain disruptions.

The Trump administration is considering restricting exports to China of goods containing or produced with U.S. software, including laptops and jet engines.
The plan aims to retaliate against China’s recent restrictions on rare earth exports and could involve broad measures affecting all products made with U.S. software.
The proposed controls could disrupt global trade, impact U.S. industry, and resemble restrictions previously imposed on Russia, with potential implementation before the November 1 deadline.

UK CMA Grants Apple and Google Strategic Market Status to Boost Competition

UK CMA granted Apple and Google SMS designation, enabling oversight of their mobile platforms due to their substantial market power and user lock-in, with potential targeted interventions to promote competition.

UK’s Competition and Markets Authority (CMA) designated Apple and Google with “strategic market status” (SMS) under the UK Digital Markets, Competition and Consumers Act 2024.
The designation grants CMA broad oversight of their mobile app stores and platforms, recognizing their substantial and entrenched market power.
CMA’s investigation, initiated in January, involved over 150 stakeholders and concluded both companies hold positions of strategic significance in UK mobile markets, with users predominantly accessing content via their respective platforms.

Microsoft CEO Satya Nadella Earns $96.5M in FY2025 Amid Performance-Based Pay

Microsoft CEO Satya Nadella earned $96.5 million in FY2025, with over 95% performance-based, amid a 480:1 pay ratio to median employee, driven by stock awards tied to long-term value.

Microsoft CEO Satya Nadella received total compensation of $96.5 million for fiscal year 2025, a 22% increase from the previous year.
Compensation includes a $2.5 million base salary, $84 million in stock awards, $9.5 million in non-equity incentives, and $196,000 in other pay.
Since 2014, Microsoft’s revenue has tripled, net income quadrupled, and EPS quintupled; FY2025 revenue was $281.7 billion with a net profit of $101.8 billion.

Microsoft to End Office Online Server Support in 2026

Microsoft will end Office Online Server support on December 31, 2026, shifting focus to Microsoft 365, with no replacement for on-premises browser-based Office applications.

Microsoft will retire Office Online Server (OOS) on December 31, 2026, with no further security updates, support, or fixes.
OOS, launched in 2016 as the successor to Office Web Apps Server 2013, provides browser-based Word, Excel, PowerPoint, and OneNote for on-premises deployment.
Microsoft recommends migrating to Microsoft 365 and Office for the Web, focusing on cloud-first solutions, with no direct successor announced for OOS.

Texas Instruments Q3 Revenue Up 14% but Stock Falls on Cautious Outlook

Texas Instruments reported Q3 revenue of $4.74 billion (+14% YoY), with sector recovery slowed by US trade policy uncertainty; expects Q4 revenue of $4.22–$4.58 billion amid cautious customer investment.

Texas Instruments Q3 revenue was $4.74 billion, up 14% YoY, but stock fell 9% after missing analyst expectations and issuing a cautious outlook
Operating profit increased slightly to $1.66 billion from $1.55 billion; Q4 revenue forecast is $4.22–$4.58 billion
US trade policy uncertainty, tariffs, and macroeconomic factors slow semiconductor market recovery; TI reports a sluggish sector with a moderate rebound
Data center components grew over 50% year-to-date, representing TI’s fastest-growing market, despite its small revenue share
TI CEO Haviv Ilan noted customers are hesitant to invest due to unclear tariff rules and trade tensions, with ongoing delays in factory expansion decisions
TI expects an effective tax rate of about 13% in Q4, influenced by new US tax legislation; company emphasizes long-term value investments

Tenfold Launches Free Community Edition for Identity Governance Under 150 Users

tenfold’s free Community Edition enhances identity governance with automated onboarding, self-service requests, cloud sharing oversight, detailed file permissions, and audit logging for organizations under 150 users.

tenfold offers a free, full-featured Community Edition of its IGA solution for organizations with under 150 users
The platform supports out-of-the-box integration with key IT systems, enabling quick setup and streamlined identity governance processes
Features include role-based access automation, self-service access requests, monitoring of Microsoft 365 shared content, detailed file server permissions reporting, and an audit event log

Meta Cuts 600 AI Roles Amid Restructuring and Focus Shift to TBD Lab

Meta is reducing 600 roles across its AI division, including FAIR, while hiring for its superintelligence team, TBD Lab, amid restructuring and decreased focus on AI research.

Meta plans to cut approximately 600 roles in its AI division, including the Fundamental AI Research (FAIR) unit and AI product/infrastructure teams.
The layoffs coincide with ongoing restructuring, while Meta continues hiring for its superintelligence team, TBD Lab.
Meta’s AI research team has diminished in prominence, with FAIR leader Joelle Pineau leaving earlier this year; Meta aims to integrate FAIR research into TBD Lab.

▶️ Technology

Researchers Develop Safe Rectal Oxygen Delivery for Respiratory Failure

A Cincinnati team demonstrated rectal oxygen delivery in humans as a safe method to potentially treat respiratory failure, inspired by fish and animal studies, with ongoing efficacy evaluation.

Researchers won the 2024 Ig Nobel Prize for discovering mammals can breathe through their anus via intestinal capillaries.
A new study tested rectal oxygen delivery (enteral ventilation via anus, EVA) in mice, micro-pigs, and 27 healthy human men, showing safety and feasibility.
Human trials involved intra-anal administration of perfluorodecalin, with doses from 25 to 1,500 mL; participants experienced mild, temporary abdominal discomfort with no major adverse effects.
The study aims to evaluate EVA as a method to give lungs rest and aid patients with severe respiratory failure, potentially reducing lung injury from mechanical ventilation.
The research builds on earlier work with perfluorocarbon liquids like Oxycyte and is inspired by loach fish’s intestinal breathing under hypoxic conditions.

Samsung Galaxy XR: First Android XR Headset with Gemini AI and Immersive Features

Samsung Galaxy XR, the first Android XR headset, leverages Android XR OS with Gemini AI, supporting apps, immersive videos, 3D exploration, and multi-app workspace, starting at $1799.

Samsung Galaxy XR is the first device built on Android XR, powered by a new operating system for next-generation headsets and glasses.
Android XR combines Gemini’s helpfulness with environmental awareness, supporting apps from Google Play, immersive content, and open standards like OpenXR, WebXR, and Unity.
Galaxy XR offers features such as an infinite screen, mixed reality interactions, voice, eye, and hand navigation, and integration with Gemini AI assistant, with a starting price of $1799 or $149/month.

Google’s Willow Chip Achieves Verifiable Quantum Advantage with Quantum Echoes

Google’s Quantum Echoes algorithm, run on the Willow chip, achieves verifiable quantum advantage by executing the OTOC algorithm 13,000 times faster than classical supercomputers, enabling practical quantum applications in molecular and physical system modeling.

Google’s Willow quantum chip achieved the first verifiable quantum advantage by running the out-of-order time correlator (OTOC) algorithm, surpassing classical supercomputers by 13,000x.
The Quantum Echoes algorithm demonstrates repeatable, verifiable quantum computations, enabled by low error rates and high-speed operations on a 105-qubit array.
The algorithm models physical experiments, such as molecular structure analysis via Nuclear Magnetic Resonance (NMR), validated through experiments with molecules of 15 and 28 atoms, matching traditional NMR results.

Starcloud to Launch 5-GW Space Data Center with NVIDIA H100 GPUs

Starcloud aims to deploy a 5-GW orbital data center using NVIDIA H100 GPUs, enabling sustainable, high-performance AI processing in space with 10x lower energy costs and water conservation benefits.

Starcloud, an NVIDIA Inception startup, plans to launch a satellite in November to create a 5-GW orbital data center with 4 km-wide solar and cooling panels.
The satellite will feature an NVIDIA H100 GPU, providing 100x more compute power than previous space-based systems, marking its cosmic debut.
Space-based data centers can utilize the vacuum of deep space as an infinite heat sink, reducing water use and leveraging nearly unlimited solar power, with projected 10x lower energy costs than terrestrial options.

Ovi: A 5B Model for Synchronized Text-to-Video and Audio Generation

Ovi is a 5B parameter cross-modal model that generates synchronized 5-second videos and high-quality audio from text or text+image inputs, supporting resolutions up to 960×960.

Ovi is a video+audio generation model capable of producing 5-second videos at 24 FPS with 720×720 resolution, supporting higher resolutions like 960×960.
Supports text-only or text+image conditioning, with high-quality audio generated by a pretrained 5B audio branch.
Generates synchronized video and audio content, accessible via wavespeed.ai and HuggingFace.

Rivian Criticizes Tesla’s Sensor Stance Amid Falling LiDAR Costs

Rivian’s James Philbin criticizes Tesla’s sensor stance amid falling LiDAR costs, with Musk dismissing LiDAR as unnecessary; in 10 years, top-level autonomous features will be standard in new vehicles.

Rivian’s James Philbin criticizes Tesla’s “rigid point of view” on sensors, citing declining LiDAR costs and engineering realities.
Elon Musk has repeatedly dismissed LiDAR, claiming it is expensive and unnecessary once vision-based AI is solved.
LiDAR prices have dropped from $75,000 to approximately $200 over 15 years, improving sensor affordability and performance in various lighting conditions.

Google’s Quantum Computer Breaks Speed Records, Boosting Medicine and Cryptography

Google’s quantum computer near Santa Barbara executes quantum algorithms 13,000 times faster than classical supercomputers, enabling breakthroughs in medicine and cryptography, supported by recent Nobel-winning quantum physics research.

Google’s quantum computer near Santa Barbara achieved a quantum algorithm that runs 13,000 times faster than software on traditional supercomputers
The development supports advances in medicine, cryptography, and other fields, building on foundational quantum physics discoveries recognized by the 2025 Nobel Prize in Physics
The quantum system’s capabilities may enable new medicines, vaccines, and breaking current encryption techniques

Metagenomi Slashes AI Costs 56% with AWS Inferentia 2 Accelerators

Metagenomi cut AI costs by 56% using AWS Inferentia 2 accelerators instead of Nvidia GPUs, leveraging batch processing and spot instances for efficient gene therapy enzyme discovery.

Metagenomi reduced AI operational costs by 56% by switching from Nvidia GPUs to AWS Inferentia 2 accelerators
Inferentia 2 features 32GB HBM, 820 GB/s memory bandwidth, and 190 teraFLOPS of 16-bit performance; Nvidia L40S offers 48GB GDDR6, 864 GB/s bandwidth, and 362 teraFLOPS
Cost savings stem from AWS’s batch processing, spot instances, and lower interruption rate (5% vs. 20%) for Inferentia, enabling more frequent enzyme discovery

GM Unveils Gemini AI, Level 3 EVs, and Robotics Amid Market Challenges

GM plans to introduce a Gemini-powered AI assistant in 2026, a new EV platform in 2028 with Level 3 autonomy, and expand robotics and energy solutions amid market and regulatory shifts.

GM CEO Mary Barra and Chief Product Officer Sterling Anderson discussed AI, autonomy, and EV strategies in a Decoder podcast episode on Oct 22, 2025.
Announced a Google Gemini-powered AI assistant launching in 2026, a new hardware/software platform for the 2028 Escalade IQ enabling Level 3 autonomy and enhanced Super Cruise.
GM is developing a robotics division, advancing battery chemistry (notably lithium manganese-rich LMR batteries in 2028), and expanding energy solutions including vehicle-to-home power.
GM is ditching Apple CarPlay and Android Auto in EVs to prioritize integrated, seamless in-vehicle systems, citing customer feedback and safety considerations.
The company faces market challenges like a $1.6 billion EV writedown, regulatory shifts, tariffs, and the end of US EV tax credits, influencing production and pricing strategies.

Alvaro Lopez Ortega / 2025-10-22 Briefing

▶️ Internet Infrastructure

▶️ Open Source

▶️ Software Development

▶️ Management and Leadership

▶️ Technology