Today’s tech landscape saw significant advances in AI with OpenAI launching autonomous ChatGPT agents capable of executing complex tasks, while Meta and the EU remain at odds over AI regulation and safety guidelines. Meanwhile, TSMC and Rapidus accelerate cutting-edge chip manufacturing in the US, and the FCC proposes new rules to secure undersea internet cables amid geopolitical tensions.
▶️ Internet Infrastructure
GM Teams Up with Redwood to Reuse EV Batteries for AI Data Center Power
GM is repurposing EV batteries for energy storage in AI data centers through a partnership with Redwood Materials, supporting grid stability and manufacturing independence amid increasing electricity demand.
- GM partners with Redwood Materials to supply energy for AI data centers using second-life EV batteries and new US-manufactured batteries.
- The collaboration includes repurposing retired EV batteries for grid-scale energy storage, supporting AI data centers like Crusoe’s Nevada microgrid.
- The initiative addresses rising electricity demand, emphasizing the industrial trend of converting retired assets into strategic resources amid supply chain stresses.
Cisco patches critical remote code execution flaw in ISE and ISE-PIC
Cisco released a security update for CVE-2025-20337, a critical remote code execution flaw in ISE and ISE-PIC, enabling attackers to gain root access through crafted API requests.
- Cisco issued a patch for CVE-2025-20337, a critical 10/10 severity vulnerability in ISE and ISE-PIC affecting versions 3.3 and 3.4
- The flaw allows unauthenticated remote attackers to execute arbitrary code with root privileges via crafted API requests
- The vulnerability is related to insufficient validation of user input and was disclosed alongside two other max-severity flaws in June
FCC Proposes Rules to Block Adversary-Controlled Undersea Cables
The FCC plans to implement rules to block Chinese and other adversary-controlled entities from building or leasing undersea internet cables, aiming to enhance infrastructure security amid geopolitical threats.
- FCC chairman Brendan Carr announced a rule proposal to block “certain foreign adversary-controlled” applicants for undersea cable projects and prohibit “covered” equipment in infrastructure.
- The rule aims to secure US-connected submarine cables from Chinese and Russian threats, with a vote scheduled for next month.
- The proposal includes default denials for certain foreign-controlled applicants and restrictions on specific equipment, though “covered” equipment definitions remain undisclosed.
Google Sues 25 Chinese Operators Over Massive BadBox 2.0 Botnet
Google sued 25 Chinese operators for BadBox 2.0, infecting over 10 million devices globally, disrupting cybercrime infrastructure, with efforts to sinkhole C2 domains and dismantle the operation.
- Google filed a lawsuit against 25 unnamed individuals in China for operating the BadBox 2.0 botnet, infecting over 10 million devices worldwide as of April 2025
- The botnet comprises primarily AOSP-based smart TVs, tablets, projectors, and car infotainment systems, making it the largest CTV botnet uncovered
- The BadBox 2.0 Enterprise includes multiple groups managing C2 servers, malware, and schemes like ad fraud and proxy services, with interconnected infrastructure
TSMC Accelerates US Chip Production with 3nm and 2nm Fab Expansion
TSMC plans to ramp up US production at its Arizona fabs, targeting 30% of high-end chips, with accelerated schedules for 3nm and 2nm nodes amid strong demand and geopolitical uncertainties.
- TSMC has completed construction of its second Arizona fab, set to produce 3nm chips, and plans to accelerate its ramp-up earlier than initially expected.
- The company aims to shift nearly 30% of high-end wafer output to the US, with US production expected to include 2nm and smaller process nodes.
- TSMC’s Q2 revenues increased 38.6% YoY to $31.7 billion USD, with 74% of wafer revenue from leading-edge nodes; net income exceeded $13.5 billion USD.
CMA to Take Swift Action Against Cloud Giants to Boost Competition and Innovation
The CMA plans to enforce swift, binding remedies against AWS and Microsoft Azure’s market dominance to foster competition and AI innovation, avoiding delays that could entrench incumbents and hinder UK’s digital growth.
- The UK’s Competition and Markets Authority (CMA) is set to publish a market health report in August 2025, focusing on cloud market dominance.
- AWS and Microsoft Azure control approximately 80% of UK cloud spending, exhibiting entrenched market power through practices like high exit fees, vendor lock-in, and discriminatory licensing.
- The CMA has new powers under the Digital Markets, Competition and Consumers Act 2024, enabling targeted interventions such as designating firms with Strategic Market Status (SMS) and mandating interoperability, open APIs, licensing reforms, and caps on exit fees.
Oracle Database@AWS Launch Offers Up to 76% Cost Savings on Autonomous Database
Oracle introduced Oracle Database@AWS, allowing users to run Oracle Autonomous Database on Exadata hardware in AWS datacenters, incentivizing perpetual licenses with 76% savings over subscriptions.
- Oracle launched Oracle Database@AWS in US East and US West regions, with plans for additional regions
- Enables running Oracle Autonomous Database on Exadata hardware within AWS datacenters, managed via Oracle Cloud Infrastructure
- Oracle promotes BYOL licensing, offering up to 76% cost savings compared to subscription models, requiring a private offer and negotiation
Rapidus Targets 2027 for 2nm Chip Production Using IBM Technology
Rapidus aims for 2nm chip production by 2027, having licensed IBM’s tech, installed 200+ equipment units, and developed a single-wafer AI-driven process, two years after TSMC and Samsung.
- Rapidus announced successful prototyping of 2nm GAA transistors at its IIM-1 plant in Chitose, Hokkaido, after trialing ASML’s EUV lithography in April 2025
- The company licensed 2nm process technology from IBM, demonstrated in 2021, and has installed over 200 semiconductor equipment units at its facility
- First PDK expected in Q1 2026, with mass production targeted for 2027, approximately two years behind TSMC and Samsung, which are ramping 2nm production in 2025
VMware Extends Cloud Foundation Support to Six Years and Slows Release Cycle
VMware extended Cloud Foundation support to six years and slowed release cadence to every three years, with updates every nine months, aligning with Broadcom’s profit and cost strategies.
- VMware announced extended product lifecycle from two to three years between major releases and support duration increased from five to six years.
- Major release cadence slowed from every two years to every three years, with updates every nine months, providing more time for installation and maintenance.
- Support for Cloud Foundation (VCF) extended to 27 months per update, with the last update supported for 45 months; support lifecycle visualized here.
- Support support for major releases extended to six years, with two years of extended support offered to some customers; subscription model favors three-year procurement cycles.
- VMware’s support document indicates desktop hypervisors Workstation and Fusion will require manual updates via Broadcom Support Portal.
- Broadcom aims to increase VMware profitability by reducing update frequency and exploring “supervisor services” for adding features without large upgrades, potentially enhancing future VCF updates.
▶️ Open Source
Meta Uses AI to Accelerate Sustainable Concrete Development
Meta used AI with Bayesian optimization to create low-carbon concrete, enabling faster formulation validation for datacenter construction, reducing environmental impact, and sharing its code openly.
- Meta developed an AI model utilizing Bayesian optimization to predict concrete compressive strength curves.
- The model accelerated the process of formulating low-carbon concrete for a datacenter in Rosemount, Minnesota.
- Meta published its concrete formulation code under the MIT License, aiming to improve sustainability and performance in construction.
Rescuezilla 2.6.1 Adds Ubuntu 25.04 Support and Virtual Disk Compatibility
Rescuezilla 2.6.1 introduces a new build based on Ubuntu 25.04 “Plucky Puffin” and updates existing versions, supporting virtual disks and including Firefox; it coincides with Ubuntu 24.10’s end-of-life.
- Rescuezilla 2.6.1 adds a new Ubuntu 25.04 “Plucky Puffin” build alongside existing versions based on Bionic (18.04), Focal (20.04), Jammy (22.04), Noble (24.04), and Oracular (24.10)
- The release supports virtual disks from VirtualBox, VMWare, QEMU, Hyper-V, and others via QEMU Disk Network Block Device
- Ubuntu 24.10 reached end-of-life on July 10, 2025, shortly before the release
▶️ Management and Leadership
SEC Considers Innovation Exception to Boost Tokenization and Blockchain Securities
SEC is exploring an ‘innovation exception’ to facilitate tokenization and blockchain securities, aiming to enable novel trading and tailored regulatory reliefs, amid recent stablecoin legislation.
- SEC Chairman Paul Atkins stated the agency is considering an ‘innovation exception’ to regulations to promote tokenization, following the US House’s passage of a stablecoin bill on July 17, 2025.
- The proposed exemption would allow for novel trading methods and tailored reliefs to support a tokenized securities ecosystem.
- Staff is evaluating potential regulatory adjustments to incentivize blockchain-based securities development within existing frameworks.
Meta Rejects EU AI Code Over Overreach and Legal Concerns
Meta refuses to sign the EU’s AI code, citing overreach and legal uncertainties, criticizing it as exceeding the scope of the AI Act and potentially hindering AI development.
- Meta announced it will not sign the EU’s AI code of practice, citing concerns over overreach and legal uncertainties.
- Meta’s head of global affairs Joel Kaplan described the guidelines as exceeding the scope of the AI Act and introducing legal ambiguities.
- The EU’s AI code aims to regulate AI development and deployment, but Meta considers it an overextension that could hinder model developers.
Jamie Siminoff Rejoins Ring as CEO to Focus on AI and Safety Innovation
Jamie Siminoff returned as Ring CEO in April 2025, restoring its crime-prevention mission, prioritizing AI integration, product innovation, and operational efficiency amid broader Amazon cultural shifts.
- Jamie Siminoff rejoined Amazon as Ring CEO in April 2025, replacing Liz Hamren, and led a company-wide overhaul.
- He reinstated Ring’s original mission: “Make neighborhoods safer,” shifting away from the community-focused branding.
- Siminoff emphasizes AI-driven innovation, operational efficiency, and product roadmap changes, including a new home surveillance drone and real-time activity alerts.
Fujitsu Wins Over Half a Billion in UK Contracts Post-Horizon Scandal
Fujitsu, after apologizing for its role in the Horizon scandal, has secured over £510 million in UK public sector contracts since 2024, including a £220 million HMRC deal, despite ongoing controversy and voluntary bidding restrictions.
- Fujitsu has received approximately £510 million ($682 million) in UK public sector contracts since the Horizon Post Office scandal dramatization, including a £220 million ($294 million) contract with HMRC awarded without competition.
- The Horizon scandal involved wrongful prosecutions of around 736 subpostmasters from 1999 to 2015 due to flawed IT systems developed by Fujitsu, leading to multiple suicides and public inquiry findings.
- Fujitsu CEO Takahito Tokita, CFO Takeshi Isobe, and UK/Europe head Paul Patterson have apologized; Fujitsu committed to halting new public sector bids during the inquiry but has continued winning large contracts, including a recent £220 million deal with HMRC in June.
- HMRC’s existing services rely on Fujitsu-hosted infrastructure, preventing migration to new contractors, with a tender launched in June for up to £417 million ($560 million) for hyperscaler services to replace Fujitsu datacenters.
- Other notable contracts include a £125 million ($167 million) land registry system for Northern Ireland, though the government later confirmed it did not request Fujitsu to continue bidding; Fujitsu’s continued bidding despite pledges to stop has been questioned by MPs and peers.
Intel to Cut 5,000 Jobs Amid Market Decline and Cost Cuts
Intel plans to cut 5,000 jobs, mainly in North America and Israel, as part of a 20% headcount reduction to improve efficiency amid declining market dominance and uncertain US CHIPS funding.
- Intel plans to lay off approximately 5,000 staff, mainly in California and Oregon, as revealed in WARN notices.
- The layoffs include nearly 2,000 in California (Folsom and Santa Clara), around 2,500 in Oregon (Hillsboro and Aloha), and additional cuts in Arizona and Israel.
- The company aims to reduce headcount by 20% and cut costs amid declining dominance in x86, GPU markets, and underperforming foundry services; over 20,000 staff have been shed in a year.
Microsoft Extends Security Support for Exchange and Skype Until 2026
Microsoft extends security updates for Exchange Server 2016/2019 and Skype for Business 2015/2019 by six months until April 14, 2026, but only via paid Extended Security Update program.
- Microsoft extends security update support for Exchange Server 2016 and 2019, and Skype for Business 2015 and 2019, by six months until April 14, 2026
- The extension covers Critical- or Important-rated security updates released after October 14, 2025, the products’ support end date
- Updates will only be available to customers who sign up for the Extended Security Update program, with costs determined through Microsoft account teams
AWS Cuts About 100 Jobs Amid Automation and Strategic Shift
AWS laid off around 100 employees across various teams due to automation efforts and strategic shifts, with layoffs described as abrupt and driven by AI integration and marketing changes.
- AWS layoffs involve approximately 100 jobs across multiple teams, primarily in marketing and outreach, with no specific team disclosures.
- Amazon confirmed the layoffs are due to a company-wide automation initiative and a strategic review of organizational priorities.
- Employees reported the layoffs were abrupt, with no warning, and driven by a focus on high-level optics rather than detailed role assessments; AI adoption and marketing restructuring are cited as contributing factors.
CISPE Secures Pay-As-You-Go Microsoft Licensing Amid Criticism of Lock-In
European cloud providers’ group CISPE gained pay-as-you-go licensing and privacy protections from Microsoft, but critics say it delays broader competition and maintains restrictive ID management.
- CISPE secured pay-as-you-go licensing for Microsoft software, including Windows Server and SQL Server, at prices comparable to Azure, with privacy protections.
- The agreement allows deployment of Microsoft 365 Local suite pending general availability in CSP-H, but retains the technical tie-in between Entra ID and Microsoft 365.
- Critics, including the Coalition for Fair Software Licensing, argue the deal is a stalling tactic that prolongs Microsoft’s customer lock-in and anticompetitive practices.
WeTransfer Clarifies No Current AI Use After User Backlash Over ToS Changes
WeTransfer added language in July 2025 granting rights to use user content for AI training, prompting user backlash; the company clarified it does not currently employ AI for content processing.
- WeTransfer updated its Terms of Service on July 18, 2025, adding language granting license to use user content for AI model training, including content moderation improvements.
- The company clarified it does not currently use AI or machine learning to process shared content and rolled back the contentious language after user backlash.
- Similar legal adjustments are occurring industry-wide as companies modify ToS to accommodate permissions for AI data processing, often citing future AI moderation or development needs.
Meta Rejects EU AI Safety Guidelines Citing Legal Uncertainties
Meta refuses to sign EU AI safety guidelines, citing legal uncertainties, while the EU mandates compliance with the AI Act for large-scale AI models trained with over 10^23 FLOPs, effective August 2, 2025.
- Meta declined to sign the EU voluntary AI safety guidelines two weeks before the AI Act takes effect, citing legal uncertainties beyond the law’s scope.
- The EU guidelines support the AI Act, which categorizes AI risks into four levels and aims to prevent illegal, extremist, or harmful content.
- Meta’s stance includes ignoring the GPAI code, which requests transparency, copyright, and safety commitments from providers of large-scale AI models trained with over 10^23 FLOPs, including Meta’s own Llama 4 (5e25 FLOPs).
- Meta criticizes the EU’s approach, claiming it introduces legal uncertainties and exceeds the AI Act’s scope, potentially hindering AI development in Europe.
- The European Commission states all GPAI providers must comply with the AI Act from August 2, 2025, with voluntary codes serving as benchmarks; non-signatories must demonstrate alternative compliance methods.
▶️ Technology
Mistral AI Unveils Le Chat Updates with Deep Research Voice and Multilingual Tools
Mistral AI introduced Le Chat enhancements in July 2025, featuring Deep Research for structured reports, Voxtral voice mode, multilingual reasoning via Magistral, project organization, and integrated image editing.
- Mistral AI announced Le Chat updates including Deep Research (Preview), Voice mode powered by Voxtral, multilingual reasoning with Magistral, Projects organization, and advanced image editing in July 2025.
- Deep Research enables structured, source-backed reports on complex topics, leveraging tool-augmented agents for transparent research.
- New voice interaction allows natural speech recognition, facilitating brainstorming, quick answers, and transcription without typing.
Business Insider Seeks Nominations for 2025 AI Power List
Business Insider is calling for nominations for its 2025 AI Power List, highlighting leaders in AI infrastructure, influence, and societal impact, with submissions due by August 29, 2025.
- Business Insider is accepting nominations for the 2025 AI Power List, focusing on influential figures in AI and machine learning.
- The list emphasizes power in terms of infrastructure, influence, leadership, and societal impact, with nominations due by August 29, 2025.
- Candidates include technologists, ethicists, entrepreneurs, and individuals from companies like Google, Microsoft, Amazon, as well as startups and academics.
OpenAI Unveils ChatGPT Agent for Autonomous Complex Task Execution
OpenAI COO Brad Lightcap described AI agents as autonomous systems capable of executing complex, previously unseen tasks at high proficiency, exemplified by the new ChatGPT Agent, which integrates web engagement, information synthesis, and conversational skills.
- OpenAI COO Brad Lightcap defined AI agents as systems capable of autonomously performing complex tasks they’ve not previously encountered, with high proficiency.
- OpenAI introduced its latest ChatGPT Agent, combining Operator, Deep Research, and ChatGPT’s conversational abilities.
- The ChatGPT Agent can independently execute complex tasks from start to finish using its own computing resources.
AWS Launches Bedrock AgentCore for Secure Scalable AI Agent Deployment
AWS introduced Bedrock AgentCore, a service for deploying and managing AI agents securely at scale, supporting multiple frameworks, models, and tools, with free access until September 16.
- AWS previewed Bedrock AgentCore at the 2025 AWS Summit, enabling scalable, secure deployment of AI agents across frameworks and models.
- AgentCore offers services including runtime, memory, authentication, tool discovery gateway, code interpreter, web browsing, and observability.
- Currently available in US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney), and Europe (Frankfurt), free until September 16, with pricing details on AWS.
Researchers Achieve 2.8x Lossless Speedup in AI Token Generation
Researchers introduced three algorithms, including Token-Level-Intersection and String-Level Exact Match, enabling lossless 2.8x speedup in large language model token generation without specialized draft models.
- Researchers at Weizmann Institute, Intel Labs, and d-Matrix developed algorithms to enhance AI performance losslessly by up to 2.8x
- Presented at the International Conference on Machine Learning; detailed in paper
- Algorithms improve speculative decoding by enabling any model to serve as a draft without matching vocabularies, eliminating the need for specialized draft models
Qualcomm Unveils Snapdragon X Series for Power-Efficient Laptops
Qualcomm’s Snapdragon X Series introduces power-efficient, multi-core processors with modular architecture, delivering up to 22 hours battery life and 90% faster unplugged performance, disrupting the PC market.
- Qualcomm Technologies launched the Snapdragon X Series, a non-x86 processor family for laptops, emphasizing power efficiency and performance.
- The architecture uses Qualcomm Oryon cores, with 12 cores on Snapdragon X Elite and 8-10 cores on Snapdragon X Plus, designed with a mobile-first, power-efficient mindset.
- Features include modular multi-engine architecture, independent modules like GPU and NPU, enabling up to 90% faster performance when unplugged and up to 22 hours of battery life.
OpenAI Launches Autonomous ChatGPT for Paid Users with Advanced Tools and Safety
OpenAI introduced ChatGPT as an autonomous agent for paid users, capable of executing multi-step tasks with integrated tools, enhanced safety controls, and resistance to prompt injection.
- OpenAI announced ChatGPT can now act as an autonomous agent capable of executing complex tasks using local and online resources, available to paid subscribers.
- The agent integrates OpenAI’s Operator, deep research tools, web browsers, terminal, APIs, and connectors to services like Gmail and GitHub, running in a dedicated virtual machine.
- Safety measures include permission prompts for real-world actions, safeguards against prompt injection, and restrictions on high-risk activities; the system resists 99.5% of prompt injection attempts, dropping to 95% in adversarial scenarios.