Chrome and Edge to embed AI in the browser

inkl.com Aug 23, 2026

Chrome and Edge to embed AI in the browser

Curated by Shubham · TechCapsules · Aug 23, 2026

AI-summarised brief · reviewed before publication

Google and Microsoft are integrating on-device artificial intelligence directly into their Chrome and Edge web browsers. This background feature requires users to maintain at least 20GB of free storage space on their devices. Google confirmed that Chrome will automatically download generative AI models, specifically Gemini Nano, provided the device meets performance thresholds and uses an unmetered internet connection. These local models power features such as writing assistance, text summaries, tab management, and enhanced security tools. The 20GB requirement is a minimum threshold rather than the actual file size, which varies. Users can disable this functionality in settings to prevent downloads and delete existing models. Microsoft Edge employs a similar strategy, requiring 20GB of free space before downloading models like Phi-4-mini. Edge includes an automatic fail-safe that deletes downloaded AI models if available disk space drops below 10GB. This move reflects a broader industry trend of browsers pre-loading local AI models without explicit user prompts, necessitating routine storage checks for users with limited disk capacity.

💡 Why It Matters

· Browser vendors are shifting AI processing from the cloud to local hardware, fundamentally altering how users manage digital storage.
· This silent pre-loading of models forces consumers to treat disk space as a critical resource for modern web navigation.

AI Capsules Google Microsoft

Read full article on inkl.com

HTX Research Examines U.S. AI Equities: Technology Remains Early, While Capital Expenditure…

HTX Research released a report analyzing U.S. artificial‑intelligence equities, noting that technology stocks remain in an early‑stage growth phase while capital expenditures and valuation metrics have shifted into a late‑cycle environment. The study highlights that AI‑driven companies are still benefiting from rapid adoption and innovation, yet their spending on infrastructure and the price‑to‑earnings ratios have begun to reflect a more mature market. The report underscores the importance of distinguishing between the developmental momentum of AI technology and the broader economic pressures affecting investor sentiment and corporate spending.

💡 The report pinpoints a critical juncture where AI firms must balance continued innovation with realistic spending and valuation expectations, reshaping investment strategies in a rapidly evolving sector.

egamers.io · Aug 23

Instagram Quietly Routes Your Off-App Activity Into Ads and AI — The…

Instagram does not offer a single “kill switch” for data collection; instead, users must navigate seven dispersed settings across the app, device system menus, and a hidden web page to limit tracking. Meta’s “Activity from other businesses” toggle, introduced in July 2026, lets users block businesses from using off‑platform activity—such as website visits or purchases—to personalize ads, feed content, and AI responses, but it does not stop Meta from using on‑platform behavior, device data, location signals, or information shared between linked accounts. The setting is rolled out country‑by‑country, with European users seeing an additional “personalized vs. less‑personalized ads” option in the Accounts Center. Despite these controls, Meta continues to train its AI models on a broad array of user signals, meaning complete privacy is unattainable.

💡 Users thinking a single toggle can shield them are misled, as Meta still harvests extensive on‑platform and device data to fuel its ad engine and AI.

cyberera.com.ng · Aug 23

AI Is Already On The Job, And Census Is Keeping Count

A recent Census Bureau survey found that 55 % of U.S. workers have used AI for at least one of 11 job-related tasks, with the most common uses being information search (37 %), writing communications (32 %), idea generation (32 %), and data interpretation (31 %). About 27 % employed AI for administrative duties. Workers reported time savings: 31 % saved one to two hours, 15 % saved three to four hours, and another 15 % saved over four hours. However, 10 % saw no time benefit and 3 % experienced increased time. The data illustrate everyday AI adoption beyond advanced model development.

💡 The findings show that AI is already embedded in routine work, offering measurable productivity gains for many employees. This trend signals a shift toward broader, practical AI integration across both private and public sectors.

techcompanynews.com · Aug 23

Rundoo Raises $30 Million In Series B Funding Round

Rundoo, a Redwood City startup founded in 2021 by Stanford alumni, closed a $30 million Series B financing round led by Battery Ventures, with participation from existing backers Bessemer Venture Partners and CRV, bringing total capital raised to $48 million. Battery Ventures’ Michael Brown joined the company’s board. The AI‑native operating platform consolidates point‑of‑sale, e‑commerce, CRM, loyalty, accounting, inventory and payments for independent hardware, paint, lawn‑and‑garden, farm‑feed and lumber retailers. Its core AI agent, Dooey, generates purchase orders, suggests promotions, delivers daily business recaps and provides price comparisons using store‑specific data, weather forecasts and local demand signals. Rundoo now serves over 500 stores in the U.S., Canada and the Caribbean, reporting first‑year sales growth of 17 %, receivables reductions exceeding 25 % and marginal gross‑margin gains of 1–2 %.

💡 By embedding AI into the back‑office of small‑format retailers, Rundoo gives independent stores a technology edge traditionally reserved for large chains, potentially reshaping the competitive dynamics of the hardware and home‑improvement market.

sourcetrail.com · Aug 23

Mastering Edge AI with Raspberry Pi LiteRT and Gemma

LiteRT-LM enables efficient execution of Google’s Gemma 4 language models on edge hardware, specifically the Raspberry Pi. This orchestration layer utilizes mixed-precision quantization to reduce memory requirements to as little as 0.8 GB of RAM, making complex large language models viable for resource-constrained devices. The system leverages hardware acceleration through XNNPACK and experimental WebGPU support to optimize inference speeds. Additionally, Multi-Token Prediction drafters accelerate processing by up to three times. The integration supports multimodal inputs, including vision and audio, alongside function calling for agentic workflows. Developers can implement self-healing firmware with dual-bank updates to ensure stability during over-the-air deployments. This combination allows IoT devices to process data locally, maintaining privacy and reducing latency without requiring constant cloud connectivity. The setup facilitates private, low-latency applications that generate human-like text directly on the device, expanding the capabilities of edge computing beyond simple scripting to sophisticated generative AI tasks.

💡 Localized AI processing eliminates cloud dependency, ensuring sensitive data remains on-device while delivering instant responses. This shift empowers developers to build resilient, private intelligent systems that function reliably even with intermittent internet connectivity.

eurogamer.net · Aug 23

Twitch faces a class action lawsuit over using streamers’ content to train…

Twitch streamer Warren Pandiscia filed a proposed class-action lawsuit against Twitch and Amazon in the Northern District of California, alleging the companies used creator content for generative AI training without consent or payment. The 37-page complaint claims this practice breaches implied agreements and violates state unfair competition laws. Twitch recently announced it would train AI models on streams, clips, chat logs, and photos, sparking immediate backlash from the creator community. Chief Product Officer Mike Minton defended the default opt-out setting, stating an opt-in approach would yield no participation. The lawsuit follows Twitch’s admission that the opt-out is not retroactive and that content used prior to the setting’s introduction remains in training datasets. Pandiscia argues the companies quietly utilized content as early as 2024. The legal action highlights growing tensions between content platforms and creators regarding AI data usage. Neither Twitch nor Amazon has commented on the filing. Amazon acquired Twitch in 2014 for approximately $970 million.

💡 The non-retroactive nature of the opt-out mechanism exposes creators to liability for past content usage, setting a precedent for how digital platforms handle historical data in AI development. This legal challenge tests the boundaries of implied consent in user agreements.