Back to MergeSort

AI News Digest - August 25, 2026

15 stories · August 25, 2026

Listen to the podcast
3.8 MB · Download MP3

Top Stories

partnership

SpaceX and Nvidia Partner to Launch Orbital AI Data Centers by 2027

SpaceX and Nvidia have announced a partnership to build space-based Starmind data centers utilizing Nvidia's Vera Rubin NVL72 systems, aiming to deploy the first AI-compute racks in orbit by late 2027. The orbital infrastructure will power agentic AI applications and support the Grok model, bridging terrestrial AI hardware with space-optimized computing.

Sources: Ben's Bites, HuggingNews, Michael Parekh AI, OpenAI Blog, The Neuron, The Rundown AI

product_launch

Apple Unveils M6 and M5 Ultra Chips to Power Next-Generation On-Device AI

Apple has announced its first 2nm chip, the M6, alongside the quad-die M5 Ultra chip, designed to power refreshed Mac mini and Mac Studio desktops. These new systems support up to 512GB of unified memory and deliver up to 4.3x faster AI performance, significantly accelerating local large language model inference and professional workflows.

Sources: Hacker News, HuggingNews, Lobsters AI, Reddit AI

product_launch

OpenAI Unveils Custom Jalapeño AI Inference Chip to Challenge Nvidia

OpenAI has announced its new Broadcom-designed custom inference chip, named Jalapeño, which is engineered to deliver faster, more power-efficient AI model execution with lower latency. In internal tests, the 700-watt processor outperformed Nvidia's GB300 in power efficiency and inference speed, marking a significant step in OpenAI's strategy to deploy proprietary silicon and reduce data center costs.

Sources: HuggingNews, OpenAI Blog

product_launch

Alibaba Releases Qwen3.8-Flash-Next Model with 120 Billion Parameters

Alibaba has introduced its Qwen3.8-Flash-Next model on ModelScope, featuring a multimodal Mixture-of-Experts architecture with 120 billion total parameters and 6 billion active parameters. The release has generated significant interest in the open-source community, offering developers an early look into the structural improvements of the upcoming Qwen4 architecture for local LLM deployment.

Sources: HuggingNews, Reddit AI

research

Qwen3.8-27B Excels on Code Arena as Industry Shifts Focus to AI Engineering and Better Evaluation Metrics

The Qwen3.8-27B model has achieved the 9th position on the Chatbot Arena coding benchmark, significantly outperforming expectations for its size class. Meanwhile, industry developments highlight a shift toward practical application, with DeepLearning.ai relaunching to focus on AI Engineering, Anthropic introducing enterprise-managed authentication for its Model Context Protocol, and NVIDIA proposing a new 'Skill Lift' metric to better evaluate AI agent utility.

Sources: Latent Space, Reddit AI

More Stories

policy

OpenAI Bans Russian Accounts Using ChatGPT to Run Fake Think Tank Disinformation Campaign

OpenAI has disrupted and banned a cluster of Russia-origin accounts that utilized ChatGPT to generate social media posts promoting a fake Israel-based think tank called the International Burke Institute. The covert influence operation used plagiarized academic work and a fabricated sovereignty index to praise Russia while criticizing Western nations.

Sources: OpenAI Blog, Reddit AI

product_launch

Google Cloud Launches Gemini Enterprise with Specialized Legal and Financial AI Agents

Google Cloud has introduced Gemini Enterprise AI agents designed to automate complex workflows like contract review and regulatory monitoring for the legal and financial sectors. To deliver these domain-specific capabilities while ensuring strict data privacy, Google has partnered with industry leaders including Thomson Reuters, Moody's, Harvey, and DocuSign.

Sources: HuggingNews

policy

Taiwan Indicts Nine Individuals Including Nvidia and Super Micro Staff for Smuggling AI Servers to China

Taiwanese prosecutors have indicted nine people, including an Nvidia employee and two former Super Micro staff, for allegedly smuggling 74 high-end B300 AI servers to China in violation of US export controls.

Sources: HuggingNews

product_launch

OpenAI Revenue Surges to $40 Billion as Greg Brockman Consolidates Control Ahead of IPO

Driven by the launch of its lower-priced GPT 5.6 model, OpenAI's annualized revenue has jumped 35% to over $40 billion, narrowing the gap with rival Anthropic. Simultaneously, co-founder and president Greg Brockman has expanded his operational control following a series of high-profile executive departures. These developments come as the company actively prepares for a targeted public offering in 2027.

Sources: Michael Parekh AI

product_launch

IBM releases Granite 4.2 reasoning model family

IBM released Granite 4.2, a new family of dense, decoder-only reasoning LLMs available in 3B, 8B, and 30B sizes with native tool calling and context windows up to 512K tokens. All models in the family are open-sourced under the Apache 2.0 license.

Sources: Hugging Face Blog

product_launch

DeepSeek adds vision capabilities to V4 Flash model

DeepSeek has added multimodal input and visual-agent capabilities to its low-cost V4 Flash model tier.

Sources: The Neuron

research

NVIDIA's AVO completes all ARC-AGI-3 public levels

The AVO agent system by NVIDIA successfully completed all 183 public levels of ARC-AGI-3, highlighting the impact of model harnesses on long-running tasks.

Sources: The Neuron

product_launch

Cerebras unveils CS-4 rack-scale inference system

Cerebras introduced the CS-4 inference system, claiming speeds up to 30 times faster than competing GPUs and 10 times the throughput of its predecessor.

Sources: The Neuron

policy

Apple Music to introduce mandatory AI-generated content labels

Apple Music announced it will explicitly label tracks, artwork, compositions, and music videos that are materially generated using artificial intelligence.

Sources: The Neuron

product_launch

Personal agent Instinct under scrutiny over persistent email data storage

Invite-only personal assistant Instinct faced privacy concerns after users discovered that disconnecting Google accounts did not immediately purge previously synced emails from its records.

Sources: The Neuron

From the Podcasts

Parallel’s Parag Agrawal: Building a New Web for AI Agents

Parag Agrawal is making a bet that goes against two decades of web search: agents will query the web a thousand times more than humans ever have, and the infrastructure built around human clicks is…

Training Data · Aug 25

Stealing Reasoning Traces from Proprietary LLM APIs — Ilia Shumailov & Alexander Panfilov

Tim Scarfe speaks with Ilia Shumailov and Alexander Panfilov about their paper, Stealing Reasoning Traces from Proprietary LLM APIs.The core bug sounds deceptively simple: providers return encrypted…

ML Street Talk · Aug 22

AI in the AM — Weekly Highlights: Relaunch Week (Aug 17–20, 2026)

Relaunch week of AI in the AM brings together highlights from four live mornings and nine guests, centered on who checks frontier AI, how wide the gap is between lab-internal systems and public…

The Cognitive Revolution · Aug 22

The Evolution of the Agent Harness

Models keep absorbing the harness into their weights — soon, it will be a harness for human attention rather than for the model.

Latent Space · Aug 22

Simulation: the new Scaling Law — Joon Sung Park, Simile AI

Simile’s CEO about his journey from the viral Generative Agents to creating 8 Billion Digital Twins of every living human... and why it’s gone from fun exploration to very serious business.

Latent Space · Aug 21

The /wayfinder Skill: Navigating the “Fog of War” of Planning

Matt Pocock tells us about his /wayfinder skill, for greenfield projects or for when the way forward is unclear.

Latent Space · Aug 20

Every Exponential Ends — Silicon Valley Forgot — Adam Becker

Astrophysicist Adam Becker, author of "What Is Real?", joins Tim Scarfe to take apart the futures Silicon Valley keeps selling: the 2045 singularity, mind uploading, Mars colonies, and the AI…

ML Street Talk · Aug 20

From Restoring Sight to Reimagining the Brain, with Max Hodak

Max Hodak, co-founder and CEO of Science Corporation, joins Sarah Guo to discuss the future of vision, brain-computer interfaces, and the human experience. Max explains how Science’s PRIMA retinal…

No Priors · Aug 20

Frontier Model Cost and Open-Weights Popularity is Driving Demand for Model Routing

Glean CEO Arvind Jain explains why model routing helps control AI costs for organizations, and how human feedback loops at scale improve its routing systems.

Latent Space · Aug 18