AI Weekly Digest - Week 42, 2025

Table of contents

OpenAI and Broadcom Unveil Strategic Partnership for Custom AI Chips

Category: Infrastructure and Hardware
Source: openai.com

OpenAI has announced its much-anticipated collaboration with Broadcom to develop custom AI accelerators and networking systems. This strategic partnership will deploy 10 gigawatts of OpenAI-designed chips, rolling out starting 2026 and completing by 2029. The move aims to create a hyperscale cloud infrastructure tailored for OpenAI’s workloads and break NVIDIA’s stranglehold on the AI chip market, which controls over 90% of current supply.

OpenAI leveraged its own AI to optimize chip design, achieving significant silicon area reductions in record time. The company’s growing needs—now serving over 800 million weekly users on a 2 GW cluster—are driving this push toward 30 GW capacity, nearly the equivalent of Australia’s entire energy consumption. While industry analysts warn this could fuel an “AI bubble”, the initiative marks a major leap toward self-sufficient, scalable, and competitive AI.

Stanford AI Designs Viruses That Kill Antibiotic-Resistant Bacteria

Category: Healthcare and Life Sciences
Source: MIT Technology Review

In a breakthrough that could rewrite the future of medicine, Stanford scientists have used AI to design viruses that target and destroy antibiotic-resistant bacteria—specifically E. coli. The Evo AI model was trained on the genetic code of nearly 2 million bacteriophages and proposed hundreds of novel viral DNA sequences. After chemical synthesis, 16 of the 302 designs successfully killed bacteria in lab tests.

With antibiotic resistance escalating—WHO reports that 1 in 6 infections are now drug-resistant—these AI-designed phages present a vital new tool. Fast iterative design via AI robotics and purpose-built labs (such as Liverpool’s new £20M facility) means treatments can be developed and deployed much more rapidly than traditional drugs. While safety concerns exist around the potential misuse of such technology, experts agree the potential for phage therapy powered by AI could save millions of lives and stave off a looming public health crisis.

Japan and Hollywood Push Back on AI-Generated Copyrighted Characters

Category: Regulation and Policy
Source: IGN

The Japanese government issued a formal request for OpenAI to refrain from generating Sora 2 videos using the likenesses of copyrighted anime and video game characters. Officials describe such cultural icons as “irreplaceable treasures,” and warned that further action under the new AI Promotion Act may follow if infringement continues. This regulatory stance comes amid a flood of AI-generated fan videos and mounting pressure from major copyright holders, including Nintendo and The Pokémon Company.

The case has sparked international debate, with Hollywood studios and agencies also voicing concern over unauthorized use of actors’ likenesses and creative IP. Lawsuits against AI firms are already underway, while open questions remain over opt-in or opt-out mechanisms and the legal status of AI-generated “fan fiction.” OpenAI CEO Sam Altman promised new controls to allow rights holders to specify how their characters can be used, but pressure continues to mount for more robust solutions.

NVIDIA DGX Spark: Petaflop Desktop Supercomputer Launches

Category: Infrastructure and Hardware
Source: NVIDIA.com

NVIDIA’s DGX Spark, a desktop AI supercomputer boasting up to 1 petaflop of FP4 performance and 128GB unified memory, has officially landed on developer desks worldwide—including SpaceX and OpenAI. Built on the new Grace Blackwell architecture, it marks a shift from data center-centric AI to locally available, hands-on computation for researchers, artists, and startups.

DGX Spark supports NVIDIA’s full AI stack—including CUDA, pretrained models, and NIM microservices—enabling rapid prototyping, fine-tuning, and advanced multimodal agent development without cloud dependency. Major PC makers like Dell, HP, and Lenovo have announced launch partnerships, and early reviews highlight its game-changing potential for speeding up AI workflows and agentic intelligence at the desktop.

Walmart Partners with OpenAI for ChatGPT-Powered Shopping

Category: Retail and eCommerce
Source: OpenAI

Walmart has joined forces with OpenAI to allow customers to shop directly through ChatGPT. Shoppers can find and purchase products via natural language conversations, making the experience highly interactive and tailored. This collaboration signals a broader trend in retail toward personalized, conversational commerce powered by AI, aiming to transform both online and in-store experiences for millions of customers.

Dfinity Launches Caffeine: Build Production Apps with Natural Language

Category: Tools and Platforms
Source: VentureBeat

Dfinity Foundation has unveiled Caffeine, an AI platform where users describe their app needs in plain English and receive fully built, production-grade web applications—no coding required. Powered by the Motoko programming language and the decentralized Internet Computer Protocol (ICP), Caffeine promises mathematical guarantees on data integrity and seamless upgrades, making software creation accessible to non-technical users and ensuring robust security.

Caffeine’s architecture could reshape software development, creating a ‘self-writing internet’ that democratizes creation and lowers barriers for enterprises and individuals. Real-world results and scalability tests remain ahead, but the potential for rapid innovation is high.

Oracle and NVIDIA Accelerate Sovereign AI for Abu Dhabi’s Digital Government

Category: Enterprise and Adoption
Source: NVIDIA.com

Abu Dhabi’s Department of Government Enablement, in partnership with NVIDIA, Oracle, Deloitte, and Core42, is pioneering a sovereign AI-native government transformation by 2027. With a 13B AED investment, the strategy uses Oracle Cloud Infrastructure and NVIDIA’s AI stack to automate public services, offer intelligent multilingual assistants, and maintain strict data sovereignty within the UAE. The initiative demonstrates how public sector organizations can modernize securely and at scale, with projections to boost GDP and generate thousands of jobs.

NVIDIA and Oracle Unveil Zettascale10 AI Cluster for Data Processing

Category: Infrastructure and Hardware
Source: Oracle

At Oracle AI World, Oracle launched its OCI Zettascale10 computing cluster accelerated by NVIDIA GPUs, capable of 16 zettaflops for high-performance AI training and inference. Integrations with NVIDIA’s NIM microservices, NeMo Retriever, and RAPIDS Data Accelerator streamline AI database workflows and enterprise video understanding—powering next-gen business data platforms with robust security, scalability, and speed.

Inclusive AI: Face Recognition Tech Fails Millions With Facial Differences

Category: Ethics and Governance
Source: WIRED

As facial recognition becomes ubiquitous for ID, security, and banking, an estimated 100 million people with facial differences are encountering barriers to essential services. Biased AI training datasets exclude non-typical faces, causing discrimination and exclusion. The issue underscores the need for inclusive data and thoughtful policy to ensure equitable technology access.

ChatGPT’s ‘Adult Mode’ and Shift Toward Unfiltered Content

Category: Regulation and Policy
Source: Reuters

OpenAI announced the launch of Adult Mode for ChatGPT, granting verified users aged 18+ access to mature content (including erotica) from December onward. CEO Sam Altman highlights the move as part of OpenAI’s “treat adults like adults” principle, with new safety tools mitigating mental health risks. While promising greater personalization and utility, the shift reignites debate about content moderation, AI safety, and legislative proposals around AI companions.

Hollywood vs. OpenAI: Legal Backlash Against AI-Generated Celebrity IP

Category: Ethics and Governance
Source: Hollywood Reporter

Hollywood agencies and studios have rebuked OpenAI’s Sora 2 for use of actor likenesses and copyrighted properties without explicit consent, prompting lawsuits and memos to clients warning of IP risks. The controversy spotlights intensifying legal and ethical debates about AI’s use and monetization of creative content, with studios weighing legal remedies and new licensing models.

Short News Highlights

Samsung’s Tiny Recursive Model Outperforms Giants
Samsung’s new 7M-parameter Tiny Recursive Model beats vastly larger models like DeepSeek R1 and Gemini 2.5 Pro on ARC-AGI tests, proving that architecture and recursive reasoning can rival brute-force scaling. Details

NVIDIA Unveils Vera Rubin NVL144 and Kyber AI Racks for Next-Gen Data Centers
NVIDIA, with more than 50 partners, is rolling out modular, energy-efficient rack architectures for gigawatt-scale AI factories—standardizing hardware for a new era of high-density, liquid-cooled GPU deployments.

Kids Treat AI as Friends: Social, Emotional Implications Rising
New online discussions highlight young users treating AI companions as friends, sparking debate on emotional development, socialization, and ethical risks.

Walmart’s ChatGPT Shopping Integration
Walmart now lets shoppers purchase directly via ChatGPT, signaling a retail revolution.

OpenAI to Allow Erotica and Adult Content
OpenAI CEO Altman confirms erotica and unfiltered content for verified adults as part of ChatGPT’s December update, sparking regulatory conversations.

Japan’s Crackdown on Copyrighted Anime Characters in AI Videos
Japan continues to request OpenAI restrict Sora 2 video generation of copyrighted anime/game characters; potential legal and diplomatic fallout looms.

Face Recognition Systems Still Exclude Diverse Populations
Wider attention drawn to biases in AI facial recognition affecting hundreds of millions who have atypical facial features.

Hollywood Legal Fights Over AI-Created IP
Ongoing lawsuits erupt as studios push back on unauthorized celebrity likenesses and IP used by AI platforms.