Reports

NVIDIA Develops AI Chips for China Amid Export Controls

Hak

23 Apr 2025 — 4 min read

Good morning

It's Wednesday, April 23, 2025, and you're reading the Agentive Daily Report, where we cut through the noise of the AI sphere to bring you what actually matters. Let's dive into what's caught our eyes the most today.

TL;DR for busy people

NVIDIA is reportedly developing custom AI chips for China using local supply chains to bypass US export restrictions
Meta is expanding AI-powered age detection on Instagram to place suspected teen users under stricter privacy settings automatically
Anthropic published a comprehensive framework for understanding and addressing AI harms across physical, psychological, economic, societal, and autonomy impacts
Google's new "Cell-to-Sentence" model can convert raw cell biology data into natural language, allowing researchers to "talk" to individual cells
AWS faces criticism over "arbitrary" limits placed on Anthropic's AI, raising questions about cloud provider control
ChatGPT search has grown dramatically in Europe, expanding from 11M to 41M monthly active users in just six months

Today's Top Stories

NVIDIA Developing China-Specific AI Chips to Bypass Export Controls

NVIDIA reportedly partnered with Chinese startup DeepSeek to develop new AI chips tailored specifically for the Chinese market after US Commerce Department export restrictions threatened its H20 chip sales. The new chips would rely entirely on China's local supply chain, including memory, processing nodes, and packaging, allowing NVIDIA to maintain its presence in a market that accounts for 13% of its total sales.

This move highlights the escalating tech sovereignty battle as nations seek to secure independent AI supply chains. For NVIDIA, which had already secured $18 billion in H20 chip deals in China this year, the stakes are enormous. However, rising tensions between Washington and Beijing mean this plan will likely face serious regulatory scrutiny from US authorities intent on controlling advanced AI hardware exports.

Meta Expands AI-Based Teen Protection on Instagram

Meta is rolling out enhanced AI systems to identify teenage users on Instagram, automatically placing suspected teens into restricted account settings with stricter privacy protections. The system analyses user behaviour patterns like birthday messages to infer real ages of users who may have misrepresented themselves, then switches those accounts to safer default settings that require parental approval to change.

This initiative comes amid mounting legal and regulatory pressure on Meta to better protect young users. With ongoing investigations in Europe and lawsuits in the US, the move demonstrates how platforms are increasingly deploying AI for safety and compliance rather than just engagement. The approach reflects a growing trend of using AI to solve content moderation and user protection challenges at scale, though questions remain about accuracy and user autonomy.

Anthropic Unveils Comprehensive AI Harm Assessment Framework

Anthropic has published a multidimensional framework for assessing and mitigating AI harms, addressing both catastrophic risks and everyday impacts. The approach evaluates five dimensions of potential harm: physical, psychological, economic, societal, and autonomy impacts, providing a structured methodology to guide policy, testing, and enforcement strategies.

This initiative represents one of the most thorough attempts by an AI developer to create a systematic approach to safety across different impact levels and timescales. By acknowledging both immediate harms and long-term risks, Anthropic is pushing the industry toward more nuanced responsibility standards. The framework complements Anthropic's Responsible Scaling Policy and demonstrates how frontier AI companies are working to establish governance structures ahead of potential regulation.

Other Developments Worth Noting

Google's Cell2Sentence Model: Google researchers have created a breakthrough model that translates complex cellular gene expression data into natural language, allowing scientists to "talk" to cells using plain English. This technology could transform biomedical research by making cell biology more accessible and enabling predictive modelling of how cells might respond to treatments.
AWS Rate Limits on Anthropic: Amazon Web Services is facing criticism over "arbitrary" rate limits applied to Anthropic's Claude AI service. The controversy highlights growing tensions between cloud providers and AI companies, raising questions about infrastructure control and fair access in the AI ecosystem.
ChatGPT Search Growth in Europe: OpenAI reported that ChatGPT search usage in Europe has grown from approximately 11 million to 41 million average monthly active users in just six months. This rapid adoption suggests ChatGPT is becoming a serious challenger to traditional search engines in the region.
AI Impacts on Google CTRs: Google's AI Overviews are significantly hurting click-through rates to websites, according to recent data, potentially threatening the traditional web ecosystem that relies on traffic from search engines.
o3 and o4-mini Hallucination Concerns: Recent analysis shows that OpenAI's GPT-o3 and o4-mini models hallucinate significantly more than older models, raising questions about the tradeoffs between reasoning capabilities and factual reliability in newer AI systems.

New Tools Discovered

Agent Simulate: A platform that lets developers test AI applications with thousands of simulated digital humans to identify potential issues before deployment.
Universal Memory MCP: An open-source tool that synchronises AI memories across different platforms, maintaining context between conversations with different AI assistants.
All Voice Lab: A service offering ultra-realistic AI voice synthesis and cloning with high fidelity output.
Open Codex: A fully open-source command-line AI assistant inspired by OpenAI Codex that runs locally using models like phi-4-mini with no API key required.
Shotup: A mobile app that turns your screenshots into a searchable knowledge base where you can ask questions about any image you've saved.

Discover more tools at Agentive.Directory

That's a wrap for today! Thank you for reading this report. Have thoughts on today's developments? Hit reply and let me know what you're thinking. Or if you've discovered a cool AI tool we should feature, drop me a line.

Until tomorrow,
Hak from Agentive.Studio