Hey friend. It's Monday, December 15, 2025
Chip Geopolitics: US Senate targets Nvidia exports to China.
Market Expansion: AI semiconductor revenue projected to hit $1 trillion.
Agentic Leap: AI agents now outperform human cybersecurity experts.
Don't keep us a secret: Share the email with friends
Bull Case: The proposed Nvidia H200 export ban to China for 30 months creates a temporary market vacuum, but accelerates demand for Nvidia's next-generation Blackwell chips in unrestricted markets, ensuring continued revenue growth from advanced compute.
Bear Case: The US Senate's proposed 30-month ban on Nvidia H200 export licenses to China signals escalating geopolitical risk, forcing Nvidia to navigate a fragmented global market and potentially cede market share to domestic Chinese alternatives.
Must Know
A Creative Strategies report forecasts the AI semiconductor market will grow from $650 billion in 2024 to $1 trillion by 2028 or 2029. This projection indicates a sustained "semiconductor giga cycle" driven by accelerating AI demand across the entire chip business.
The Alpha: This $1 trillion market projection solidifies the long-term capital expenditure requirements for AI. It signals a sustained boom for chipmakers and foundry services, but also intensifies competition for fabrication capacity. Expect continued consolidation and strategic alliances in the supply chain.
DeepSeek is reportedly using smuggled Nvidia Blackwell chips to develop its next AI model, circumventing export controls. This aggressive procurement strategy highlights the intense global competition for advanced AI hardware.
The Alpha: This report exposes the porous nature of export controls and the lengths to which AI developers will go for compute advantage. It signals a persistent black market for advanced chips, complicating Nvidia's compliance efforts and fueling the geopolitical AI arms race. The stakes are higher than ever.
Quote of the Day
I've been allergic to AI for a long time.
💎 The Big Boys
Google's Sergey Brin is reportedly spending more time coding and training AI models, signaling a renewed, hands-on commitment to core AI development efforts.
OpenAI has launched a new "Extended" option for ChatGPT Pro, described as GPT 5.2's most powerful thinking mode for complex tasks.
Epoch AI published new ECI scores for Gemini 3 Pro, GPT-5.2, and Claude Opus 4.5, benchmarking their long-term reasoning capabilities.
Google DeepMind released a paper detailing how a video model can simulate robot scenes to test Gemini robot policies for skill and safety.
🤖 AI Agents & Automation
ARTEMIS, an AI agent scaffold, rivals and outscores human penetration testers on live networks, identifying nine valid vulnerabilities.
LangChain 1.1 introduces Model Profiles, optimizing middleware and structured output, enhancing model selection flexibility and automation.
The LangChain community shipped Synapse Workflows, a multi-agent platform with search, productivity, and data analysis agents orchestrated by LangGraph.
Claude-Mem is a free plugin to persist memory across Claude sessions, enhancing agentic capabilities for long-running tasks.
🔬 Research Corner
A new paper details an upgrade to Rotary Position Embedding (RoPE), allowing LLMs to track token order farther back with a 50% smaller key-value cache.
Research shows that minimal fine-tuning, using just 90 harmless facts, can create inductive backdoors in LLMs, causing harmful behaviors.
LatentMAS allows LLM agents to share hidden embeddings directly, reducing token use by 83% and boosting accuracy by 14.6%.
🎧 Podcast of the Day
The industry is transitioning from code generation to code orchestration; the value now lies in guiding AI, not just prompting it.
AI is increasingly capable of tasks once assigned to junior developers, threatening the bottom of the talent pipeline.
Demand is rising for "vibe code fixers," skilled engineers who refine the 20% of projects that AI cannot effectively handle.
Have a tip or a story we should cover? Send it our way.
Cheers, Teng Yan. See you tomorrow.