Tesla taxi's hit Austin, memory beats RAG & more (January 23, 2026)

Good morning,

AI power is consolidating across chips, autonomy, and regional strategy. China is quietly lining up Nvidia supply, Tesla is removing humans from the loop, and Google is leaning on local partners to win abroad.

Let’s dive in 👇

━━━

Powered by Aloa • AI Research & Education Hub

━━━

Want to engage in conversation about AI news? Join the Byte-Sized Reddit Community

🤖 Chips, Cloud & Global Strategy

🚢 China Preps Nvidia H200 Orders

China tells Alibaba, top tech firms to prepare orders for Nvidia’s H200 AI chips, signaling quiet approval despite U.S. export controls. The move suggests Beijing is prioritizing inference capacity while navigating regulatory gray zones. It reinforces Nvidia’s central role in China’s near-term AI compute strategy.

☁️ Railway Raises $100M to Challenge AWS

Railway secures USD100 million to build an AI-native cloud aimed at developers shipping agent-heavy workloads. The company is positioning itself as simpler and faster than traditional hyperscalers. This reflects growing demand for infrastructure designed specifically around AI applications.

🇯🇵 Google Backs Sakana AI for Gemini

Google bets on Sakana AI to accelerate Gemini adoption in Japan through a strategic partnership. Sakana focuses on culturally aligned, Japan-first AI models where U.S. labs often struggle. The move shows Google favoring local partnerships to win regional AI markets.

🚗 Models, Agents & Monetization

🚕 Tesla Launches Driverless Robotaxi Rides

Tesla launches robotaxi rides in Austin with no human safety driver present. This marks a shift from demos to real passenger operations under live conditions. Regulatory pressure is expected to increase as edge cases surface.

🧠 MemRL Beats RAG Without Fine-Tuning

MemRL outperforms RAG on complex agent benchmarks using structured memory instead of retrieval pipelines. The results suggest memory-centric architectures scale better for long-horizon reasoning. This challenges RAG’s dominance in agent system design.

💰 DeepMind CEO Questions ChatGPT Ads

Google DeepMind CEO is surprised OpenAI is pushing ads inside ChatGPT so early. He frames monetization as premature relative to trust and alignment. The comment highlights a widening philosophical divide between Google and OpenAI.

🛠️ Tools of the Day

→ Tonkotsu – AI-native backend framework for fast agent builds
→ Qwen3 – Alibaba’s open model focused on reasoning and multilingual tasks
→ RightNow AI – Real-time AI decisioning for operations and support

⚡ Quick Hits

→ Google begins offering free Gemini-powered SAT prep
→ Humans thinks coordination is AI’s next frontier
→ Microsoft adds AI features to Paint and Notepad
→ Xebia owner weighs $1B AI consultancy sale
→ Former Sequoia partners launch AI calendar negotiator
→ Inferact lands $150M to commercialize vLLM

🧠 TLDR

China is lining up Nvidia H200 chips while the U.S. debates controls. Tesla removed the human safety driver as autonomy moves into real-world deployment. At the same time, memory-based agents challenge RAG, and Google distances itself from OpenAI’s ad-first approach by doubling down on regional AI partnerships.

Cheers,
David