## 🔍 Key Insights AI infrastructure is experiencing a **dual shock**: an **ASIC hardware revolution** and a **precipitous decline in inference costs**. The **Taalas HC1 chip** achieves **17,000 tokens/sec** inference throughput, priced as low as **$0.0075 per million tokens**. Concurrently, **NVIDIA** has pivoted to strategic capital alignment—directly investing **$30 billion into OpenAI**, signaling a shift from being a 'pick-and-shovel' enabler to a core co-builder of the AI ecosystem. ## 🚀 Top Highlights - **Taalas launches the HC1 dedicated AI chip**: Built on an extreme ASIC architecture featuring weight-frozen silicon, it achieves ultra-high inference throughput of **17,000 tokens/sec**. - **AI video generation cost drops below $10**: Cost per video now stands at just **$9.6**, prompting Hollywood screenwriters to question the existential viability of traditional film and TV production paradigms. - **NVIDIA's $30 billion investment in OpenAI**: Abandoning its prior trillion-FLOPS-scale compute partnership framework, NVIDIA joins OpenAI's latest funding round as a cornerstone equity investor. - **Beihang University open-sources Code2Bench—a new benchmarking paradigm**: Introducing **dual-scaling dynamic evaluation** (assessing both code source quality and test rigor), it ends the era of large models 'gaming the scores' through passive benchmark optimization. - **Claude Code deeply integrates Git Worktree**: Enables AI agents to execute tasks in isolated working directories—significantly reducing conflict risks in large-scale code operations. - **Peking University releases AgentRob**: A novel framework that leverages online forums as intermediaries (via the **MCP protocol**) to enable human → LLM → physical robot cross-modal, asynchronous, and persistent task orchestration. - **Open-source plugin claude-review-loop launches**: Establishes an automated 'implement–review–fix' loop for Claude Code, enhancing industrial-grade code usability via Codex CLI integration. - **SaaS companies redefine their AI moats**: Industry-wide critique of the 'survival test' framework intensifies, with renewed focus on how AI fundamentally reshapes product stickiness and value-capture logic.