INTC "Just to push on that a little bit, you guys have your own factories. Why are you in the inventory situation that you’re in? And even if you look at it, I get the whole idea of finished goods versus other components, but you have $11.6 billion of inventory and it’s still not in the right place at the right time. How does that happen?" Intel's management team has been accumulating difficulties as always, and has been directly exposed and criticized ， So it's really hard to find a CEO who knows how to allocate cyclical capital. It's better to dismiss more people and continue the transformation Overall, the performance expectations are still too high, but the more concerned issue of strong CPU demand has been confirmed, and the supply side is not unable to increase production, so everyone cannot expand production in the short term. Looking at a previous report from Huatai: "From the progress of major overseas companies, Meta acquires manus, Anthropic releases Cowork, OpenAI's ChatGPT Agent, Google's Gemini CLI, Chrome Auto Browse; From a domestic perspective, Alibaba's Qianwen Agent and Doubao Mobile Assistant show that the mainstream agents of major manufacturers have been basically polished, and the PC and mobile personal life and work assistant oriented agents have become the mainstream product form. The penetration pace of C-end agents in 26 years may exceed expectations. From the Claude Code workflow perspective, during the startup phase (where only one project and a few small files are opened), the intelligent agent consumes approximately 1-2w tokens per thought; In the deep work stage, users open more than 20 files, and the agent accumulates a large number of attempt records (encoding, debugging, logging, comparing code differences), consuming 200000 tokens for each step. Claude Code recommends that users set a token limit of 20-30 million tokens per minute, which is sufficient to demonstrate the huge demand for computing power (compared to Chatbot, which requires about 2 interactions in 1 minute and consumes about 2k tokens per transaction). Overall, quantitatively speaking, completing a small project consumes millions of tokens, which is three orders of magnitude higher than a single Chatbot transaction. Based on the Openrouter data from the past week, there has been a significant increase in call volume for Claude 4.5, and the next token acceleration point will soon appear. The two major structural impacts brought by agents are storage and CPU. From the perspective of CPU, from the working principle of Agent, the importance of CPU has increased: 1) The server receives numerous inference tasks, and the first step is to batch, which is completed by the CPU; 2) Before entering the GPU, data needs to be decompressed, decrypted, and vectorized. For tail inference tasks that do not require parallel computing capabilities, the CPU will be the first choice; 3) Under the virtual machine technology roadmap, agents require CPUs to create a large number of sandbox environments, rewards, and validators, and to build various tool environments; 4) The allocation of Agent task steps and multi-threaded tool calls are executed on the CPU, and context storage is prioritized to overflow to the CPU. '' Computing power, transportation capacity, storage power, CPU power, it's really taking turns. Since the CPU demand has been confirmed, it actually feels more advantageous for INTC wafer fabs to adjust their production lines themselves. It's a bit like before when Micron: Tian Ji raced horses, cut down waiting horses, and supplied grain and forage to the top horses, thereby increasing production to high gross profit production lines Keep holding on