Anthropic has begun involving Claude in the design of the chips that will power future versions of Claude. According to The Information, Anthropic has assembled a team of approximately 15 engineers who are using Claude to assist in designing custom AI chips. It is reported that Claude has already significantly reduced the time required for certain design tasks. Interestingly, even before the chip has entered mass production, Anthropic has already shared this development with potential IPO investors. I think this signals a noteworthy shift. In the past, AI companies grew stronger primarily by buying more chips and hiring more engineers. Now, they are beginning to use their own models to improve the efficiency of chip development. If this approach succeeds, it could create an intriguing feedback loop: stronger models help design more efficient chips, and more efficient chips lower the cost of training and inference for the next generation of models. Of course, faster design does not guarantee successful tape-out, nor does it imply the ability to challenge NVIDIA. Chip verification, manufacturing, and mass production remain entirely different challenges. But for investors, the story Anthropic is now telling has changed. Previously, it was: “Give me more money, and I’ll buy more compute to train stronger AI.” Now, it’s becoming: “Give me more money, and my AI will help me design the next generation of compute.” The most imaginative business model for AI companies may not be selling intelligence to others—but using intelligence to transform themselves first.