OpenAI has just announced the launch of its first self-developed AI chip, "Jalape ñ o"! Collaborated with @ Broadcom to design from scratch, tailored specifically for LLM inference. Early testing showed that the power efficiency is significantly better than current top accelerators, with a development cycle of only 9 months. This is a crucial step for OpenAI's "end-to-end layout" from models, products to infrastructure, with the goal of making ChatGPT, APIs, and other services faster, cheaper, and more reliable for global users. Is the AI industry about to enter a new era of customized chips? The greatest and most direct positive impact on OpenAI: Significantly reducing costs and dependencies: Currently, one of the biggest expenses for AI companies is "inference" (the computation that actually answers user questions). Jalape ñ o is a customized ASIC (non general-purpose GPU) optimized for this aspect, and early testing has shown that the special effects per watt can be significantly better than existing top solutions (including Nvidia Blackwell). This can significantly reduce long-term operating costs. • Supply chain autonomy and scalability: Reduce high dependence on Nvidia GPUs (Nvidia supply was once scarce), and collaborate with Broadcom+Celestica to begin large-scale gigawatt level deployment by the end of 2026. Help OpenAI expand ChatGPT and future Agentic products faster to serve more users. Full end strategic advantage: OpenAI is no longer just a "model company", but is moving towards a complete stacking layout from chips, cores, servo systems to products. This is the path that Google (TPU), Amazon, and Meta have already taken, which will bring greater competitiveness and bargaining power in the long run. Technological Leadership Demonstration: Using our own model to accelerate chip design, we completed tape out in just 9 months, demonstrating strong execution capabilities and laying the foundation for future generations of chip platforms. Impact on other stakeholders: Broadcom: A Major Victory! Acquiring OpenAI as a heavyweight customer, combining chip and network technology (Tomahawk), will significantly increase future revenue and market share. Nvidia: Long term warnings and challenges. Although Nvidia still has advantages in training and versatility, the inference market is beginning to be dominated by customized ASICs (cheaper and more energy-efficient). This is a clear action of 'Strike at Nvidia'. The entire AI industry: accelerating the trend of "self-developed chips". More companies will follow suit, and in the future, AI computing costs may decrease and efficiency may improve, indirectly benefiting end users and developers. However, in the short term, it is still necessary to observe actual performance data and production progress. Summary: For OpenAI, this is a very positive and strategic step that helps to reduce costs, enhance control and scalability in the long term, and is a key milestone in transforming from a "model leader" to an "AI infrastructure leader".