OpenAI's AI Designed Its Own Chip in 9 Months — And It Beat Nvidia!

OpenAI’s new Jalapeño AI chip could become a serious challenge to NVIDIA’s dominance in AI inference. Co-developed with Broadcom, Jalapeño is OpenAI’s first custom AI inference chip, designed specifically to run large language models while using significantly less power than traditional GPU-based AI infrastructure. In this video, we break down the OpenAI Jalapeño chip, its HBM4 memory, AI inference performance, power efficiency, and benchmark results against NVIDIA AI GPUs. On the public InferenceX benchmark discussed in the video, Jalapeño delivered roughly 1.5–1.9× higher performance per watt at peak throughput and 1.7–3.6× lower end-to-end latency across models including GPT-OSS 120B, DeepSeek R1 and Kimi K2.5. We also explore how OpenAI used AI models to help design and optimize its own custom silicon.

But OpenAI isn't abandoning NVIDIA. Jalapeño is about reducing dependence on outside AI hardware and building a more vertically integrated infrastructure stack for future ChatGPT, AI agents, and OpenAI models. We also cover the latest Anthropic Claude developments, Alibaba’s new Qwen 3.8-Flash model, and the rapidly intensifying battle over AI models, custom chips, and inference costs.

#OpenAI #Jalapeno #AIChip #NVIDIA #Broadcom #ChatGPT #ArtificialIntelligence #AI