BlockBeats news, on August 26, Citrini analyst Jukan published an analysis suggesting that OpenAI's "Jalapeño" AI chip may not reach the scale of NVIDIA's Rubin platform. He noted that if OpenAI's HBM4 high-bandwidth memory relies entirely on Samsung for supply, its production ramp-up could face certain capacity limits.
Jukan further noted that if OpenAI wishes to scale up the deployment of "Jalapeño," it may need to lower requirements for specifications such as HBM4 transfer speed, enabling the adoption of HBM4 products from other suppliers and alleviating supply chain constraints.
BlockBeats reported earlier today that OpenAI has released the latest test results for its in-house AI inference chip, Jalapeño, showing that it outperformed NVIDIA's GB200 and GB300 superchips on the InferenceX benchmark. On models such as GPT-OSS 120B, DeepSeek R1, and Kimi K2.5 1T, the chip delivers 1.5 to 1.9 times the AI workload per watt compared to competitors and reduces end-to-end latency by 1.7 to 3.6 times.