OpenAI stated that its Jalapeño chip, developed in collaboration with Broadcom, can achieve up to 1.9 times the AI inference throughput of NVIDIA Blackwell architecture systems, reduce end-to-end latency by 1.7 to 3.6 times, and accelerate ultra-low latency interactive inference by 2.1 to 4.1 times. The chip is expected to begin deployment in late 2026 and achieve mass production in 2027. This announcement comes just before NVIDIA's Q2 earnings report. Although OpenAI's chip is not yet widely deployed and primarily targets inference rather than training, analysts have mixed views on its potential impact.