Broadcom이 분석한 OpenAI의 Jalapeño 추론 최적화 칩은 범용 GPU 대비 성능과 비용 이점을 강조한다.

OpenAI has partnered with Broadcom to develop Jalapeño, its first custom inference chip. The collaboration represents a far more cost- and power-efficient approach to AI monetization compared to relying on NVIDIA GPUs.

Designed using an AI-optimized architecture, Jalapeño delivers lower latency, higher throughput, and industry-leading performance while consuming significantly less power than existing alternatives. The chip costs less than half the price of comparable GPUs.

The evolution of AI infrastructure can be understood in two phases. Phase I, now mature, focused on GPU compute capacity for large language model training. Phase II is shifting focus to inference—a substantially larger market opportunity. Broadcom's XPU business is positioned to capture significant value in this transition.

Broadcom projects $350 billion in AI semiconductor revenue over the next two years, driven primarily by TPU and XPU designs for Google, OpenAI, and Anthropic.

Following Q3 results—marked by net income, EPS, and free cash flow growth exceeding 95% year-over-year—combined with the new two-year $350 billion AI semiconductor revenue guidance, the case for Broadcom remains compelling.