OpenAI's custom chip likely first of many, Wedbush says

OpenAI and Broadcom this week unveiled the Jalapeño chip, the startup's first custom AI design. Wedbush analysts called it a starting point, not a one-off. They expect more custom silicon to follow.

The chip targets inference – the stage where trained models answer user queries. OpenAI runs millions of those through ChatGPT and its API every day. Inference costs are its biggest operational line item after headcount. A chip purpose-built for that workload could cut per-token costs over time, the Wedbush note said.

Broadcom's ASIC team developed the design. That gives the chipmaker a marquee customer in the fast-growing custom chip market. Google has the TPU. Amazon runs Trainium and Inferentia. Microsoft has the Maia chip. OpenAI's entry puts Broadcom alongside those efforts.

Wedbush pointed to the strategic logic. OpenAI spent years relying on Nvidia's GPUs for both training and inference. Building its own silicon lets it optimize for the specific math its models use and avoid paying Nvidia's margin. The Jalapeño chip is the first step. Later designs could cover training workloads too, the analysts said.

Broadcom (AVGO) carries an Alpha Score of 56 out of 100, a Moderate rating, within the technology sector. The stock has already ridden the AI infrastructure buildout higher. The OpenAI partnership adds a long-term customer with deep pockets and a real need to control costs. Read more on the AVGO stock page.

Execution risk is real. Chip design and manufacturing timelines slip routinely. OpenAI did not disclose volume commitments or financial terms. Broadcom declined to comment beyond its public filings.

The Jalapeño chip is expected to reach production in the second half of the year.