level: business
nvidia has long been the main supplier of ai chips, but that is starting to change. openai recently announced jalapeño, a custom inference chip built with broadcom. this move puts openai in a group with google, apple, and spacex, all of which are making their own silicon. the aim is not to completely replace nvidia but to have an alternative. custom chips let companies tune hardware to their exact needs and avoid relying on a single source.
building custom silicon gives companies more control and can lead to better performance. apple showed this when it switched from intel to its own chips. for ai workloads, custom designs can be faster and more efficient. openai's jalapeño chip focuses on inference, the process of running trained models. by handling this in-house, openai can potentially lower costs and speed up its services. other firms are likely watching these efforts closely.
the trend toward custom chips could reshape the ai hardware market. if more big players design their own, nvidia's dominance may weaken. however, nvidia still has a strong ecosystem and years of experience. the shift is gradual, and many companies will keep using nvidia gpus for training. still, the growing list of firms investing in custom silicon signals a long-term change in how ai infrastructure is built and managed.
why it matters: custom ai chips can lower costs and improve performance for data-heavy tasks, making ai services more efficient and accessible.