Why everyone from OpenAI to SpaceX is building their own chips (and turning up the heat on Nvidia)
TL;DR
Nvidia still dominates AI accelerators, but major customers no longer want total dependence on one supplier, one roadmap, and one pricing model. OpenAI is working with Broadcom on Jalapeño, a custom inference chip, according to TechCrunch. That puts it alongside Google, Apple, and SpaceX in the custom-silicon push. This is not an instant break with Nvidia. It is a hedge: custom chips can give companies more control over cost, supply, and performance for specific workloads.
Nauti's Take
The headline sounds like a Nvidia killer story, but the reality is more practical: custom silicon mainly pays off for companies with massive, repeatable workloads and enough capital to absorb mistakes. For OpenAI, an inference chip can make real sense because inference cost hits product margins directly.
Nvidia still has deep advantages in training, software, and supply. The real shift is that AI giants do not just want to build models anymore; they want to control the factory behind them.
Briefingshow
AI is becoming an infrastructure game: running models at scale is less about software alone and more about compute, power, and reliable supply. Custom chips can lower costs and reduce dependence, but they are expensive, slow to build, and only make sense when workloads are large and predictable enough.
Sources