Nvidia’s grip on the AI chip market is loosening. OpenAI just revealed Jalapeño, its first custom inference chip built in partnership with Broadcom, joining a growing wave of tech companies designing their own silicon.
The trend isn’t about ditching Nvidia entirely — it’s about hedging. Custom chips mean more control, hardware tuned to specific workloads, and the kind of performance gains Apple unlocked when it moved away from Intel. Google, Apple, and SpaceX are all pursuing similar strategies.
OpenAI’s Jalapeño is an inference chip, meaning it’s optimized for running trained models rather than training them. This is a strategic move: as AI models grow larger and more expensive to run, inference costs become the bottleneck. Custom silicon can deliver better performance per watt and per dollar.
The announcement, discussed on TechCrunch’s Equity podcast, signals a broader shift in the AI hardware landscape. Companies that once relied entirely on off-the-shelf GPUs are now investing in custom architectures to reduce supplier risk and gain competitive advantages in cost and performance.
For the industry, this means the era of one-size-fits-all AI hardware is ending. The companies building the most advanced AI systems are increasingly the same ones designing the chips those systems run on.
