OpenAI says its AI chip rivals Nvidia in inference

OpenAI built the chips in record time by using its own AI models to speed up the design and verification phase of development. OpenAI unveiled benchmarks for its Jalapeño inference chip Tuesday and they are… fast. So fast that they compete with Nvidia’s best chips. OpenAI built the chips in record time by using its own AI models to speed up the design and verification phase of development. It still needed to work with Broadcom to build it, and it will still rely on fabs like TSMC to make it, but it’s a sign of things to come. If AI companies can use their own models to rapidly develop chips which they can then use to run more powerful AI models to develop the next chip, that’s a flywheel that leads to even faster capability acceleration. Sure, there are still bottlenecks galore (memory, power, a limited number of fabs). But the market will iron those out in time. It also shows how disruption is coming across industries. Smaller companies will soon learn that they can be more ambitious and more bold, challenging their larger competitors. And while that might be most visible in software, it is quickly spreading to the physical world.