Why everyone from OpenAI to SpaceX is building their own chips (and turning up the heat on Nvidia)
TL;DR
Nvidia still sets the pace for AI data centers, but major customers are building alternatives. OpenAI is working with Broadcom on Jalapeño, a custom inference chip meant to reduce single-supplier dependence. Google, Apple, and SpaceX are moving in the same direction with custom silicon: more control over cost, supply, and performance rather than a clean break from Nvidia.
Nauti's Take
This is not Nvidia’s obituary; it is a power move. If you spend billions on AI data centers, you do not want your entire business tied to one GPU supplier forever.
Custom chips are expensive, risky, and hard to scale, but they give OpenAI and others strategic insurance. The interesting shift is that once inference becomes the mass-market workload, efficiency matters more than raw GPU hype.
Briefingshow
The AI boom is constrained by power, chips, delivery timelines, and margins, not just model quality. Companies with their own hardware can tune models more tightly to infrastructure and negotiate harder with Nvidia. For the market, that means more competition, but also more vertical integration by the biggest platforms.
Sources