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 the biggest customers no longer want their compute plans tied entirely to one supplier’s prices, capacity and roadmap. OpenAI is reportedly working with Broadcom on Jalapeño, a custom inference chip. Google, Apple and SpaceX are part of the same broader move toward in-house or tightly customized silicon. This is less about ditching Nvidia overnight and more about hedging: better control over costs, availability and performance for narrow, high-volume workloads.
Nauti's Take
This sounds like a Nvidia killer story, but it is more realistically a pressure valve. Nvidia will not lose the market tomorrow, yet the myth of total irreplaceability is starting to crack.
For OpenAI and peers, this is less about hardware pride and more about margins, supply control and negotiating leverage. The hard part: a chip is not a blog post.
Custom silicon means years of operational complexity.
Briefingshow
AI infrastructure is becoming a strategic control layer, not just a cloud procurement problem. Companies running models at massive scale do not want every bottleneck, price move and product decision dictated by Nvidia. Custom chips shift some power back to the platforms, but only for companies with huge capital, deep talent and enough volume to justify the bet.
Sources