OpenAI and Broadcom's Jalapeño Chip Rewrites Inference Infrastructure
OpenAI and Broadcom unveiled Jalapeño, a reticle-scale AI inference ASIC built in nine months using OpenAI's own models to accelerate chip design — and the networking architecture inside the chip is what changes the inference cluster picture. Plus: NetBox goes read-write for AI agents with governance built in, ninety-three percent of organizations have already hit AI-caused infrastructure incidents, and a valleytronic photonic circuit from Nature Photonics that does something nobody has managed at room temperature before.
Welcome to Amaze Networks for Wednesday, June twenty-fifth. Quick question before we get into anything: if the company that consumes more AI inference compute than anyone else on the planet just built its own chip to do it, what does that tell you about the GPU market?
It tells you the GPU was always a general-purpose tool being used for a specific job. And at some point the specific job gets its own hardware.
That's the Jalapeño story, and it dropped yesterday from OpenAI and Broadcom. We'll start there because it's the kind of announcement that looks like a product launch but is actually a supply-chain strategy shift. Stick around — we've also got NetBox going read-write for AI agents with some genuinely interesting governance architecture, a survey that says ninety-three percent of organizations have already been burned by AI-generated infrastructure changes, and a science story out of Nature Photonics that does something valleytronic researchers have been trying to do for a decade.
New episodes, every weekday.
Amaze Networks drops at 4 AM CT, Monday through Friday. Spotify and Apple Podcasts submissions in progress.