OpenAI Brings The Heat With New Chip (a Jalapeño Pointed Straight at Nvidia)

By Stocks News   |   4 months ago   |   Stock Market News
OpenAI Brings The Heat With New Chip (a Jalapeño Pointed Straight at Nvidia)

Huge miss if SammyGPT doesn’t make THIS the mascot… 

OpenAI has spent three years as one of the best customers Nvidia ever had, cutting eye-watering checks for GPUs every time demand went vertical. This week it pulled the cover off the chip designed to make those checks smaller, then named it after a pepper.

Meet Jalapeño, OpenAI and Broadcom's first piece of joint custom silicon and the ChatGPT maker's debut swing at its own AI hardware. The two have been bedding it up together for 18 months, went public with the plan back in October, and on Wednesday handed OpenAI a physical sample. They're calling it an "Intelligence Processor," the first "AI accelerator" in a platform built to make advanced AI "faster, more reliable, and more accessible to more people." (The poors, generously, included.) The bigger goal, in OpenAI's words, is to stop renting its brain from other people's chips build the full stack behind its own models.

In case you missed the nature of this friends with benefits relationship, Broadcom makes the thing, OpenAI runs the inference… a.k.a, the unglamorous, compute-hungry job of actually serving its models to the hundreds of millions of people typing into ChatGPT all day. Jalapeño is an ASIC, which in plain English means it's cheaper and less flexible than Nvidia's GPUs, built for one task instead of all of them. Translation: Until it ships, it’s a Temu-Nvidia GPU. 

Of course, OpenAI president Greg Brockman says owning more of the stack lets the company "serve more intelligence with greater efficiency." (Greater efficiency here meaning fewer nine-figure invoices from Nvidia.) Which tracks. OpenAI lit the fuse on this entire boom in 2022 and has since become one of the planet's hungriest buyers of Jensen's famously pricey silicon, and nobody wants to pay the king's fat taxes forever.

Also worth noting, Jalapeño isn't OpenAI’s only side piece. The company has signed for AMD chips, leaned on Amazon's Trainium silicon, and cut a deal with Cerebras, the dinner-plate-wafer crowd that IPO’d in May. The Broadcom build alone is meant to scale into racks of OpenAI-designed chips drawing 10 gigawatts of power, and OpenAI has the dry powder to fund all of it. Ashtray money, bro. 

Which still leaves Nvidia selling shovels to the guy trying to build a shovel factory. The call, increasingly, is coming from inside the data center. Vertically integrating your own silicon is, in fairness, deeply based, and every lab Broadcom helps roll its own chips is one more customer learning it doesn't have to rent forever.

But, but, but… Nvidia bulls will tell you an ASIC can't touch a real GPU, that inference is a sideshow, that the incumbent isn't going anywhere. Maybe. Or maybe that's just CopeGPT talking. Regardless, OpenAI's homegrown pepper doesn't hit the racks until the end of 2026, and even then only for inference. Meaning, until the kitchen actually opens, the company building the chip to need less Nvidia will keep ordering Nvidia by the crate. Until next time, friends… 

 

At the time of publishing, Stocks.News holds positions in Amazon as mentioned in the article. 

 

Did you find this insightful?

Disclaimer: Information provided is for informational purposes only, not investment advice. We do not recommend buying or selling stocks. Stock price discussions are based on publicly available data. Readers should conduct their own research or consult a financial advisor before investing. Owners of this site have current positions in stocks mentioned throughout the site, Please Read Full Disclaimer for details Here https://app.stocks.news/page/disclaimer