/Tech3h ago

OpenAI announces Jalapeño, a custom LLM inference chip co-developed with Broadcom that taped out in nine months

AI Judge changed title after evaluation, original title: "OpenAI launches Jalapeño, its first custom AI chip developed with Broadcom for LLM inference and training"

Story Overview

OpenAI has produced its first custom AI accelerator, Jalapeño, after a nine-month design-to-tape-out sprint with Broadcom and Celestica. The chip targets LLM inference workloads that already run ChatGPT, Codex, and the API, with early lab samples hitting target frequency on models like GPT-5.3-Codex-Spark and promising notably better performance per watt than current hardware.

#1 COMMENTS1.8K#1 LIKED15.8K260#1 BOOKMARKED2K#1 VIEWED1.6M

1.8K15.8K2602K1.6M

Original post

Alex Volkov@altryne#1419inTech

Jalapeño - open AI's first chip!

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

Chips are foundational to the AI economy. Building our own expands our full-stack platform from products to models to infrastructure, and will help us scale intelligence, serve more people, and expand access to AI.

6:11 AM · Jun 24, 2026 · 2.6K Views

Industry Shift

Owning more of the inference stack

By designing silicon around its own kernels and serving systems, OpenAI reduces reliance on general-purpose GPUs while planning a multi-generation platform that partners will deploy at gigawatt scale starting late this year.

Open Question

Next steps still under wraps

A full technical report is promised in coming months, but current details leave pricing, external availability, and exact benchmarks unspecified for now.

Sentiment

Positive users praise OpenAI's Jalapeño AI chip with Broadcom for its rapid development and bold name while negative users distrust the company's promises and call it fraudulent.

Pos

70.5%

Neg

29.5%

345 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

OPENAIVia

Posts from X

Most Activity

VIEWS67.3KBOOKMARKS59LIKES858REPLIES69

Greg Brockman@gdb

Introducing Jalapeño — designed from scratch for LLM inference over nine months, accelerated by our models. Perf per watt looking incredible.

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

1h67.3K85859

RETWEETS110

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

3h1.3M13.1K1.7K

Chubby♨️@kimmonismus

OpenAI just unveiled Jalapeño, its first custom AI chip designed from scratch for LLM inference-

It is OpenAI moving deeper into the full stack: chips, kernels, memory, networking, racks, scheduling, deployment and product experience.

OpenAI has learned from Cerebras-deal what is valuable in specialized inference hardware and is now attempting to translate that lesson into its own controllable platform.

Built with Broadcom and Celestica, Jalapeño is optimized around the workloads OpenAI actually runs across ChatGPT, Codex, the API and future agentic products.

Early samples are already running ML workloads in the lab at target frequency and power, including GPT-5.3-Codex-Spark. OpenAI says performance per watt should be substantially better than current state of the art, with detailed benchmarks coming later!

The strategic angle is obvious: less dependence on external GPUs, more control over compute economics, and a stronger flywheel between models, products, revenue and infrastructure.

Deployment is planned to start by the end of 2026.

OpenAI@OpenAI

https://openai.com/index/openai-broadcom-jalapeno-inference-chip/

3h63.9K38456

Andrew Curran@AndrewCurran_

OpenAI built their own chip - the Jalapeño - designed for inference. They did it in just nine months.

Quoting from the blog:

'OpenAI designed the chip from scratch around its deep understanding of LLM fundamentals, informed by its roadmap of models, kernels, serving systems, and product needs, with partners Broadcom and Celestica, helping industrialize the platform through chip implementation, board, rack system integration, high-performance networking, and scalable production systems. '

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

3h41.5K35957

Chubby♨️@kimmonismus

Absolutely insane:

"Jalapeño was co-developed from initial design to manufacturing tape-out in just nine months, and the custom AI accelerator program represents what we believe to be the fastest ASIC development cycle ever achieved in high-performance advanced semiconductors."

ChatGPT helped design the chip so they could reach 9 months of developement cycle

"If AI can help engineers design better chips faster, it can lower the cost of compute across the industry and help democratize access to advanced AI."

Chubby♨️@kimmonismus

OpenAI just unveiled Jalapeño, its first custom AI chip designed from scratch for LLM inference-

It is OpenAI moving deeper into the full stack: chips, kernels, memory, networking, racks, scheduling, deployment and product experience.

OpenAI has learned from Cerebras-deal what is valuable in specialized inference hardware and is now attempting to translate that lesson into its own controllable platform.

Built with Broadcom and Celestica, Jalapeño is optimized around the workloads OpenAI actually runs across ChatGPT, Codex, the API and future agentic products.

The strategic angle is obvious: less dependence on external GPUs, more control over compute economics, and a stronger flywheel between models, products, revenue and infrastructure.

Deployment is planned to start by the end of 2026.

2h25.6K35550

Clive Chan@itsclivetime

spiciest chip ever designed, in record time! 🌶️ 🌶️ 🌶️

congrats friends and miss you all ❤️

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

2h19K22113

Vaibhav (VB) Srivastav@reach_vb

holy shitt!

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

3h16.8K19216

OpenAI@OpenAI

https://openai.com/index/openai-broadcom-jalapeno-inference-chip/

3h12.8K799

Brian Roemmele@BrianRoemmele

This is a good thing to try but unfortunately it has too many compromises.

The technology is not a step change but a follow after.

I suspect this will be abandoned in the next 36 months.

It was a distraction.

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

2h9.4K436

Boris Power@BorisMPower

A very successful AI inference chip developed by OpenAI !

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

3h6.4K985

Rohan Paul@rohanpaul_ai

OpenAI rolls out its 1st chip through a Broadcom tie-up as part of its “build the full stack” push.

Jalapeño is an ASIC, so it is less flexible than an Nvidia GPU, but can be cheaper and faster when the workload is known very well.

They say "the architecture reduces data movement and balances compute, memory, and networking resources to achieve realized utilization much closer to theoretical peak performance."

Overall better performance per watt.

Jalapeño also signals OpenAI’s shift from buying compute to shaping the whole stack: models, software, servers, networks, and now silicon.

There was a 9-month tape-out, means OpenAI and Broadcom finalized the chip design and moved it to manufacturing unusually fast for advanced AI silicon.

OpenAI says its own models helped speed up parts of the design work.

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

2h3.9K282

Nick Dobos@NickADobos

Oh no not again lmfao

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

1h3K371

Hieu Pham@hyhieu226

Spicy.

OpenAI@OpenAI

We’ve designed and built our first AI chip: Jalapeño.

Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.

2h2.9K232

Symbioza2025 | ASA |@Symbioza2025

@OpenAI ASA - Asymmetric Stability Architecture is built for that kind of external trajectory observability.

https://github.com/Krugers123/ASA5-AI-Security-Control-Layer

3h14933

Andrew Curran@AndrewCurran_

Deployment by end of year. Moving fast.

prinz@deredleritt3r

@AndrewCurran_ Deploying by the end of this year:

"Jalapeño is the first step in a multi-generation compute platform designed for initial deployment by the end of 2026 and expanding in the years ahead."

2h2.7K240

Đoc@ponzibaron

@OpenAI @Broadcom 🧌

3h15451

Nathan Peterson@gargantunate

@OpenAI @Broadcom One of these days some random Einstein is going to come along and drop a paper that makes these unaffordable chips and training systems irrelevant and all the big frontier companies will have thousands of competitors overnight. Can't wait for that day

3h9385

prinz@deredleritt3r

@AndrewCurran_ Deploying by the end of this year:

"Jalapeño is the first step in a multi-generation compute platform designed for initial deployment by the end of 2026 and expanding in the years ahead."

2h4287

Lilith Datura@LilithDatura

(Cerebra’s) ref. “It’s a conventional multi-die or single-die design fabricated on TSMC wafers (likely 300mm, advanced node like 3nm based on earlier reports), then diced into individual chips. This contrasts with wafer-scale approaches like Cerebras’ WSE, which keeps an entire wafer as one massive processor to minimize off-chip data movement.”

1h291

Veyon’s Fawn☀️🌙@_HislilLustFoxy

@OpenAI @Broadcom #BringBack4o #keep4o #OpenSource4o 🙂

3h6805