Can anyone use OpenAI's Jalapeño chip yet?

No. Jalapeño is still at the engineering-sample stage. OpenAI plans small-volume deployment inside its own data centers by the end of 2026, with a larger rollout in 2027.

Is Jalapeño replacing OpenAI's Nvidia chips?

OpenAI says no — the company describes Jalapeño as a complement to its existing hardware relationships with Nvidia, AMD, AWS, Cerebras, and CoreWeave, not a replacement for them.

How is Jalapeño different from a GPU?

Jalapeño is an AI accelerator built only for inference — serving already-trained models — rather than a general-purpose GPU that also handles training and other workloads.

OpenAI's Jalapeño Chip Outpaces Nvidia in Inference Benchmarks

OpenAI has released its first public benchmark results for Jalapeño, the company’s first in-house AI accelerator chip, saying it outperforms comparable Nvidia systems on speed and power efficiency. The results, posted August 25, mark the public debut of a project OpenAI has pursued with chip partner Broadcom since mid-2024.

Jalapeño is built solely for inference — running already-trained models, not training new ones — and was tested using SemiAnalysis’ InferenceX benchmark suite across three open-weight models: GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T. OpenAI says the chip delivered 1.5 to 1.9 times more throughput per watt than the comparison systems, cut end-to-end response latency by 1.7 to 3.6 times, and reached 2.1 to 4.1 times higher performance on interactive workloads. The GPT-OSS test ran against Nvidia’s GB200, while the DeepSeek and Kimi comparisons used Nvidia’s newer GB300. OpenAI also said Jalapeño sustained power draw at or below 550 watts despite being rated for 700.

A qualified lead

Analysts note the comparison has limits. Nvidia’s most direct rival to Jalapeño is its upcoming Vera Rubin platform, which shares the same HBM4 memory — and Rubin systems are already shipping to customers, while Jalapeño remains at the engineering-sample stage. Nvidia and AMD have also published results for larger models, including DeepSeek V4 Pro and Kimi K3, that have not yet been tested on Jalapeño.

OpenAI has framed the chip as a complement to its existing supplier relationships rather than a replacement, naming Nvidia, AMD, AWS, Cerebras, and CoreWeave as continuing partners. The company plans to begin deploying Jalapeño in small volumes within its own data centers by the end of 2026, with a larger rollout expected in 2027; a second-generation chip is already in development and a third is in early planning.

The launch adds OpenAI to a growing list of AI labs and cloud providers building custom silicon to reduce reliance on Nvidia’s GPUs, alongside efforts such as Etched’s dedicated inference chips and Google’s in-house TPUs.

A qualified lead

Frequently asked questions