- Published on
OpenAI’s In-House Jalapeño AI ASIC Enters Mass Production and Deployment, Chosen as a “Pragmatic” Option to Pair With...
- Authors
- Name
- DeepHorus
IT Home reported on October 3 that Tom's Hardware disclosed on October 2 that OpenAI has already deployed its in-house-developed Jalapeño ASIC chip internally, using it together with AMD EPYC Turin hosts equipped with 1.5 TB of memory. This combination is currently intended only for OpenAI's internal requirements and has not yet been sold externally. Richard He, OpenAI's vice president and head of hardware, disclosed the platform-selection rationale and the principles guiding the project's execution.
During the design phase of the Jalapeño project, the OpenAI team identified "de-risking" and "rapid implementation" as two core principles from the outset. The team aimed to be aggressive in both performance targets and cost control, while ensuring that platform selection followed the existing engineering decision-making framework. He said this approach was intended to keep the project's execution under control and reduce uncertainty during deployment.
He explained in an interview that choosing Turin was primarily based on "pragmatic" considerations. He praised AMD Turin as extremely powerful, saying it had already met all of the Jalapeño project's requirements. By contrast, NVIDIA's new Vera CPU was "somewhat behind" in terms of maturity.
OpenAI also valued industry experience and deployment efficiency. He said that partners already had extensive experience with the Turin platform, which would help accelerate the actual deployment of the Jalapeño platform. This CPU choice was not based solely on performance; it also incorporated chip maturity and engineering feasibility into the evaluation.
Based on the information disclosed so far, Jalapeño has entered actual internal deployment at OpenAI and is operating in conjunction with high-memory-configured Turin hosts. However, OpenAI has not yet announced chip production volumes, specific performance, cost targets, or use cases. It remains unclear to outside observers whether the project has expanded to large-scale deployment.
Jalapeño is currently for OpenAI's internal use only, and the company has no plans to sell it externally. OpenAI has also not disclosed a timetable for further expanding deployment. He said that, if resources permit later, the company does not rule out extending Jalapeño to a broader range of use cases.
At this stage, OpenAI will continue to advance work centered on internal deployment and its established project objectives. Issues that will attract attention subsequently include chip production capacity, platform scale, and whether external supply will be established, but the relevant details have not yet been announced. Whether Jalapeño can move from a custom chip for internal use into broader use cases will still depend on the project's progress and OpenAI's resource allocation.
