Chip designer and lead brand, sets the architecture and benchmarks, plans to deploy Jalapeno in its own data centers while remaining a customer of Nvidia hardware.
Silicon co-design and implementation partner, also supplies the Tomahawk6 networking switches used in Jalapeno's rack fabric; CEO Hock Tan claims the chip runs inference at roughly half the cost of a typical AI GPU.
Fabricates the chip using advanced process and packaging; its advanced packaging capacity is reported sold out through 2026, a constraint shared with Google and Meta's own custom-silicon programs.
Reportedly supplies the HBM4 memory stacks used in the Jalapeno package.
Incumbent GPU supplier and Jalapeno's primary benchmark rival via GB200/GB300; OpenAI continues to buy Nvidia hardware even as Jalapeno aims to reduce reliance on it for some inference workloads.
Deployment partner - reported to receive Jalapeno deployment at gigawatt-scale data centers, alongside continued Azure infrastructure investment from OpenAI.
Handles board, rack, and system integration for Jalapeno deployments.
Independent analysis firm whose InferenceX benchmark suite was used for the published comparisons; concludes Jalapeno beats Blackwell broadly but notes the fairer long-term rival is Nvidia's upcoming HBM4-based Rubin chip.