SemiAnalysis has published a detailed look at Jalapeño, an application-specific integrated circuit that it describes as OpenAI’s self-designed chip developed with Broadcom, according to a Techmeme summary of the report. The central claim is that Jalapeño was developed in 16 months and beat chips from Nvidia, AMD and Google on multiple top open-source models. The public summary does not specify the benchmark methodology, the exact models tested, the workloads involved, or the comparison configurations, so those details should remain attributed to SemiAnalysis rather than treated as independently confirmed. Techmeme’s summary also says the SemiAnalysis report compares OpenAI’s ASIC with Nvidia Rubin and discusses Jalapeño’s total cost of ownership and throughput. Those are the right axes for evaluating custom AI silicon, but the available cluster does not include the figures behind them. For now, the story is best read as a developing chip-industry report: a detailed single-source analysis claims OpenAI and Broadcom produced a competitive custom ASIC on an unusually compressed timeline. The material claims are not contradicted in the cluster, but they are not corroborated by another independent item here. Who benefits: OpenAI and Broadcom would benefit if the reported performance and development timeline hold up. A credible custom ASIC could strengthen OpenAI’s leverage in AI infrastructure planning. Who's exposed: Nvidia, AMD and Google are the named comparison points in the SemiAnalysis report. The degree of exposure cannot be assessed from this cluster because the public summary does not provide benchmark scope or deployment scale.