Custom and specialized AI silicon challenge Nvidia’s inference dominance

Reported OpenAI Jalapeño benchmarks, Intel’s improving MLPerf and Arc Pro results, Apple’s possible Nvidia-networked server initiative, Positron’s reported $875M inference-chip financing, and Marvell’s custom-silicon and networking opportunity broaden evidence that specialized architectures may pressure Nvidia in inference. Marvell’s upside case remains highly assumption-dependent, while independent benchmarks, software validation, production scale, and customer disclosure remain insufficient to establish a durable competitive shift; Intel’s sharp rally raises valuation risk.