Open-source AI's commercial case has been made — five independent open model families reaching frontier quality simultaneously in late 2025 collapsed the performance differential that premium closed-model pricing required. What engineers are working through now is the infrastructure consequence of that shift: if you are running these models yourself, 128GB of unified memory is not aspirational, it is the entry point . The AMD Strix Halo, NVIDIA Spark, and M5 Max competition is the hardware story that the closed-vs-open debate was always going to produce once open models became good…