GPU and CPU Co-Packaging Redefines Edge AI Compute Paradigms

When a laptop weighing less than 2 kg can smoothly run a large model with 120B parameters, do we still need to send every inference request to the cloud? The live debut of the RTX Spark at Bilibili World straightened out that question mark—it is no longer an armchair concept but a "super chip" with CPU and GPU fused together, currently shaking up the underlying logic of AI edge computing.