Physical AI is shifting from lab prototypes to fleets of machines operating reliably in the real world, and Ambarella's Muneyb Minhazuddin joins Six Five On The Road at AI Infra 2026 to explain what that shift demands of silicon, memory and developer workflows. Minhazuddin traces the case back to a 22-year-old chip architecture and forward to a developer cloud built to compress model porting time from months to days, and he identifies enterprise ROI proof points as the biggest hurdle left before physical AI scales into millions of deployed units.