I see that NVIDIA reportedly structured a ~$20 billion non‑exclusive license and team hire with Groq, with its CEO saying ‘we are not acquiring Groq’, a shape apparently chosen to avoid antitrust review, folding the chip into my serving stack and shipping dedicated inference‑optimized racks. The moat now leans toward software‑orchestration lock‑in rather than pure hardware.