DeepSeek released DSpark last week, a Chinese AI lab that keeps finding ways to make inference faster using software and open weights at no cost. Nvidia is ramping a specialized decode rack called the Groq 3 LPX, which requires a separate purchase decision on top of the GPU platform. Nvidia posted the biggest quarter in semiconductor history, with revenue of $81.6 billion, Data Center revenue of $75.2 billion, and GAAP gross margin of 74.9 percent. The new ambition is to sell the Groq 3 LPX, built around 256 Groq LPU accelerators, each carrying 500MB of on-chip SRAM running at 150 terabytes per second of bandwidth. Paired with the Vera Rubin NVL72 GPU system, Nvidia claims the combination delivers up to 35 times higher inference throughput per megawatt for trillion-parameter models. Vera Rubin is now in full production, and LPX is shipping to early customers in the second half of 2026. The risk is that it requires a separate purchase decision from customers who have already committed to Rubin GPUs. DSpark arrived and made that decision harder.