The discount makes B200-based SFT and RL runs cheaper for teams that want full control of their training loop rather than a managed pipeline. Startups fine-tuning open-source models should compare this against renting B200 capacity on clouds like AWS or CoreWeave before the eight-week window ends. Pricing is in US dollars and available to customers globally, including Southeast Asia.
Serving teams running MoE models can treat 4:8 sparsity as a practical optimization, since the whole-model 1.18x gain shows the expert-layer speedups survive end-to-end in vLLM. Engineers benchmarking inference should compare these numbers against their own NVFP4 baselines on B200 hardware. Kimi-K2.5 serving as the test model also gives Moonshot (Kimi) users a direct performance reference.
一组挺反直觉的数据：OpenAI 降价 62%，B200 租金反而涨 50%，想理解推理成本逻辑的可以看看。
The B200's resale value surge highlights the ongoing high demand for AI computing power, which is crucial for anyone investing in or using AI hardware.
vLLM拆开预填充和解码，TileRT提供低延迟解码，单用户618 tok/s，适合实时助手。
