Inference.net touts 'a ton' of idle B200/B300 capacity at discount rates

MatthewBerman · x · 2026-11-20

Sam Hogan of Inference.net says the company has "a ton" of B200/B300 GPU capacity available at discounted managed-inference rates with flexible short-term commitments, ideal for hosting post-trained Kimi, DeepSeek, and GLM models. Commentator Matthew Berman notes he's never seen a provider publicly claim massive spare capacity — a telling signal on GPU supply-demand.

Original post →

More from Infra

Infra channel →