DeepSeek hikes API prices as demand outstrips limited GPU capacity

Chinese AI startup DeepSeek announced an upcoming substantial increase in the fees for all its API services. The company said the move is driven by a surge in demand that has strained its limited compute infrastructure, which relies on roughly 20,000 Nvidia H100 GPUs. DeepSeek’s V4 Flash model, with 284 billion parameters, was priced at $0.14 per million input tokens and $0.28 per million output tokens, undercutting rivals such as OpenAI. After the model’s launch, traffic grew beyond the capacity of the hardware, causing slower response times and prompting the price revision.

DeepSeek also disclosed plans to build a large‑scale data centre in Inner Mongolia, targeting about 1 gigawatt of AI processing power to support future models like V4 Pro. No specific percentage or effective date for the new pricing was provided, leaving developers to reassess operational costs. Competitor pricing – OpenAI, Anthropic, Google Gemini – was cited for context, highlighting DeepSeek’s historically aggressive rates.

The announced changes are expected to affect companies and developers that rely on DeepSeek’s APIs, requiring them to recalculate budgets for high‑volume applications.