DeepSeek in advanced talks for Series B at $71B valuation; AI agents drive data center capex

Chinese AI firm DeepSeek is in preliminary discussions for a Series B funding round at a targeted $71 billion valuation, just weeks after closing its first-ever Series A in May 2026. The company raised $7 billion at a $52 billion valuation in May; the new round would represent a 37% valuation increase. Multiple sources told the Financial Times on July 14 that deal details remain preliminary, but the unusually rapid funding schedule reflects DeepSeek's capital demands as it pivots from model development to infrastructure ownership.

DeepSeek is planning to build its own data center and acquire additional AI chips to support its expanding agent-based systems. The company's shift toward agentic AI is driving significantly greater compute demand than its earlier model-only offerings. By moving to in-house infrastructure—rather than relying entirely on rented cloud capacity—DeepSeek mirrors the playbook of US frontier labs (OpenAI, Anthropic, Google, Meta), which are all securing long-term chip supplies and custom silicon.

The pace of fundraising reflects the capital intensity of agentic AI infrastructure. As autonomous systems proliferate, the compute needed per model increases dramatically, pushing frontier labs to secure supply agreements and in-house capacity before growth accelerates. For practitioners, the signal is clear: the transition from inference-optimization to training-plus-deployment infrastructure is expensive and increasingly capital-gated.

DeepSeek's rapid fundraising also underscores continued investor confidence in Chinese AI despite US export controls on advanced chips. The company famously trained its flagship models on domestic Chinese silicon (not Nvidia H100/H200), signaling both technological progress in non-US chip stacks and continued access to capital despite geopolitical friction.