DeepSeek plans $7.4B raise at $74B valuation, building own inference chip ahead of STAR IPO

China's DeepSeek is moving fast on both capital and silicon. Reuters reports the company is targeting a fresh raise near 50 billion yuan (~$7.4 billion) at a roughly $74 billion post-money valuation, and is preparing IPO documentation for a Shanghai STAR Market filing as soon as the end of 2026, eyeing a 2027 debut. This comes just weeks after DeepSeek closed its first-ever external financing — about $7 billion at a $52 billion valuation near the end of May — a striking valuation jump in under two months.

The strategic tell is hardware: DeepSeek is developing its own inference-focused AI chip to cut dependence on NVIDIA and Huawei parts, a bet aimed at controlling cost-per-token as its models scale. That vertical-integration move mirrors what OpenAI, Google (TPU) and Amazon (Trainium) have done, and reads as a hedge against U.S. export controls that constrain access to top NVIDIA accelerators.

DeepSeek's traction underpins the valuation: in June it accounted for nearly 23% of tokens processed through enterprise gateway Vercel, evidence its open-weight models are winning real production workloads, not just benchmark buzz. Community sentiment frames the chip-and-IPO combo as a geopolitical bet against NVIDIA lock-in and a sign of Chinese AI maturation — though skeptics note DeepSeek's implied market cap is still roughly a thirteenth of OpenAI or Anthropic.

What to watch: whether the STAR Market listing actually lands on schedule given regulatory complexity for AI firms, and whether DeepSeek's in-house inference chip reaches tape-out on a timeline that matters. The raise also signals China's willingness to fund a homegrown frontier challenger through public markets rather than Western venture capital.