DeepSeek Raises Record ¥50 B in First Round, Backed by Liang Wenfeng’s ¥20 B Commitment, V4.1 Set for June

DeepSeek’s valuation jumped five‑fold in 21 days, reaching ¥350 billion, and the company announced a record‑size first‑round financing that could total ¥500 billion. According to The Information, entrepreneur Liang Wenfeng personally contributed up to ¥200 billion, accounting for 40% of the round, making it potentially the largest funding event for a Chinese large‑model firm.

The valuation milestones unfolded rapidly: early April 2026 the round began at roughly ¥100 billion; by April 22 discussions with Tencent, Alibaba and others pushed the figure past ¥200 billion; on May 6 the National Integrated Circuit Industry Investment Fund was reported to lead a ¥450 billion negotiation; and early May reports suggested a final valuation could hit ¥500 billion.

Historically, DeepSeek branded itself as a research‑first AI lab that avoided fundraising, commercialization, and roadshows. The new financing signals a shift, driven by three core pressures.

01 Compute

Frontier models now require far more than simple training—advanced inference, agent capabilities, ultra‑long context (1 M tokens), and enterprise‑grade stability push compute needs upward. DeepSeek’s April V4 series already extended context to 1 M tokens and began testing visual mode, offering developer‑friendly features that nonetheless demand massive training and continuous inference resources, especially if the company moves toward enterprise services.

02 Talent

The firm has lost several star researchers—including Guo Dayi, Wang Bingxuan, and Wei Haoran—who have taken higher‑pay positions elsewhere. As competition for top AI talent intensifies, idealism alone cannot retain staff; the financing provides a basis for pricing employee stock options and aligning incentives.

03 Productization

DeepSeek now stresses that a model’s strength must translate into chargeable products and services. Employees are already promoting the models to industry customers, and the company acknowledges that a heavy‑asset AI firm must manage customers, revenue, delivery, cost, and talent structure, not just model metrics. The new capital will help fund these product‑focused initiatives, though the next question is where the money will be allocated.

The V4.1 upgrade, expected in June, will add more tools, better support for the industry‑standard MCP protocol, and capabilities for simultaneous image and audio processing. According to The Information, the financing pressure is accelerating DeepSeek’s release cadence, moving the lab from a leisurely, research‑centric timeline to one that aligns with typical industry speeds.

Overall, DeepSeek is transitioning from a light‑asset research team to a heavy‑asset AI company, where compute, data centers, product teams, enterprise customers, equity incentives, and release rhythm intersect. This mirrors a broader industry shift: large‑model competition is evolving from a purely lightweight model race to a stage where compute, talent, capital, and commercialization are all on the table.