How Liang Wenfeng’s DeepSeek Propelled Chinese AI Unicorns Past the Trillion‑Yuan Mark

01

May 2024 saw an unprecedented surge in primary‑market financing for Chinese AI unicorns. DeepSeek, the domestic super‑AI unicorn, announced its first external fundraising, with the National Integrated Circuit Industry Investment Fund and Tencent among potential lead investors. Early reports suggested a valuation above $10 billion; within weeks the valuation rose to $45‑50 billion, and the round is expected to reach $3‑4 billion, mainly for compute infrastructure, R&D, and employee incentives.

DeepSeek’s valuation multiplied five‑fold in a few weeks, driven by its status as a core competitor in the domestic AI base‑model arena and by market expectations that the company would finally open its financing door after years of “technical idealism.”

At the same time, Moon Shadow (月之暗面) secured a roughly $2‑4 billion round, pushing its post‑money valuation past $20 billion. The round was led by Meituan Longzhu, with participation from China Mobile, CPE and other investors, marking the largest single private financing for a domestic large‑model startup to date.

Jueci Xingchen (阶跃星辰) was reported to be closing a near $25 billion financing, attracting capital from mobile‑ODM leaders Huawei‑Qin, Longqi, Haowei and telecom giant ZTE, indicating deep industry‑chain involvement.

02

Two weeks ago DeepSeek‑V4 Preview was launched and open‑sourced. V4‑Pro and V4‑Flash both support a 1 million token context length; V4‑Pro’s parameter count reaches an astonishing 1.6 trillion. Pricing is $0.435 per million input tokens and $0.87 per million output tokens, roughly one‑tenth the cost reported for Anthropic’s Claude Code (≈$13 per day for developers). The model’s cost‑performance attracted strong community feedback, with Hacker News users noting that V4 Pro runs code‑review workloads in about twice the time of Opus/GPT but at less than 10 % of the cost.

V4’s significance extends beyond model upgrades: it is tightly integrated with Huawei’s Ascend AI chips. Huawei announced full support for DeepSeek V4 on its Ascend 950 series, and other AI giants such as ByteDance, Tencent and Alibaba are racing to acquire Ascend chips following the V4 release. This alignment of a domestic base‑model with domestic compute hardware marks a pivotal convergence of the AI stack.

03

By the end of May, DeepSeek’s valuation was estimated at ¥3 200‑3 600 billion (≈$45‑50 billion), Moon Shadow at ¥1 400 billion (≈$20 billion) and Jueci Xingchen targeting a $10 billion valuation as its $25 billion financing closes. The new market norm for top Chinese base‑model companies is now a valuation threshold of roughly ¥1 trillion.

The re‑valuation is driven by several factors: the binding of DeepSeek V4 to Huawei Ascend chips, a rapid rise in domestic compute‑chip indices (cloud services +5.77 %, semiconductor +5.48 % on the first trading day of May), and analyst forecasts that AI‑infra revenue will multiply in 2025‑2026.

Despite the financing boom, uncertainties remain. Secondary‑market gains for Zhipu and MiniMax have already been tempered by price corrections after V4’s launch, and high valuations bring volatility. Red‑chip (VIE) structures required for Hong‑Kong IPOs are being dismantled by some firms (e.g., Jueci Xingchen) while others (Moon Shadow) are still negotiating restructuring, adding another layer of risk.

Overall, the “crazy May” financing wave has reshaped the valuation landscape for Chinese AI base‑model companies, linked tightly to domestic compute capabilities and early commercial use cases such as Vibe Coding and Agent tools, while also highlighting the limited number of players capable of sustaining the massive R&D spend required for next‑generation models.