Moonshot AI has reportedly filed an A1 document confidentially with the Hong Kong exchange and begun its IPO process, raising a Pre-IPO round at a US$50 billion pre-money valuation. The company declined to comment.
It is not alone. StepFun is said to plan a Hong Kong listing by year end, and DeepSeek is reported to be preparing a domestic IPO, possibly filing this year for a 2027 debut. As they queue for the public markets, all three have fundraised aggressively through 2026, with capital racing for the last seats on the large-model ship.
Alongside the IPO report, Moonshot is running a Pre-IPO round through a designated direct-investment fund at a US$50 billion pre-money valuation, with a reported RMB 30 million minimum ticket. Since founder Yang Zhilin’s US$500 million raise announcement late last year, Moonshot has run five rounds and disclosed about US$7.9 billion (roughly RMB 53 billion) in total, backed by Alibaba, Tencent, Meituan, IDG and 5Y Capital.
Its valuation climbed from US$4.3 billion to US$50 billion in just eight months, more than tenfold. In July it completed a US$3.5 billion-plus F round and converted its domestic entity into a joint-stock company, with leadership and shareholder changes. Yang holds 51.83 per cent, and he, with co-founders Zhou Xinyu, Wu Yuxin and Zhang Yutao, hold 65.63 per cent. Nine state-backed domestic institutions joined as shareholders in late June.
StepFun, after AI veteran Yin Qi took control, raised RMB 5 billion early in the year, nearly US$2.5 billion in May, then about RMB 600 million in strategic investment from three listed firms. DeepSeek, after a first round of RMB 51 billion led by Liang Wenfeng with Tencent, CATL, NetEase, JD, IDG and the state AI industry fund, has opened a second round targeting another RMB 50 billion at a RMB 500 billion pre-money valuation, making it China’s most valuable independent model company.
The window is narrowing. As one AI investor put it, everyone wants more bullets before the window shuts. The real test after the capital frenzy is whether models that top global coding leaderboards can turn that rank into recurring enterprise revenue. Moonshot’s K2.5, K2.6 and K3 models placed it third globally and first in frontend coding at one point, but commercial conversion remains the open question.
Editor’s note: This is an adapted translation of the original Sohu IT report. It has been trimmed and restructured for readability for an international business audience.
Translated and adapted from Sohu IT (https://www.sohu.com/a/1071912163_115565).