This is a synthesis of a major article published by Zhidongxi via 36Kr on May 14, 2026. The original was written in Chinese. Crossing the River exists to surface the best Chinese-language writing on entrepreneurship in China for a Western audience.

In a single week in May 2026, China’s foundation model startups collectively went into valuation hyperdrive. The article asks a sharp question: of these billion-dollar-plus AI companies, whose valuation actually has substance behind it?

May 7: Moonshot AI (Kimi’s parent) closed US$2B at a US$20B valuation. May 8: StepFun reportedly raising US$2.5B, restructuring for a Hong Kong IPO. The same day, DeepSeek’s first outside round reportedly hit US$7.3B, valuing the company at US$51.5B. China’s largest-ever AI fundraise was broken twice in one day. May 13: Zhipu’s Hong Kong-listed stock surged 37%, closing above US$65.7B market cap for the first time. MiniMax jumped 18%, reaching US$32.9B.

The “Six Little Tigers” of China’s foundation model wave — Zhipu, MiniMax, Baichuan, Moonshot, StepFun, and 01.AI — now have three members valued above US$14.7B. The article applies three tests to evaluate whether these valuations hold up

Three capital types are fueling the boom. State capital provides ballast: China’s National IC Investment Fund (“Big Fund”) may lead DeepSeek’s round, its first-ever investment in a foundation model company. This would elevate LLMs to a strategic priority on par with semiconductor manufacturing. DeepSeek founder Liang Wenfeng reportedly plans to personally invest up to US$2.9B (40% of the round), keeping control with the founding team.

Industrial capital provides synergy: StepFun’s US$2.5B round drew investment from handset supply chain players (Huaqin, Longcheer, OmniVision, ZTE). Tencent has invested across three consecutive rounds. Moonshot attracted China Mobile, potentially giving Kimi distribution through the carrier’s massive consumer base.

VCs and internet strategics chase growth and exit: MiniMax is backed by Hillhouse, IDG, Sequoia, plus Alibaba, Tencent, and Xiaohongshu. Companies that attract all three capital types have the most durable valuation foundations.

MiniMax has the clearest data via its IPO prospectus. Revenue grew from US$0.25M (2023) to US$30.5M (2024) to US$53.4M (first 9 months of 2025). Gross margin improved from -24.7% to 23.3%. C-side AI products (Talkie, Hailuo AI) drove 71% of revenue, with 70%+ from overseas. But net loss was US$512M in the same period.

Zhipu is the government-enterprise story. Revenue: US$8.4M (2022) to US$45.9M (2024). But losses hit US$347M with R&D at 835% of revenue. The striking comparison: Zhipu has less than US$118M annual revenue yet trades at US$65.7B. Meanwhile iFlytek earns US$3.99B annually with US$123M net profit, yet is valued at only US$15.2B.

Kimi’s ARR reportedly broke US$100M in March 2026, then doubled to US$200M+ by April. At US$200M ARR against a US$20B valuation, that’s roughly 100x forward P/S. StepFun has shipped models into 42M+ devices but install base does not equal revenue. DeepSeek has never disclosed revenue, but its pricing (V4-Flash at US$0.003 per million input tokens) is setting the cost benchmark for the entire industry. It also completed full migration from Nvidia CUDA to Huawei’s Ascend CANN — the first frontier AI model independent of Nvidia.

Baichuan last raised US$735M at US$2.9B valuation with US$441M cash remaining. 01.AI reportedly exceeded US$14.7M revenue in 2024. Both lack audited public financials.

DeepSeek’s position is the most unique: open-source infrastructure, Nvidia independence, and lowest pricing. Hardest to replicate, but open-source doesn’t automatically equal a business moat. Zhipu is “China’s Palantir” with 12,000 enterprise clients and 80M endpoints, but depends on government budgets. Kimi is betting on being China’s ChatGPT/Claude. MiniMax has the strongest global product matrix. StepFun is embedded in hardware. Baichuan has pivoted to healthcare AI; 01.AI to enterprise platforms.

China’s AI valuation story has two halves: the first was about who had a model; the second is about who has a business. No company has fully proven commercial viability yet. To hold a US$73.5B+ valuation, a company must pass all three tests simultaneously: capital that’s patient enough, revenue that’s real enough, and a market position that’s hard enough to replace.

This is an original English synthesis, not a verbatim translation. Read the full original (in Chinese) at 36kr.com/p/3807808383344384.