The Shenzhen humanoid robotics startup says it has sold thousands of units, mostly overseas. The question is whether that 'order book' survives a public market audit in a category with no profits.
A Shenzhen-based humanoid robotics startup, LimX Dynamics, has closed a roughly $200M Pre-IPO financing at a ~$2.1B post-money valuation. The round formalizes a bet that an order book the company describes as "thousands of units, mostly overseas" can survive the public-market audit an IPO in mainland China will eventually require.
The round was first reported by 36kr / 硬氪, which gave the size as "nearly $200M" and the post-money as 15 billion yuan. The investor list is a cross-section of China's industrial and capital establishment: IDG Capital, Lens Technology, GGG Group, Redstone VC, Huashan Capital, and Hefei Binhu Industrial Development Group, joined by Stone Venture, an Abu Dhabi-based firm that has participated in multiple prior rounds. Existing shareholders Oasis Capital, Cornerstone Capital, Nanshan Strategic Investment, Shangqi Capital, and NIO Capital reportedly over-subscribed. (36kr, thepaper.cn)
Cumulative funding across the last six months is now reported at $400M. Earlier rounds drew in a SAIC-affiliated private equity fund (jiemian), Alibaba and China Merchants (headscm), and, per a separate report, Liu Qiangdong (rccaijing). The Pre-IPO is the last private round before a public listing.
The product line carrying the round has two parts. LimX Luna, a 160-centimeter, 27-degrees-of-freedom humanoid, launched in May 2026 and reportedly reached batch deliveries within a month. TRON 2 is a multi-form robot that switches between dual-arm, bipedal, and dual-wheel-leg configurations, aimed at industrial and research buyers who want one hardware platform in multiple form factors. The company says it has sold "thousands of units" since launch, with more than half going to overseas customers. It has not disclosed unit prices, named the buyers, or split firm orders from softer pipeline. (36kr)
The autonomy stack behind the products is where the technical claims get load-bearing. LimX describes a three-layer architecture: System 0, a full-body motion foundation model; System 1, a vision-language-action layer for humanoid capabilities; and System 2, an embodied-agent operating system called COSA, built on large language models and world models. The company also runs FluxVLA, an open-source platform it positions as an enterprise-grade training pipeline spanning data processing, simulation, real-machine iteration, and hardware deployment. Zhang Wei, who founded LimX Dynamics in 2022, frames the design as a "大小脑融合" bet, with high-level reasoning from foundation models wired into low-level whole-body control. (rccaijing, 36kr)
These are company claims, not third-party validation. None of the System 0/1/2 architecture, the COSA OS, or the FluxVLA pipeline has been audited or benchmarked externally in the public record.
LimX completed a joint-stock reform in March 2026, the formal step Chinese companies take ahead of a domestic A-share listing. Per informed sources cited by 36kr, the company has now started the IPO process, though no prospectus has been filed. The capital raised in this round is earmarked for three things: pushing the "big brain + small brain" architecture into product; scaling the deployment of "thousands of fully autonomous humanoid robots"; and a global market push focused on the Middle East, Europe, and other Asian markets. (36kr)
China's humanoid robotics field is crowded. Unitree, Fourier Intelligence, UBTech, Agibot, and Galbot have all raised at the hundred-million-dollar scale or above in the last 18 months. None has filed audited financials in a public-market setting. The $400M cumulative that LimX points to is roughly the round volume a category leader can now command, not a unique signal. A roughly $2.1B post-money on a company that has disclosed no unit prices, no named customers, and no revenue is a private-market bet on category leadership rather than a fundamental pricing event.
The IPO, when it comes, will be the first real test. Public-market disclosure in China requires segment revenue, gross margin, and order-backlog detail that LimX has not volunteered in private. If the order book is mostly firm, the listing can clear at or near this private mark. If it softens between Pre-IPO and prospectus, the proxy breaks and the next private round re-prices.
The Pre-IPO is closed. The next data point on the company's path to listing is a prospectus.