Scope: From 36Kr Pitchhub 2026 (January–May), among which Frontier Technology (188) + Tool Software (17) industries, totaling 201 events(accounting for 18% of the overall market).
In one sentence: Capital is flowing along three lines
- Large Models (5 leading players) — Moonshot AI stands out with a US$2 billion Series D; Step / Zhipu / Baichuan / 01.AI proceed cautiously
- Embodied Intelligence (Explosive Growth Phase) — 5+ Series B/A rounds exceeding RMB 1 billion: Qianxun / Zibianliang Robotics / Lingxin Qiaoshou / Jijia Shijie / LimX Dynamics
- Quantum / Computing Power / AI for X Application Layer — Bose Quantum / Jiuzhou Xingji / Guanglun Intelligence / Aishi Technology
Large Models: Leaders Locked In + Second-Tier Players Exiting
Moonshot AI = the only company with "two consecutive US$2-billion-level rounds"
| Date | Round | Amount |
|---|---|---|
| 2026-02-19 | Series C+ | Over US$700 million |
| 2026-05-06 | Series D | US$2 billion |
Total US$2.7 billion in a single year. The May 2026 Series D is the largest single funding round for domestic large model companies largest single funding round among domestic large model companies, directly widening the gap with Step and Zhipu.
Moves of other large model companies
| Company | H1 2026 Moves | Notes |
|---|---|---|
| StepFun | 2026-01-26 Series B+ RMB 5 billion | Multimodal + end-to-end voice; partnership with Volcano Engine |
| MiniMax | 2026-01-09 Hong Kong IPO HK$4.818 billion | Listed before OpenAI; the first publicly traded Chinese large model stock |
| Zhipu AI | No large deals disclosed by 36Kr | Pivoting to B-end (government / central state-owned enterprises) + state capital rounds |
| Baichuan Intelligence | No large deals disclosed by 36Kr | Pivoting to medical large models |
| 01.AI | No large deals disclosed by 36Kr | Reported to have pivoted |
Why "5 leading players locked in + second-tier players exiting": The threshold of training GPU cluster + data costs exceeds US$1 billion per year, and second-tier companies can no longer keep up with scaling. Moonshot AI is supported by Kimi's C-end daily active users and Alibaba's strategic investment; MiniMax went public relying on Talkie's overseas revenue.
Special significance of MiniMax's IPO
Hong Kong IPO on 2026-01-09 raising HK$4.818 billion (≈ US$600 million). This is the first IPO of a large model company globally in 2026 (OpenAI's US listing is expected in Q3–Q4). Implications:
- Chapter 18C of the Hong Kong Stock Exchange accepts unprofitable large model companies
- The overseas IPO path for Chinese AI companies is opened
- Valuation anchor: HK$4.818 billion market cap, far below the US$10 billion primary and secondary market valuations set by US dollar funds
Embodied Intelligence: The Real Protagonist of H1 2026
Domestic embodied intelligence / humanoid robot companies collectively raised large sums in H1 2026.5 companies raised over RMB 1.5 billion in a single round:
| Company | Date | Round | Amount | Notes |
|---|---|---|---|---|
| Zibianliang Robotics | 2026-04-20 | Series B | Nearly RMB 2 billion | Dual-arm manipulation large model |
| Qianxun Intelligence | 2026-02-24 | Series A | RMB 2 billion | Extremely rare for a Series A round to raise RMB 2 billion |
| Qianxun Intelligence | 2026-04-07 | Series A+ | RMB 1 billion | Same company raised again within 2 months |
| Jijia Shijie | 2026-04-16 | Series B | Nearly RMB 1.5 billion | Dexterous hands |
| Jijia Shijie | 2026-03-05 | Pre-Series B | Nearly RMB 1 billion | Same company, 1 month earlier |
| Lingxin Qiaoshou | 2026-02-12 | Series B | RMB 1.5 billion | Dexterous hands |
| LimX Dynamics | 2026-02-02 | Series B | US$200 million | Whole-body control |
| Guanglun Intelligence | 2026-03-11 | Series A+ | RMB 1 billion | Synthetic data |
| Volant Aerotech | 2026-04-27 | Series C | US$300 million | eVTOL low-altitude |
Why embodied intelligence is exploding:
- Outside of large models, it is a track where "leaders have not been locked in"; VCs have FOMO from "missing out on large models"
- National policy set the tone ("embodied intelligence" was written into the 2025 Government Work Report)
- Hardware supply chain / GPU training costs are controllable (single prototype at the RMB 1 million level vs large model clusters at the US$1 billion level)
- Exit expectations: High IPO probability within 5 years (benchmarked against Figure / Optimus / 1X)
Why Qianxun Intelligence closed Series A + A+ back-to-back in 2 months: Consecutive equity financing is a hallmark of hot deals; raising RMB 2 billion in Series A followed by RMB 1 billion in Series A+ equals total fundraising of RMB 3 billion (about US$400 million) in H1 2026.
Quantum Computing Power + AI for X Application Layer
| Company | Date | Round | Amount | Notes |
|---|---|---|---|---|
| Bose Quantum | 2026-03-31 | Series B | RMB 1 billion | Quantum coherent optical computing |
| Jiuzhou Xingji | 2026-04-03 | Strategic financing | RMB 1 billion | AI video generation |
| Aishi Technology | 2026-03-12 | Series C | US$300 million | PixVerse, large overseas user base for AI video |
| iFlytek | 2026-04-22 | Directed share issuance | Approximately RMB 4 billion | A-share listed company replenishing capital |
Why the application layer accounts for a low share:
- 36Kr focuses on the company level rather than the tool level; a large number of AI application startups (with funding < RMB 100 million) are not listed on Pitchhub
- Early-stage application layer financing is mostly from individuals / angels, and often "undisclosed"
Profile differences between state capital and US dollar funds
State capital preferences (embodied intelligence + manufacturing)
| State capital institution | H1 2026 moves in the AI track |
|---|---|
| Chengdu Science and Technology Venture Capital Group | Qianxun Intelligence / Bose Quantum |
| Suzhou Yuanhe | Zibianliang Robotics / Jijia Shijie |
| Shenzhen Venture Capital Group | Step / Moonshot AI |
| CAS Star | Bose Quantum / Guanglun Intelligence (CAS-affiliated hard technology preference) |
| China Structural Reform Fund | Moonshot AI Series D |
US dollar fund preferences (anchoring leading players)
| US dollar fund | H1 2026 moves |
|---|---|
| Sequoia China | Moonshot AI Series D / Qianxun |
| Hillhouse | Moonshot AI / Step |
| Qiming Venture Partners | Lingxin Qiaoshou / Jijia Shijie |
| Shunwei Capital | Zibianliang Robotics / Jijia Shijie |
| 5Y Capital | Qianxun / Guanglun |
Why state capital and US dollar funds appear together: Large AI deals in H1 2026 generally adopt a syndicate structure of "state capital leading + US dollar funds following", seeking both political correctness (state capital participation) and professional judgment (US dollar funds evaluating the track).
Boundary of companies tagged "Frontier Technology" by 36Kr
36Kr's industry tags are not strict; some AI companies are classified under other industries:
- MiniMax(IoT/Hardware): Because it has physical Talkie products
- StepFun(Culture & Entertainment): Because it built AI agent applications
- Kujiale / Coohom(Enterprise Services + Frontier Technology): 3D design SaaS + AI generation
- iFlytek(Frontier Technology): Listed company focused on ASR/TTS, actually closer to "mature technology"
Estimated under "broad AI" (including those classified elsewhere), real AI funding events in H1 2026 are approximately 220–240 events, accounting for 20% of the total.
Three judgments
1. AI is not a "track", but an "application tool"
In H1 2026, there were only ~30 real "large funding rounds for AI companies" (Moonshot AI / Step / embodied intelligence + 7–8 quantum companies). The remaining 170+ "frontier technology" events are AI applications / tools / data infrastructure.AI is not an independent track, but a capability embedded in every industry.
2. Large models: winner-take-most
Large model companies in H1 2026:
- 1 company pulling ahead (Moonshot AI)
- 2–3 companies sustaining (Step / MiniMax / Zhipu)
- 5+ companies exiting / pivoting
This landscape should remain stable over the next 12 months; second-tier companies will no longer have opportunities to "train large models".
3. Embodied intelligence = capital from the previous generation of AI application layer reallocated
After missing out on large models, VCs are placing their "AI-era bets" anew on embodied intelligence.Single-round amounts are already an order of magnitude larger than those in the AI application layer (Series B RMB 1.5–2 billion vs AI application Series A RMB 100–300 million).
Risks: The commercialization path for embodied intelligence is unclear (C-end is expensive, B-end is slow); whether exits can be achieved within 5 years remains to be seen.
Data queries
See sources/reference/36kr-pitchhub-2026/ (Local reference materials):
- Moonshot AI Series D:
events/<projectId>/summary.md(Local reference materials) + articles/ (associated 36Kr in-depth reports) - Qianxun Intelligence / Zibianliang Robotics / Jijia Shijie same as above
- Full filtered industries:
grep -l '前沿技术' events/*/summary.md
Cross-wiki references
- Crawling method:36Kr Pitchhub crawling methodology
- Full industry overview:Panoramic Snapshot of China Venture Capital H1 2026
- Wang Kai's perspective on AI industry judgments:Wang Kai's Observations on AI Industry Judgments and Big Tech Competition Landscape
- newtype's perspective on the Agent market:AI Agent Market Investment Trends 2025 (Data Report)
As of 2026-05-19, data after May for the large model / embodied intelligence tracks are pending supplementation.