AI Market Watch Asia · Fortnight 18: Moonshot AI files confidential A1 form for Hong Kong IPO targeting $50B valuation

Moonshot AI files confidential A1 form for Hong Kong IPO targeting $50B valuation

STRATEGIC SHIFT: converts Moonshot's previously disclosed $50B pre-IPO fundraising ambition into a formal Hong Kong Stock Exchange listing process, the furthest any China-based frontier AI lab has gotten toward a public listing.

Moonshot AI confidentially filed an A1 form with the Hong Kong Stock Exchange this fortnight, targeting a $50 billion valuation through a pre-IPO round — the first time any China-based frontier AI lab has converted a stated fundraising ambition into a formal public listing process. The filing, reported on September 3, formalizes the company's late-July disclosure that it was seeking a $50 billion pre-money valuation following a $3.5 billion raise at a $35 billion valuation, making it the clearest regulatory milestone in the Chinese foundation-model capital cycle to date. This ranks above other developments because it engages underwriters and disclosure counsel, giving outside capital a defined liquidity path rather than a private-round target.

The A1 submission lands weeks after Moonshot leadership was reported citing a $160 billion valuation tied to its flat organizational structure and Kimi K3 momentum, a gap between the private-market figure circulating in press coverage and the $50 billion number now attached to the public listing.

The filing follows a compressed run of capital milestones: a $3.5 billion Series F at a $35 billion valuation closed in early August, with the National AI Industry Fund as a lead investor.

For investors, the A1 filing signals Moonshot is far enough into Hong Kong listing preparation to have engaged underwriters and disclosure counsel, giving outside capital a nearer-term liquidity path than a typical late-stage private round. The startup is also in talks with Microsoft, Amazon, and Google on revenue-sharing arrangements, per TechCrunch, a distribution context that would matter for how the company's revenue mix evolves ahead of a public debut.

Separately, the filing lands amid a broader Hong Kong listing wave for Chinese AI companies: Zhipu AI trades at HK$1,033, down 65.3% from its all-time high, while MiniMax is down 84% from its peak.

XPeng Robotics raises $900M+ at $6.3B valuation for Iron humanoid mass production

RECORD/MILESTONE: XPeng states this is the largest single-round private financing in China's embodied AI industry, backed by IDG Capital, Tencent, and Alibaba, funding Iron humanoid mass production targeted for end-2026.

XPeng Robotics raised more than $900 million in its first external funding round, valuing the unit above $6.3 billion post-money, in what XPeng called the largest single-round private financing in China's embodied AI industry (source). The round was led by IDG Capital, with Gaorong Ventures participating and Tencent and Alibaba providing strategic backing. This is the top story of the fortnight because the dollar amount and valuation reset the capital bar for OEM-backed humanoid programs: roughly $600 million of the committed capital is external, with about $200 million from an XPeng subsidiary and $100 million from leadership.

XPeng stated the proceeds will fund physical-AI model training, hardware development, and manufacturing capacity ahead of mass production of the Iron humanoid robot, targeted for end-2026 with commercial deliveries in China and overseas planned for 2027 (source). The company set an interim monthly capacity goal above 1,000 units.

The raise was executed through XPeng's Dogotix subsidiary; XPeng retains controlling ownership and will keep the unit consolidated (source).

XPeng is selling a manufacturing thesis: reuse EV supply chains, factories, and owned retail campuses to move embodied AI into volume production (source).

For builders and investors, the implication is that automaker-backed humanoids now compete on factory ramp, fleet and store data loops, and in-house silicon—not gait videos alone (source). Watch whether Iron hits the end-2026 capacity target and whether claimed on-device Physical AI autonomy survives outside campus pilots. Pure-play robotics startups without automotive-grade production muscle face a steeper capital and distribution hurdle against raises of this scale.

Notable Developments

• MediaTek — Secured a $3.5 billion investment from NVIDIA via convertible bonds, deepening a partnership spanning NVLink Fusion custom silicon, RTX/DGX Spark PC chips, and DRIVE AGX automotive AI. The deal ties MediaTek's manufacturing scale to NVIDIA's rack-scale architecture across three separate hardware markets.
• DeepSeek — Is reportedly seeking a $7.4 billion round at a $74 billion valuation, restructuring investor equity toward direct stakes ahead of a targeted 2027 Shanghai IPO. The valuation implies roughly 148 times its disclosed ~$500 million ARR.
• Sharpa — Raised more than $668 million at a $3.3 billion valuation, backed by Alibaba, Meituan, Tencent, and JD.com. Its dexterous robot is running a live 55-step Blizzard Treat preparation workflow inside an unmodified Shanghai Dairy Queen outlet.
• Alibaba — Priced a HK$80 billion (~$10.2 billion) Hong Kong share placement, with all proceeds earmarked for AI infrastructure and full-stack AI capability expansion. The raise follows 75% year-over-year capex growth disclosed the same week.
• Unitree — Shares surged 460% on their Shanghai Stock Exchange trading debut, one of several outsized first-day pops among recent Chinese tech IPOs. The pop values the humanoid robot maker well ahead of its roughly $248 million in prior private funding.
• Naver Cloud — A Naver Cloud-led 32-member consortium was selected over a rival SK Telecom-led bid to build South Korea's state-funded cybersecurity AI foundation model, backed by 256 Nvidia B200 GPUs for an initial five months. The win extends Naver Cloud's position as a default vendor for government AI infrastructure.
• Kakao — Board approved splitting the company into Kakao AI, which will run the KakaoTalk-and-AI platform business, and Kakao X, which retains fintech, content, and mobility units including KakaoBank and KakaoPay. Kakao AI targets over 6 trillion won (~$4.4B) in 2030 revenue.
• SK Hynix — Is evaluating a joint-venture memory fab in Japan as CEO Kwak Noh-jung warned HBM shortages could last through 2030. The company has already committed roughly $38 billion to Korean fabs plus a $4 billion-plus HBM packaging site in Indiana.
• Sakana AI — Released a Namazu-based update to its Sakana Translate service, claiming — per its own internal testing — that it now outperforms Google Translate and DeepL on Japanese-English translation. The launch follows a reported restructuring into a SoftBank- and NEC-backed consortium under the Noetra name.
• Sarvam AI — Added IndiGo Ventures as a strategic backer within its ongoing Series B round, deepening a commercial deployment of voice and agentic AI already running across IndiGo's passenger and operational workflows. The round has drawn $234 million at a $1.5 billion valuation from HCLTech, Bessemer, Khosla Ventures, and Peak XV.

By Country

🇰🇷 Korea

• AimIntelligence — AIM Intelligence joined SK Telecom's consortium to provide AI red-teaming and guardrail technology for South Korea's 'AI for All' agentic assistant, launching by year-end.
• Nota AI — Nota AI joined SK Telecom's 'AI for All' consortium to apply its NetsPresso platform, cutting compute and memory load for Korea's national public-service AI models.
• Motif — Motif is contesting its sovereign-model program exit, arguing that its AAII benchmark lead was diluted by opaque expert and user scoring.
• Wrtn — Wrtn raised KRW 100 billion in Series C funding at a KRW 1 trillion-plus valuation as its consumer AI platforms expand overseas.
• JNP Medi — JNP Medi received KRW 5 billion in Global TIPS support to build an AI platform integrating clinical-trial data, documents, regulation, and workflows.
• Liner — Korean AI agent/search startup Liner raised a 50 billion won (~$36M) Series C led by LB Investment, days after partnering with Rebellions on NPU-paired agent infrastructure.
• Atomo — Atomo, a functional materials startup spun off from LG Electronics and Bluepoint's STUDIO341, secures seed funding to pursue a fabless model, aiming to reduce costs and speed up market entry for therm
• Selectstar — SelectStar is taking Datumo to Tokyo, betting that Japanese enterprises will need recurring, service-specific AI reliability evaluation.

🇨🇳 China

• iFlytek — iFlytek's Ciyuan Xinghuo subsidiary open-sourced Spark X2.5 on-device models with a 1-million-token context window, hitting 90.3% accuracy on smart-home control tasks.
• VAST — VAST raised $446 million in Series B and B+ rounds led by Matrix Partners China, pushing its six-month funding total to roughly $700 million.
• MiniMax — MiniMax's H3 Max video model, tuned by fal.ai, generates 480p five-second clips with audio in about three seconds, priced at $0.05 per second.
• LatentVerse — LatentVerse raised several hundred million RMB in a seed round from Hillhouse Venture and others to build UTAM, a unified tactile-action model for embodied AI.
• MiniMax — MiniMax lifted its three-year Alibaba Cloud spending ceiling 220% to $1.2B after burning two-thirds of its 2026 budget by June, signaling surging compute demand for its open-weight models.
• Ant Group — Ant Group launched Ling-3.0-flash-Fin, a finance-enhanced model for investment research with planned open weights and OpenRouter access.
• 可灵 — Kuaishou's AI video model Keling is valued at ~$18B post-$3B round, matching the parent's market cap and raising questions about a potential IPO arbitrage.
• Z.ai — WSJ profiles Z.ai's Tang Jie and Moonshot AI's Yang Zhilin, once Tsinghua teacher and student, now rival founders of two Chinese LLM startups.

🇯🇵 Japan

• Turing — Turing raised KRW 7 billion (~$5.1M) in a Series B to expand its GPAI academic assistant, which has reached 1.5 million registered users, beyond STEM subjects.
• Quantum Solutions — Quantum Solutions secured about JPY 494 million to begin a Japan-based NVIDIA B300 GPU data-center buildout, starting with supply reservations for 32 servers.
• Arm — Arm disclosed its Neoverse cores shipped over 1.5 billion units in the past year and detailed a dual-instruction-set IBM mainframe chip shipping by 2029.
• LENZO — LENZO refocused entirely on its power-efficient CGLA AI inference chip architecture, led by a former Sony PlayStation chip engineer, as agentic AI demand grows.
• NIAH — Sharp launches NIAH, a metered AI agent service connecting to 52 appliance models, in Japan on September 30, with four tiers up to 1,200 conversations monthly.
• VMware — Broadcom's VMware introduced AI Factory, a Cloud Foundation extension cutting private AI infrastructure buildout from weeks to 30-45 minutes via pre-certified hardware.
• EdgeLLM — APOLLO11 launched EdgeLLM, a fixed-price on-premise generative AI platform for Japan's regulated industries, starting at ¥9.8 million (~$66,000) with zero external data transmission.
• Google — Google Cloud unveiled Gemini Enterprise for Legal, connecting via Model Context Protocol to Thomson Reuters, iManage, and e-discovery platforms, competing with Anthropic's Claude for Legal.

🇮🇳 India

• Sarvam — IndiGo Ventures plans to invest in Indian AI startup Sarvam to apply AI across aviation operations, adding a non-tech strategic investor to its 2026 sovereign-AI funding base.
• Gnani — Gnani.ai launched Artha, a sovereign AI stack pairing its open-weight 30B-parameter Evon v3.3 LLM with the Plexus agent platform for Indian enterprises and public institutions.
• Ringg AI — Bengaluru's Ringg AI extended its Series A to $15M, led by Peak XV Partners, to expand from voice AI into WhatsApp and browser-based agents that act within enterprise systems.
• Runable — Runable raised a $21M Series A to expand its AI-agent platform for small-business development, operations, and growth workflows.

Other East Asia

• MediaTek — MediaTek gained an Nvidia equity investment of undisclosed size, deepening the companies' existing supply relationship spanning the Grace Blackwell superchip platform and automotive chip design.
• MSI — MSI's XpertStation WS300 workstation launched in Japan via Ask Corporation, pairing a 72-core Grace CPU with an Nvidia GB300 Superchip for up to 748GB of coherent memory.
• Alibaba — Alibaba launched Wan3.0, an AI video model generating clips up to 30 seconds from documents, spreadsheets, and slides, alongside its $10.2 billion Hong Kong share placement for AI infrastructure.

Southeast Asia

• nSWX (SG) — nSWX, founded by NASA- and MIT-trained engineers, raised $2 million in a seed round to develop AI chip packaging technology addressing thermal management and interconnect bottlenecks.
• FileAI (SG) — FileAI received an undisclosed strategic investment from SMBC Asia Rising Fund and Singtel Innov8 to expand in Japan and launch its fileScout data-mapping tool.
• Supermom (SG) — Supermom raised S$18 million (~$13.3M) in Series B funding to expand its AI-driven parenting data platform into Indonesia, Malaysia, and Vietnam.
• fileAI (SG) — fileAI secured undisclosed SMBC and Singtel Innov8 backing for Japan expansion and launched fileScout for enterprise data mapping.
• fileAI (SG) — fileAI raised undisclosed strategic funding from SMBC and Singtel to expand in Japan and support enterprise unstructured-data workflows.
• Ant International (SG) — Ant International released Falcon TST 2.0, positioning time-series AI as a production tool for FX exposure and treasury forecasting.
• OM Venture Capital (SG) — OM Venture Capital uses AI to spot startup deals earlier, gaining a competitive edge in the venture capital landscape.
• Qoder (Alibaba) (SG) — Alibaba's Qoder launched a desktop AI coding agent with chat/voice interface and self-verifying Computer Use/Browser Use loop, targeting non-technical business users beyond the IDE.

Compiled from 41 linked sources across AI Market Watch's Asia news index. Fortnight covers 2026-08-21 – 2026-09-04.