AI Market Watch Asia · Fortnight 17: Kling AI raises a record $3 billion funding round at an $18 billion valuation

Kling AI raises a record $3 billion funding round at an $18 billion valuation

Meets the record-or-milestone criterion as a reported global record single funding event for an AI video-generation company.

Kling AI Secures Record $3B in Independent Funding for Video Generation

Kling AI, the video generation model spun out of Kuaishou, completed a $3 billion independent funding round at an $18 billion valuation, setting a global record for a single funding event in the AI video generation sector, per EqualOcean via Sohu. This is the largest single-round investment in a generative AI video company to date, surpassing the $2.8 billion round reported in early July 2026, making it the most consequential development this fortnight because it consolidates capital concentration in late-stage players and raises the barrier for new entrants in the sector.

The funding is part of a broader surge in AI video generation investment, with the sector attracting $9.147 billion across 29 deals in the first seven months of 2026, and it underscores a pivot from traditional video generation toward world models and 3D generation that offer interactive, physically plausible environments, per the Sohu report. The involvement of Alibaba and Tencent in Kling AI's funding highlights the strategic importance of controlling model layers and driving cloud consumption, as hyperscalers place multiple bets across the generative AI stack.

Kling AI's substantial round signals that pure video generation capability is no longer sufficient for a standalone venture, with success now hinging on achieving scale, integrating with broader AI ecosystems, and moving toward spatial intelligence, per the Sohu analysis. As the sector matures, the focus will shift to commercial application and sustainable revenue, marking a transition from technology competition to survival competition based on scale and distribution, which explains why capital is increasingly concentrating on proven, late-stage players.

The record round sets a new capital barrier for the video generation segment, with smaller competitors facing structural pressure from compute costs, R&D burn, and enterprise customer acquisition, likely triggering consolidation in the coming quarters.

Alibaba releases Qwen3.8-27B with reported agent benchmarks above Claude Opus 4.6 Max

Meets the quantified product-claim criterion through disclosed coding and agent benchmark results paired with reported execution on a single 24GB consumer GPU after quantization.

Alibaba released Qwen3.8-27B on August 16, a 27-billion-parameter open-weight multimodal model that scored 8.3 points higher than Claude Opus 4.6 Max on SWE-bench Pro and 15.2 points higher on QwenSWEBench, with the model running on a single 24GB consumer GPU after quantization, per the QbitAI report. This ranks above other developments this fortnight because it is the first open-weight model of this size to pair disclosed coding and agent benchmark results with reported execution on commodity hardware, a quantified product claim that directly challenges closed-model pricing assumptions.

The model supports a native 262K-token context expandable to 1M and features Gated DeltaNet linear attention, with agentic scores of 70.7 on CoWorkBench and 84.3 on OSWorld-Verified versus Opus 4.6 Max's 68.2 and 72.7, per the QbitAI report.

The model integrates with Transformers, vLLM, SGLang, and TokenSpeed, and includes adjustable reasoning depth for local agent deployment, per the QbitAI report. This follows Alibaba's August 13 release of Qwen3.8-2.4T-A95B, a 2.4-trillion-parameter MoE model with day-zero adaptation across nine AI chip platforms, per the prior Qwen release coverage.

Separately, Alibaba Cloud brought its third South Korea data center online on August 19, adding six Agentic AI services for local enterprises and expanding to 104 availability zones, per the Alibaba Cloud announcement. The Qwen app introduced paid memberships on August 10, priced RMB 19 to RMB 1,499, with third-party agent support, per the Qwen app monetization report. The company claims the model's agentic and coding performance is available on a single RTX 3090 or 4090 after quantization, pairing the prior quarter's infrastructure expansion with new consumer-hardware capabilities, per the Qwen release note.

By contrast, no other event this fortnight disclosed comparable head-to-head benchmark results against a named frontier closed model with verified consumer-hardware execution details.

Notable Developments

• Day1 Company — Acquired enterprise AX specialist SpaceY and launched DAY1 AI Deployment Company, with plans to hire about 100 specialists. The deal extends its education business into FDE-led AI implementation and operational support.
• Alibaba Cloud — Brought its third South Korea data center online and added six agentic-AI services for local customers. The expansion increases local capacity for enterprises requiring lower-latency and data-residency options.
• VeriSilicon — Reported first-half revenue growth of 91.37% to RMB 1.864 billion, while AI-computing work represented about 90% of newly signed orders. Its RMB 12.449 billion order book points to sustained domestic demand for AI chip-design services.
• Ministry of Science and ICT — Advanced the SK Telecom, LG AI Research, and Upstage teams to the next phase of South Korea's sovereign foundation-model project. The submissions included SKT's reported 32.2-point gain across domestic benchmarks and Upstage's 1-million-token model design.
• Alibaba Qwen — Passed 2 billion Hugging Face downloads and generated more than 151,000 derivative models, according to Hugging Face's 2026 Open Model Report. The reported derivative-model base exceeds Meta's Llama ecosystem by 4.7 times.
• Infiforce — Raised nearly $150 million across Series A and A+ rounds to develop embodied-AI models and expand robot deployments. It says its robots operate across more than 30 Chinese cities and 100 scenarios, creating a data loop for its training infrastructure.
• Graas — Raised $17 million and acquired Trustana to expand its retail AI-agent platform. The transaction adds product, customer, and inventory-data capabilities for e-commerce and offline retail operations.
• Motif Technologies — Open-sourced its 314-billion-parameter MoE model, including weights, training code, and libraries under an MIT license. The company reports a top South Korean ranking on the Artificial Analysis Intelligence Index while activating 13.2 billion parameters per token.
• Moore Threads — Approved a Hong Kong H-share listing plan after first-half revenue rose 147% to RMB 1.7 billion. Its net loss narrowed sharply, signaling a more mature capital phase for China's domestic GPU challengers.
• Firmus — Secured $2 billion to expand AI data-center capacity across Australia and Asia-Pacific. The buildout extends an existing 360MW Batam project developed with NVIDIA.

By Country

🇰🇷 Korea

• Liner — Liner partnered with Rebellions to market AI agent software running on NPU infrastructure in Korea and Saudi Arabia, where Liner already supplies search technology to HUMAIN.
• Superb AI — Superb AI joined LG AI Research’s K-EXAONE consortium as it advanced to South Korea’s sovereign foundation-model project third round, handling vision, video, and physical AI development.
• iHW — iHW raised ₩52 billion in Series A financing to scale production of its INFERTRON analog compute-in-memory edge AI chip and develop LLM chip prototypes.
• SECA — SECA secured undisclosed seed funding from LG Electronics and BluePoint Partners after spinning out of Studio341, while piloting its AI-for-EDA platform with Airbus and LG divisions.
• PMatch — PMatch secured undisclosed funding from the Samsung Fire & Marine Insurance–Infobank Value-Up Venture Fund No. 2 to expand its Pilltalk polypharmacy-management service.
• Medicall — Medicall raises seed funding from AI Angel Club, CNTTech, TaleVentures, and Antler Korea to expand its AI phone assistant platform for dental clinics.
• N.Dlight — Korean robotic digital-twin data company N.Dight raised ~$10.5M Series B from KDB Industrial Bank, Naver, and others to expand its physics-aware 3D training data platform for physical AI robots, curre
• Rebellions — Rebellions achieves 3 trillion won valuation as on-device AI NPUs gain prominence over GPUs, signaling a market shift.

🇨🇳 China

• Mifeng — Mifeng received funding led by China Telecom, marking state-backed telecom investment in the Chinese embodied AI startup; financial terms were not disclosed.
• DeepSeek — DeepSeek’s V4 Flash completed 53.8% of real-world agent tasks in a new evaluation, while the company raised API prices by up to 1,100%.
• Zhipu — Zhipu AI releases GLM-5.3, its latest LLM, amid a capital-intensive race for dominance in China's foundation model market.
• Westlake Robotics — Westlake Robotics raises $14.8M Series A to develop a unified whole-body AI model for humanoid robots, totaling $74M in funding.
• Pragmatik Labs — Ex-Qwen lead Lin Junyang founds Pragmatik Labs, an AI agent startup backed by Sequoia Capital China, Gaorong, Tencent, and Shanghai Future Industries Fund.
• Acorn Robot — Acorn Robot raised a multi-hundred-million-yuan angel round led by China Merchants Group and NIO Capital to advance its 'zero-data' embodied AI approach.
• MiniMax — MiniMax H3 open-weight video model matches Seedance 2.0 performance, runs locally on RTX 4090, and offers Omni Reference for multi-modal consistency.
• Om AI — Om AI raises hundreds of millions and releases edge-native VLX models, positioning itself at the forefront of physical AI.

🇯🇵 Japan

• Hmcomm — Hmcomm announced annual strategic investments of ¥100 million to ¥300 million to expand the ecosystem around its Voice AI Agent Platform.
• NexusAdvisors — NexusAdvisors raises 50 million yen seed for AVACAREER, an AI recruitment platform where AI agents interview each other to score candidate fit.
• Nitto Kogyo — Nitto Kogyo installed its direct-liquid-cooled OCP ORV3 rack, rated at 150kW, at MC Digital Realty’s DRIL AI and HPC validation facility in Chiba.
• MIXI — MIXI released the beta RomiCore SDK, enabling its Romi conversational robot to connect with IoT, healthcare, and municipal systems through a Python runtime.
• GaiXer — FIXER launched Lenovo-co-developed Sovereign GaiXer, a roughly 1-petaFLOPS on-premise generative AI workstation priced at ¥3,806,400 with a three-year warranty.
• Contract One AI Agent Powered by Claude — Contract One introduced an Anthropic Claude-powered agent that searches across contract databases and automatically reviews uploaded drafts against internal company standards.
• Preferred Networks — Preferred Networks said its PLaMo 3.0 Prime tokenizer reduces Japanese token consumption by 20–30%, while its Noetra venture targets domestic physical-AI model training by 2027.
• Google — Google argued that Go supports AI-assisted engineering through static typing, gofmt formatting, and the govulncheck security tool.

🇮🇳 India

• Consint.AI — Consint.AI raises Rs 22 crore Series A to build a fraud-detection foundation model and expand internationally across key markets.
• Tagbin — Tagbin launched Brixx, a platform that converts natural-language prompts into architectural designs, BIM models, drawings, cost estimates, and project plans.
• Nugget — Eternal spins off its AI platform Nugget into a standalone enterprise AI business, targeting banks and retailers with conversational AI and workflow automation.

Southeast Asia

• Graas (SG) — Singapore-based Graas raises $17M and acquires Trustana to expand its retail AI agent capabilities.
• Gprnt (SG) — MAS spinout Gprnt seeks $10M Series A to expand AI-driven compliance solutions across Asia.
• Acrab (SG) — Acrab raises $130M to scale its Agent Box platform for on-device inference, signaling SEA's push into AI hardware.
• ThinkAR (SG) — ThinkAR launched AiLENS V1, a 38-gram AI smart glass with dual microphones and a heads-up display for consumer augmented-reality use.
• GENCOW (SG) — GENCOW launches AI-powered software creation platform to automate repetitive backend development tasks, aiming to accelerate software delivery.
• Stratsync (SG) — Stratsync launches multi-agent AI to detect margin risks across ERP, PDFs, and WhatsApp, signaling a shift toward agentic finance operations.
• AI races beyond chatbots (SG) — AI market expands beyond chatbots into wearables and drug discovery, signaling new opportunities for startups and investors.
• Exabytes (MY) — Exabytes launches GROW AI and partners with Aurora Mobile's GPTBots.ai to bring enterprise AI to Southeast Asian businesses, targeting one million by 2030.

Compiled from 37 linked sources across AI Market Watch's Asia news index. Fortnight covers 2026-08-07 – 2026-08-21.