A chip startup most people outside China's video industry have never heard of just closed one of the larger private rounds of the season. The money is not for chasing headline model benchmarks. It is for a narrower, more commercial problem: video is expensive to move, and even more expensive to generate.
Lingchuan Technology (凌川科技), a Beijing company spun out of Kuaishou's internal chip unit, announced a RMB 1 billion (≈ US$140M / HK$1.1B) Series B. The round was co-led by the Yitang Shenghe Fund under Beijing E-Town's state investment platform, Qiming Venture Partners, and Zhaohui Capital. Roughly twenty institutional and corporate backers joined in total, with existing shareholders Shunxi Fund, Shuimu Investment Group, and Jiuzhi Capital adding follow-on money.
The company has not disclosed per-investor amounts or its post-money valuation. What it has disclosed is where the money goes: next-generation chip R&D and mass production, plus international expansion.
Why a video platform built its own silicon
Lingchuan's origin story explains its focus. The company began life as Kuaishou's heterogeneous computing and chip division, became independent in March 2024, and was jointly set up by Kuaishou and the Beijing municipal artificial intelligence fund. Founder and CEO Liu Lingzhi — a PhD from Shanghai Jiao Tong University and a University of Washington postdoc who previously held senior technical management roles at Kuaishou, Huawei, and Intel — started building Kuaishou's chip team from scratch in 2018.
The logic was defensive at first. Live-streaming and short-video platforms re-encode massive volumes of video every day for different devices and network conditions. Off-the-shelf general-purpose chips do this work at a high energy and dollar cost. Lingchuan's first product, the SL200 video chip, handles transcoding, compression, and analysis as a dedicated workload.
That bet has turned into real revenue. According to company and investor disclosures, SL200 has sold nearly 100,000 units to date, with customers including Kuaishou, Alibaba Cloud, Baidu Cloud, and Bilibili. Kuaishou alone routes about 99.7% of its live-streaming transcoding through the chip. This year the chip has also reached data centers, security, and media clients through telecom operator partnerships.
The new chip targets video generation, not just delivery
The Series B funds a second act. Lingchuan's next chip is built on a domestically produced 3D stacking process (全国产3D堆叠), taped out in April 2026 with silicon back and volume shipment expected next year. Where SL200 processes existing video, the new part is aimed at the compute-heavy side of the ledger: running large models that turn text and images into video.
On the technical side, the company sits inside the RISC-V open-source ecosystem, has published a delay-deterministic "DiPU" architecture, and claims a first for a 3D near-memory design that integrates chips of different functions and process nodes via hybrid bonding and advanced packaging. Its LUCAS unified computing platform has been adapted to mainstream open video models including Qwen, Tongyi Wanxiang, Hunyuan Video, and MiniMax, and a production platform called Nebula covers the text-to-video and image-to-video workflow end to end.
The market logic is straightforward. Video generation startups live or die on the cost per usable clip — generation speed, re-roll economics, and the share of output that actually makes it to air all hit the same budget line. Dedicated silicon that lowers that line is, in principle, easier to sell than another general accelerator. According to Li An, chief scientist at BrainNet (脑机网), dedicated silicon tuned to video-generation workloads could compress the cost curve that has kept AI video a margin-pressured business.
The pattern underneath this round
Step back from the company and the round says something about where private capital in China is going.
- Vertical, revenue-backed silicon beats general-purpose promises. Lingchuan's fundraising follows a 2026 pattern in which chip startups with shipped products and named customers raised far more easily than those selling future benchmarks. Peers in the 3D chip space closed large rounds within weeks of each other this autumn.
- State capital is anchoring hard-tech (硬科技) rounds. A Beijing E-Town state fund co-leading, alongside a provincial state insurer's fund and multiple government-adjacent vehicles, shows municipal industrial policy still functioning as first-loss capital for domestic chipmaking.
- Platform spinouts are becoming a supply chain of their own. Kuaishou previously spun out its video generation business, which raised its own large round earlier this year; now the chip arm has followed. Internet platforms in China are systematically converting internal cost centers into external, fundable companies.
None of this removes the central risk of the category. Whether Lingchuan's second chip succeeds depends on physical manufacturing yields, on customers qualifying the part, and on whether video generation demand keeps growing fast enough to need it.
Key takeaways
- Lingchuan Technology raised a RMB 1 billion (≈ US$140M / HK$1.1B) Series B co-led by Beijing E-Town's Yitang Shenghe Fund, Qiming Venture Partners, and Zhaohui Capital; valuation and per-investor amounts were not disclosed.
- The company spun out of Kuaishou in March 2024; its SL200 video chip has sold nearly 100,000 units and handles about 99.7% of Kuaishou's live-streaming transcoding.
- Proceeds fund a next-generation 3D-stacked chip for video generation workloads, taped out in April 2026, with volume shipments expected next year.
Honest limitations
- Valuation, per-investor ticket sizes, and revenue figures are not independently audited; the customer and shipment numbers come from company and investor disclosures rather than third-party verification.
- The "volume shipment next year" timeline is a company expectation, not a confirmed order book; tape-out success does not guarantee manufacturing yield or customer adoption.
- Reporting relies primarily on Chinese financial media (including China Fund News and STAR Market Daily) relaying company announcements; no international wire service has independently covered the round, and the company's own English-language materials are limited.
What readers can do now
- If you evaluate AI video vendors, ask them directly what silicon their pipeline runs on and what their cost per finished minute is — the chip layer, not the model layer, increasingly decides their pricing.
- If you track China semiconductor startups, watch for the next chip's customer qualification announcements rather than the funding headlines; qualification, not financing, is the harder gate.
- If you build on RISC-V or heterogeneous packaging, Lingchuan's public architecture materials on the DiPU and 3D near-memory designs are worth a close read as an early data point on where domestic video silicon is heading.