Log in Sign up
News AMD's August Deal for Inference Chip Startup Taalas Was Worth $2.9 Billion, Filing Shows, With $1.4 Billion Contingent Business Google Expands Voluntary Exit Buyouts to Five More Countries, With Swiss Staff Offered Up to 12 Months' Pay Apple Intelligence Apple Cuts Engineers' Daily AI Spending Cap to $200, With Deeper Cuts on the Table Meta Meta Buys Conductor, the Startup Whose App Runs Multiple AI Coding Agents in Parallel News MoWorld 2.0 Debuts at Wuzhen: a 4D World Model Trained Entirely on Huawei's Ascend Security Sentra Raises $150 Million From Capital One Ventures to Rein In What AI Agents Can Read Business Kuaishou's Chip Spinoff Lingchuan Raises 1 Billion Yuan to Cut the Cost of AI Video Productivity Y Combinator President Garry Tan Says He Spends Up to $200,000 a Year on AI Tokens News AMD's August Deal for Inference Chip Startup Taalas Was Worth $2.9 Billion, Filing Shows, With $1.4 Billion Contingent Business Google Expands Voluntary Exit Buyouts to Five More Countries, With Swiss Staff Offered Up to 12 Months' Pay Apple Intelligence Apple Cuts Engineers' Daily AI Spending Cap to $200, With Deeper Cuts on the Table Meta Meta Buys Conductor, the Startup Whose App Runs Multiple AI Coding Agents in Parallel News MoWorld 2.0 Debuts at Wuzhen: a 4D World Model Trained Entirely on Huawei's Ascend Security Sentra Raises $150 Million From Capital One Ventures to Rein In What AI Agents Can Read Business Kuaishou's Chip Spinoff Lingchuan Raises 1 Billion Yuan to Cut the Cost of AI Video Productivity Y Combinator President Garry Tan Says He Spends Up to $200,000 a Year on AI Tokens

Kuaishou's Chip Spinoff Lingchuan Raises 1 Billion Yuan to Cut the Cost of AI Video

Business · By: AIReport Admin · 3 hours ago · 👁 15
Kuaishou's Chip Spinoff Lingchuan Raises 1 Billion Yuan to Cut the Cost of AI Video

Lingchuan Technology, the chip venture that grew out of Kuaishou's heterogeneous computing unit, has closed a 1 billion yuan Series B. Its SL200 transcoding chip already runs 99.7 percent of Kuaishou's live-stream transcoding, and a next-generation chip built on fully domestic 3D stacking has returned from the fab ahead of volume shipments next year.

Lingchuan Technology, a Beijing chip startup that grew out of Kuaishou's in-house silicon team, announced on November 4 that it has closed a 1 billion yuan Series B. The round was co-led by Yizhuang Guotou's Yitang Shenghe fund, Qiming Venture Partners and Zhaohui Capital, with more than ten institutions joining, including Guotai Haitong Kaiyuan, CCB Beijing Investment and CDH Baifu, plus strategic backers such as Loongson-affiliated funds. Existing investors Shunxi Fund, Shuimu Investment Group and Jiuzhi Capital added more.

The company was founded in March 2024 as a joint effort between the Beijing municipal AI fund and Kuaishou, built on the skeleton of Kuaishou's heterogeneous computing and chip division. Founder and chief executive Liu Lingzhi, a Shanghai Jiao Tong PhD and University of Washington postdoc, previously ran the division as a Kuaishou vice president after holding technical management roles at Huawei and Intel. The pitch is that the company is a "hardware-based software company": the chip is the floor, and the software stack decides what it can run.

Its first product, the SL200, is a smart video system-on-chip for transcoding — squeezing the same stream into multiple bitrates and resolutions for different devices. The company says it can compress video to as little as a quarter of the original size at comparable quality, a figure that is company-reported and not independently verified. Commercial traction is easier to check: roughly 100,000 SL200 units have shipped to date, Kuaishou says 99.7 percent of its live-stream transcoding runs on the chip, and Alibaba Cloud, Baidu Cloud and Bilibili all buy it despite running rival video platforms of their own.

The economics of transcoding explain the niche. Video workloads are billed per stream or per gigabyte, not per peak FLOPS, so what matters is cost per concurrent channel. Nvidia's NVENC and Intel's QuickSync bundle encoders into general chips at middling efficiency; Canada's NETINT and AMD's Alveo MA35D sell accelerator cards; Tencent's Cang hai and ByteDance's in-house codecs are cloud-only and never sold standalone. Lingchuan is one of the few Chinese vendors selling a dedicated transcoder as a chip, aimed at platforms that operate their own transcoding fleets.

The next product changes process route entirely. A new chip based on fully domestic 3D stacking technology taped out in April and returned from the fab this month, the company said, with volume shipments planned for next year. It targets AI video production — cutting generation cost and latency — and pairs a deterministic-latency DiPU architecture built on RISC-V with a 3D near-memory design using hybrid bonding. The LUCAS software platform has been adapted to Qwen, Tongyi Wanxiang, Hunyuan Video and MiniMax, and the company runs an AIGV platform called Nebula on top.

The funding lands at a moment when video has become the most expensive corner of generative AI. Training and serving video models multiplies token counts by frames, and Chinese labs have been unusually aggressive about cutting the bill: Kuaishou's Kling, Tencent's Hunyuan and Meituan's LongCat all leaned on domestic silicon or domestic compute in recent launches. A transcoding specialist is a smaller bet than a frontier accelerator startup, but it attacks a bill every video platform pays every day.

What makes Lingchuan worth watching is not the headline round but the customer list. Alibaba Cloud and Baidu Cloud build their own video infrastructure, and they still buy this chip — the strongest available signal that the per-stream economics hold up outside the founding shareholder. If the domestically stacked AI video chip ships on schedule next year, the company will find out whether the same logic survives the jump from compression to generation.

Comments (0)

Log in to join the discussion

Log In

No comments yet