Sunrise
Category: AI Chips / Semiconductors
Chinese fabless chipmaker spun off from SenseTime that builds inference-only GPUs, servers and supernodes with a full software stack to cut the cost per token of serving large models and agents. Sunrise was founded in 2024. The company is led by Wang Zhan (Co-CEO; Wang Yong also serves as Co-CEO). Based in Hangzhou, China. Team size: 101-500. Total funding raised: $830M. Latest round: Series B. Key investors include Hangzhou Capital, IDG Capital, Gaorong Ventures (高榕创投), SANY Huaxu Fund, 4Paradigm, Hangzhou Data Group, CICC Capital, Wuji Capital, PICC Capital, CCB Equity, Janchor Partners, Casstar (中科创星).
- Founded
- 2024
- Headquarters
- Hangzhou, China
- Team size
- 101-500
- Total funding
- $830M
Value proposition
Lowers the unit token cost and TCO of large-model and agent inference. Its GPUs are designed for inference from the architecture stage: they drop training-related modules, use LPDDR6 memory instead of HBM, and support FP4 natively.
Products and solutions
Qiwang (启望) S1/S2/S3 inference GPUs (S3 launched Jan 2026 for LLM and agent inference), Zhiwang (智望) S1/S2 compute cards, Chenwang (辰望) AI servers and agent all-in-one appliances, Huanwang (寰望) 256-card supernode and compute clusters, Xiwang workstations, SIRE software stack (toolchain, model engine, MaaS), inference cloud / Token Factory offering
Unique value
Sunrise says it is China's first GPU company with an 'All-in Inference' strategy. It also claims S3 is the first domestic GPU to use LPDDR6 and PCIe Gen6. Its own figures put S3 at about 5x the performance of S2 with about 90% lower cost per token; the benchmark methodology has not been disclosed.
Target customer
Cloud and compute providers, telecom operators, enterprises and public-sector institutions in China that deploy LLM and agent inference
Industries served
Telecom operators, manufacturing, finance, public utilities, healthcare/pharma, green energy, entertainment/AIGC, cloud token services
Technology advantage
Builds its full stack in-house: instruction set, GPGPU IP, compiler toolchain and cluster scheduling. The company reports 100+ core patents and first-pass tape-out success, two generations (S1, S2) already in mass production, and more than 10,000 chips delivered in 2025. It inherits 8 years of R&D (about RMB 2B) from SenseTime's large-chip unit, and its software stack is compatible with mainstream AI ecosystems.
How they differentiate
Sunrise focuses purely on inference, while listed peers such as Moore Threads (full-function GPU) and Biren/Enflame (training plus inference) cover broader workloads. It takes a capacity-first, lower-cost memory route (LPDDR6 instead of HBM) and competes on cost per token rather than peak specifications.
Main competitors
Moore Threads, MetaX (沐曦), Biren Technology, Enflame, Cambricon
Key partnerships
SenseTime (parent and origin of the chip unit), strategic industrial investors including SANY Group (Huaxu Fund), 4Paradigm, Hangzhou Data Group, Chia Tai/CP Group, GCL Technology and Tongcheng Travel
Notable customers
Sony (reported S1 IP licensing customer), Xiaomi (reported S1 IP licensing customer)
Major milestones
2018: GPU R&D starts inside SenseTime. S1 and S2 reach mass production. Late 2024: spun off from SenseTime. Early 2025: Wang Zhan joins as co-CEO. Jan 2026: S3 launched at the first Sunrise GPU Summit, alongside a disclosure of about RMB 3B raised over the prior year. Apr 2026: >RMB 1B round led by Hangzhou Capital. May 2026: joint-stock conversion and start of IPO process. Aug 2026: RMB 2B Series B at about RMB 20B valuation. Oct 2026: S3 and 256-card supernode featured on CCTV's 'Dialogue'.
Growth metrics
About 500 employees, roughly 80% in R&D. More than 10,000 GPUs delivered in 2025. At least seven financing rounds since the late-2024 spin-off, totaling about RMB 6B. Valuation went from over RMB 10B (Apr 2026) to about RMB 20B (Aug 2026).
Market positioning
Leading private inference-GPU unicorn among China's domestic GPU makers, valued at about RMB 20B post-money (Aug 2026). The 'four little dragons' (Moore Threads, MetaX, Biren, Enflame) are already public. Sunrise positions itself as the AI industry's 'P&L optimizer' for token costs and has started its IPO process.
Geographic focus
China (domestic substitution for AI compute), with some presence at international exhibitions such as in Bangkok
Patents and IP
Company reports 100+ core patents and proprietary IP, including a self-developed instruction set and GPGPU IP
About Wang Zhan (Co-CEO; Wang Yong also serves as Co-CEO)
Ex-Baidu founding team member (2000-2016), Senior VP and group legal representative; known as the 'father of Phoenix Nest' (凤巢), Baidu's search ad system; Peking University BSc Physics. Joined Sunrise as co-CEO in early 2025 to run operations, product and commercialization.
Latest news about Sunrise
More AI Chips / Semiconductors companies
Official website: https://sunrise-ai.com