Company at a glance
Scale
Private, not exchange-listed
Growth prospects
Expanding customer base and product roadmap
via eu.36kr.com ↗Tencent (with concert parties) holds over 20% equity (20.26% reported), the largest external shareholder, invested since the Pre-A round; Tencent con…via eu.36kr.com ↗2023 RMB 301 million; 2024 RMB 722 million; 2025 RMB 990 million; net losses RMB 1.665B (2023), RMB 1.51B (2024), RMB 1.164B (2025), cumulative about…China competition
Competes with Huawei Ascend, Biren, Moore Threads, Cambricon in AI accelerators — peers listed at sector level.
SubstituteGlobal comparables
Global reference set: Nvidia, AMD, Google TPU.
Nvidia (H20 for China inference market)Technical gap
Sector rated substitute vs the frontier.
via qbitai.com ↗邃思 (Suisi) DSA-architecture training/inference chips and 云燧 (CloudLight) DTU accelerator cards; 云燧i20 inference card (2021-12) with 819 GB/s memory b…Obstacles
Sector bottleneck applies
CUDA-like software ecosystem, HBM supply, advanced packaging, leading-node access, and large-scale deployment proof.All sourced facts (8)
- Founded: 2018-03-19 in Zhangjiang, Shanghai, by CEO Zhao Lidong (ex-Unis/紫光 communication VP for semiconductor investment) and co-founder Zhang Yalin; both AMD alumni with 10+ years there via pitchhub.36kr.com ↗
- Funding history: 10+ rounds totaling about RMB 7 billion: Pre-A RMB 340M (2018); A RMB 300M (2019); C round ~USD 279M with Tencent (2021-01); D round RMB 2 billion led by Shanghai International Group with Tencent and Meitu (2023-09); reported valuation about RMB 16 billion via m.caixin.com ↗
- Tencent relationship: Tencent (with concert parties) holds over 20% equity (20.26% reported), the largest external shareholder, invested since the Pre-A round; Tencent contributed 33.34% of revenue in 2023, 37.77% in 2024, and 83.79% in 2025; chips run behind WeChat and Tencent Meeting workloads via eu.36kr.com ↗
- Products: 邃思 (Suisi) DSA-architecture training/inference chips and 云燧 (CloudLight) DTU accelerator cards; 云燧i20 inference card (2021-12) with 819 GB/s memory bandwidth and INT8 256 TOPS; fourth-generation L600 claimed to match and exceed Nvidia H20 via qbitai.com ↗
- Revenue trajectory: 2023 RMB 301 million; 2024 RMB 722 million; 2025 RMB 990 million; net losses RMB 1.665B (2023), RMB 1.51B (2024), RMB 1.164B (2025), cumulative about RMB 4.3 billion over three years via eu.36kr.com ↗
- IPO status: SSE STAR Market application accepted 2026-01-22; listing committee approved 2026-06-15; CSRC registration ~2026-07-09/10 for about USD 883M (RMB ~6 billion); preliminary price inquiry 2026-08-28, online/offline subscription 2026-09-02; code 688801 via reuters.com ↗
- Use of proceeds: R&D and industrialization of fifth-generation (due 2027) and sixth-generation (due 2029) AI chips via eu.36kr.com ↗
- Milestone: Built China's first 10,000-chip domestic inference computing cluster in Qingyang, Gansu Province (2025); DSA architecture is CUDA-incompatible and focused on inference via eu.36kr.com ↗
Tracked chips: Enflame Suiyuan/CloudBlazer DTU (current: S60, L600)
About
Fabless cloud AI accelerator design: sells 云燧 CloudLight DTU training and inference accelerator cards based on in-house 邃思 DSA chips (fourth generation L600), deployed in cloud data centers; increasingly reliant on Tencent procurement (83.79% of 2025 revenue).
Founded 2018
Why it matters
Enflame is among the few Chinese AI chip startups with large-scale commercial cloud deployment (revenue CAGR above 80% from 2022-2025, ~10x growth) and built China's first 10,000-chip domestic inference cluster in Qingyang, Gansu in 2025; its IPO is one of 2026's largest STAR listings.
Where it sits in the stack
Key semiconductor products
- Enflame software stack (TopsRider) for cloud deployment
- AI training and inference accelerator chips
Chip pricing
- Enflame Suiyuan/CloudBlazer DTU (current: S60, L600)~$1.9K · 2024-12-31
Ecosystem
Investors / affiliates
- Enflame Technology
Global peer set
- Nvidia (H20 for China inference market)
- Huawei Ascend
- Hygon DCU
- Cambricon (domestic listed peer)
Affiliates
- Tencent (reported backer)
Competitive position
- Cloud AI inference accelerator positioned against Nvidia H20-class China-market parts; L600 claimed to surpass H20
- Deeply embedded in Tencent's cloud infrastructure as the dominant domestic inference supplier there
- One of the few Chinese AI chip startups with four shipped generations and ~RMB 1 billion revenue scale pre-IPO
- Positioned as a domestic alternative to NVIDIA GPUs for AI training and inference; specific performance and process node not confirmed in this signals batch.
- Depends on advanced foundry access constrained by US export controls; adoption bottleneck is production-scale deployment in Chinese cloud platforms.
- Global comparables: Nvidia (H20 for China inference market)
- Global comparables: Huawei Ascend
- Global comparables: Hygon DCU
Milestones and signals
- 2018—Enflame Technology was founded.
- 2023—Valuation (Pre-IPO Round): CNY ~16 billion
- 2024—Revenue: CNY 722 million
- 2025—Revenue: CNY 990 million
- 2026—IPO Raise: CNY ~6 billion (~USD 883-892 million)
What to watch
- IPO pricing (inquiry 2026-08-28) and first-day trading around September 2026
- Customer concentration: Tencent at 83.79% of 2025 revenue and Tencent's in-house 'Zixiao' 紫霄 chip as a competitive risk
- Fifth-generation chip delivery in 2027
- Path to profitability versus ~RMB 1.2 billion annual losses
Questions
Last reviewed 2026-07-11Answered questions
What is the measurable gap versus global reference companies?
Against NVIDIA, Nvidia (H20 for China inference market), Huawei Ascend, and Hygon DCU: Positioned as a domestic alternative to NVIDIA GPUs for AI training and inference; specific performance and process node still unclear in this signals batch.
Open questions
- Which products are qualified beyond pilots in fabless?
- Which customers, fabs, or end markets show real adoption rather than policy interest?