
Company Overview
Founded in March 2018 and headquartered in Shanghai's Zhangjiang Hi-Tech Park, Enflame is the first domestic company with both high-performance cloud-based training and reasoning capabilitiesAI chipStartup Company. With “original innovation” as its technology route, the company focuses on providing artificial intelligence industry withfull-stack computing powerThe company's solutions cover chips, acceleration cards, smart computing clusters and software platforms, and its business covers the Internet, finance, smart cities, domestic smart computing centers and other fields. As of 2025, the company has completed several rounds of financing, valuation of more than 20 billion yuan, shareholders, including Tencent, the National Integrated Circuit Industry Investment Fund, CITIC Securities and other top capital, is regarded as a strong contender for “the second share of domestic AI chips”.
Milestones
- Start-up period (2018-2019)
- It was established in March 2018, completed the Pre-A round of financing of 340 million yuan in August (led by Tencent), and released the first cloud-based training acceleration card “cloud flint T10” in December, filling the blank of domestic training chips.
- The core members of the team come from AMD, Huawei Hesse and other international manufacturers, CEO Lidong Zhao once led the AMD global GPU product line research and development, deep technical genes.
- Technology breakthrough period (2020-2021)
- In May 2020, it completed the B round of financing of 700 million yuan, and in December, it released the first reasoning acceleration card “cloud flint i10”, realizing the full coverage of training and reasoning.
- In January 2021, the company completed the C round of financing of 1.8 billion yuan, and released the second-generation training products “Profound Thinking 2.0” chip and “Cloud Flint T20/T21” in July, with a single card FP32 arithmetic up to 150 TFLOPS, and the cluster performance is comparable to that of NVIDIA A100, becoming the first company to release the second-generation training products in China. A100, becoming the first domestic enterprise to release the second generation of training products.
- Explosive growth period (2022-present)
- In September 2022, it released the Cloud Flint Intelligent Calculator, and in 2023, it completed the D round of financing of 2 billion yuan, with a valuation exceeding 17 billion yuan.
- In 2024, mass production of “Suwon S60” reasoning card will be realized, with shipments and orders exceeding 100,000 pieces; in 2025, the fourth-generation training and pushing integrated chip “Suwon L600” will be released, which supports low-precision calculation of FP8, and a home-made 10,000-card reasoning cluster will be built in Qingyang, Gansu Province, to serve the national strategy of “East Counts, West Counts”. To serve the national strategy of "East Counts, West Counts".
- In November 2025, the company restarted the IPO counseling for the Science and Technology Board, and the counseling institution was changed to CITIC Securities, and the listing process continued to advance.
Products & Services
- Hardware Products
- training sideThe cloud flint T20/T21 accelerator card supports training of large models with hundreds of billions of parameters, single card FP32 arithmetic power of 150 TFLOPS, TF32 arithmetic power of 600 TFLOPS, equipped with 4 HBM2E storage chips, bandwidth up to 1.8TB/s, support 64GB memory, adapted to the deployment of ultra-large-scale clusters.
- inference endCloud flint i20 accelerator card realizes INT8 arithmetic 400 TOPS, energy efficiency ratio 1.5 TOPS/W, liquid-cooled version of power consumption is only 150W, supporting high-density deployment, has landed in Tencent cloud, operators and other scenes.
- System level productsCloudFlint provides AI computing center solutions, adapting to more than 30 models of mainstream server vendors, supporting liquid-cooled ultra-large-scale clusters, and reducing deployment costs.
- software service
- Harnessing the TopsRider platform: A computing and programming platform with independent intellectual property rights, compatible with mainstream frameworks such as TensorFlow, PyTorch, etc., supporting operator generalization and graph optimization, with a 30% increase in distributed training efficiency.
- ecological cooperationThe company has jointly launched a national training program with DeepSeek and other vendors, reducing inference latency by 35%; joined the “Hardware Ecology Co-Creation Program” of Flying Paddle and built a closed-loop ecosystem from computing power to applications with server vendors.
- Industry Solutions
- the Internet: Provide large cloud vendors with a cost-reducing and efficiency-enhancing arithmetic base to support large model training and inference.
- financial: Guarantee financial security with original and innovative arithmetic, supporting high concurrency transactions and deployment of risk control models.
- smart city: Efficient inference arithmetic clusters to support urban AI video clouds and realize cross-domain arithmetic sharing.
- National Intelligent Computing Center (NICC): Landed liquid-cooled clusters in Qingyang, Wuxi and Yichang to accelerate breakthroughs in cutting-edge technologies.
core technology
- Chip Architecture
- Designed with 3D integrated circuits for industry-leading integration and efficiency, it supports FP8/FP16/FP32/TF32 multi-precision computation to meet training and reasoning needs.
- The Profil 2.0 chip is equipped with GCU-LARE full domain interconnect technology, providing bi-directional 300GB/s interconnect bandwidth, supporting thousands of accelerated card interconnections with excellent linear acceleration ratio.
- software collaboration
- The Harness Computing Platform gives full play to the performance of the chip through the collaborative architecture design of hardware and software, supporting model training and optimization under mainstream deep learning frameworks.
- Open programming models with extensible operator interfaces provide custom development capabilities for customer model optimization.
- ecological construction
- Launched the “Prairie Fire” program to build a universal heterogeneous computing ecosystem, including technology standardization, research ecosystem, developer ecosystem and industrial ecosystem, starting with original innovation.
- Participate in the development of industry standards, build test platforms, and work with partners to build an ecosystem of continuous innovation.
development prospect
- policy dividend
- The national “East counts and West counts” strategy to promote the localization of arithmetic, Enflame in Qingyang, Wuxi and other places in the layout of the Smart Computing Center, in line with the policy guidance, is expected to obtain more resources tilted.
- The support of the Science and Technology Board for companies in key technology areas has injected new momentum into the listing and long-term development of the company.
- market
- The global AI chip market is growing rapidly, and Deloitte predicts that the scale will exceed $150 billion in 2025 and reach $400 billion in 2027. Enflame as a domestic head company, is expected to occupy a significant share of the training and reasoning market.
- The demand for computing power in the Internet, finance, smart city and other fields continues to climb, and the company's full-stack solution has cost and performance advantages, with broad commercialization space.
- Technology Iteration
- The company adheres to the strategy of rapid iteration, and the L600 chip released in 2025 supports FP8 low-precision computation, which can improve the speed of large model training and reduce costs, and further consolidate its technological leadership.
- Co-established 28nm mature process capacity guarantee system with SMIC and Huahong Group to cope with geopolitical risks and ensure supply chain stability.
- capital assistance
- Tencent has participated in six consecutive rounds of investment, holding 21.37% as the largest shareholder, the National IC Industry Investment Fund, CITIC Securities and other strategic shareholders to provide financial and resource support to help technology research and development and industrial expansion.
data statistics
Relevant Navigation

It is a high-tech listed company dedicated to providing total solutions for intelligent manufacturing with robotics technology as the core, covering industrial robots, mobile robots, special robots and intelligent service robots.

BrainCo
Focusing on the R&D and industrialization application of non-invasive brain-computer interface technology, we are committed to solving the core pain points in the fields of medical rehabilitation, health management and education through the analysis of EEG signals and intelligent interaction technology.

Huichuan Technology
Focusing on automation, digitalization and intelligence in the industrial field, we concentrate on the core technologies of "information layer, control layer, drive layer, execution layer and sensing layer".

Kunlun Chip
Focusing on the R&D of general-purpose AI computing chips and full-stack hardware/software technologies, we provide a complete product portfolio ranging from chips and accelerator cards to supernodes and scenario-specific solutions. We are committed to advancing the widespread adoption of intelligent computing through our proprietary XPU architecture.

Guoxin Technology
A chip design company focusing on the research and development of domestically owned and controlled embedded CPU technology and industrialized application, with a rich product line and strong R&D strength.

Haier Smart Home
Intelligent Manufacturing in the Home Appliance Industry

Sankyo Precision
Industrial automation equipment, precision molds, ultra-precision components and products

DJI
DJI solutions for airports and unmanned survey operations
No comments...
