Enflame Technology

Enflame Technology, officially Shanghai Enflame Technology Co., Ltd., is a Shanghai-based AI chip company founded in March 2018. The company is registered in Shanghai’s Lingang New Area and focuses on cloud-side AI compute for data centers. It was co-founded by Zhao Lidong and Zhang Yalin, with Zhang also listed as the company’s legal representative in its STAR Market filing materials.

‍

Enflame positions itself as a full-stack provider built around proprietary cloud AI chips and system delivery. Its disclosed product scope spans AI chips, accelerator cards and modules, intelligent computing systems and clusters, plus a self-developed AI computing and programming platform called TopsRider that includes drivers, compilers, operator libraries, and toolchains. The same filing highlights in-house interconnect work and a software stack designed to reduce model migration effort onto Enflame hardware.

‍

In the latest milestone, Enflame’s STAR Market IPO application was accepted on January 22, 2026. Chinese financial reporting around the filing said the company targets roughly RMB 6 billion in proceeds, with planned spending on fifth-generation and sixth-generation AI chip R&D and industrialization, plus an advanced hardware-software co-innovation project. Its sponsor document also notes that thousand-card and ten-thousand-card intelligent computing center projects have already generated revenue and that the company is working with customers on supernode designs and larger high-speed interconnect clusters.

Enflame Technology is a Shanghai-based, fabless designer of cloud AI compute hardware and software for data centers. Its product system spans AI chips, accelerator cards and modules, intelligent computing systems and clusters, plus its self-developed full-stack software platform TopsRider (drivers, compiler language and compiler, operator libraries, and toolchains). Enflame positions its architecture around a proprietary instruction set, its GCU-CARE compute unit and GCU-LARE high-speed interconnect, with the latest generation described as natively supporting FP8 for large-model training and inference workloads.

‍

How Enflame makes money

‍

1. Accelerator hardware sales (chips, cards, modules)
   Revenue is primarily tied to shipments of AI accelerator products used in cloud and enterprise data centers.
   ‍
2. Systems and cluster deliveries
   Enflame sells integrated intelligent computing systems and cluster solutions. Its STAR Market filing documents state that thousand-card and ten-thousand-card intelligent computing center projects have already generated revenue, and the company is working with customers on “supernode” designs and larger high-speed interconnect clusters.
   ‍
3. Software platform as the adoption layer (TopsRider)
   TopsRider is positioned as the bridge between Enflame hardware and AI applications, aimed at reducing model migration and development friction versus CUDA-centric workflows. This software stack supports deployments and performance tuning at scale.
   ‍
4. Direct sales plus AVAP model sales
   The filing documents break out revenue exposure via direct sales and AVAP model sales, especially in the Tencent relationship.
   ‍

Market position in China

‍

• Fast revenue growth, large losses
  Revenue rose from RMB 90.1 million (2022) to RMB 301.2 million (2023) to RMB 722.4 million (2024). For Jan–Sep 2025, revenue was RMB 540.2 million. Net loss stayed large, including RMB 1.51 billion (2024) and RMB 887.8 million (Jan–Sep 2025).
  ‍
• R&D intensity remains the defining feature
  R&D spend is disclosed as a high share of revenue, including 181.66% (2024) and 164.77% (Jan–Sep 2025).
  ‍
• Customer concentration is extreme, with Tencent at the center
  Sales to the top five customers represented 96.41% of revenue in Jan–Sep 2025. Tencent-related sales represented 71.84% in the same period, and the filing frames these as related-party transactions and expects a high Tencent share to persist for a period.
  ‍
• Competing in a market still dominated by non-China vendors
  The filing materials cite international vendors holding over 70% of China’s AI accelerator card market share in 2024, with domestic competition also coming from large local players like Huawei HiSilicon and Cambricon.
  ‍
• Capital markets milestone
  Enflame’s STAR Market IPO application was accepted on January 22, 2026, with reported fundraising target of RMB 6 billion.

China is Enflame’s full focus market, with deployments tied to domestic data centers and large internet platforms. In its STAR Market filing materials, Enflame says its accelerator cards, systems, and cluster solutions are already in large-scale commercial use across internet AI workloads, spanning traditional models and large models. It also states that thousand-card and ten-thousand-card intelligent computing center projects generated revenue during the reporting period, including participation in national “East Data West Computing” hub projects and expanded cooperation with domestic telecom operators.

‍

Commercial performance is concentrated in a small buyer set. The filing discloses that in Jan–Sep 2025, the top five customers accounted for 96.41% of revenue, and Tencent-related sales accounted for 71.84%, across direct sales and an AVAP model.

‍

A key near-term milestone is capital markets access. Enflame’s STAR Market IPO application was accepted on January 22, 2026.

Enflame’s growth plan centers on scaling domestic data center AI compute while funding a heavy R&D roadmap. Its STAR Market IPO application was accepted on January 22, 2026, and Reuters reported the company targets about RMB 6 billion in proceeds.

‍

In its STAR Market filing materials, Enflame describes a full-stack product system that spans cloud AI chips, accelerator cards and modules, intelligent computing systems and clusters, plus the TopsRider software platform (drivers, compiler toolchain, operator libraries, tools). It also states that thousand-card and ten-thousand-card intelligent computing center projects generated revenue, and that it is working with customers on supernode designs and larger high-speed interconnect clusters.

‍

Key growth drivers include

‍

1. Next-generation chip investment
   Public reporting around the filing says IPO proceeds are aimed at R&D and industrialization for fifth-generation and sixth-generation AI chip series products, plus an AI hardware-software co-innovation project.
   ‍
2. Large-model workloads and cluster software
   The filing highlights work on distributed inference and cluster operations for MoE-style models, and it cites operational targets such as stable runs for thousand-card clusters and 24-hour stability for ten-thousand-card clusters.
   ‍
3. China deployment channels beyond a single vertical
   Enflame points to participation in “East Data West Computing” hub projects and deeper cooperation with domestic telecom operators, alongside broader ecosystem building with downstream partners.
   ‍
4. Engineering scale as a moat
   The filing reports high R&D intensity and a large R&D workforce, with cumulative R&D spend disclosed for 2022–2024 and R&D staff as the majority of headcount.
   ‍

Challenges ahead

‍

• Customer concentration and related-party exposure
  The filing flags extreme concentration, with Tencent-related sales (direct plus AVAP model) rising to 71.84% of revenue in Jan–Sep 2025, and it expects a high Tencent share to persist for a period.
  ‍
• Losses and cash flow pressure during the ramp
  The filing shows continued losses, negative operating cash flow, and the risk that revenue scale does not cover fixed R&D needs in the near term.
  ‍
• Fast iteration cycles and supply-chain dependency
  The filing lists risks tied to rapid chip iteration, advanced manufacturing and packaging requirements, and supplier concentration.