--- title: "Enflame's market debut puts a self-built AI accelerator maker under the spotlight" date: 2026-10-23 category: Chips & Compute site: NeuroAI canonical: https://neuroai.site/a/na-chips-enflame-ai-accelerator-ipo language: en --- # Enflame's market debut puts a self-built AI accelerator maker under the spotlight > Enflame listed on the SSE STAR Market in September 2026 after growing revenue to RMB 990 million in 2025, though it is still posting a net loss. A Shanghai trading floor. The bell rang once, and a row of engineers standing at the back of the room finally exhaled. For a chip startup that had spent eight years building its own path, the listing was the easy part. The hard part is everything after the bell. ## A home-grown accelerator, not a clone Enflame (燧原科技) builds cloud AI accelerators — the silicon that does the heavy lifting behind a large model (大模型) and the broader build-out of compute (算力) in China. The company listed on the SSE STAR Market on 11 September 2026 under the ticker 688801, closing a journey that began in 2018. What sets Enflame apart from the GPGPU crowd is its architectural bet. Rather than follow Nvidia's CUDA-compatible road, Enflame chose a domain-specific (DSA) design. Its products span three layers: - The 邃思 (Suisi) series AI chips and accelerator cards. - 云燧 (Yunsui) intelligent computing clusters for data centers. - 驭算 (TopsRider), a self-developed software and programming platform that replaces CUDA with its own compiler, operator library, and toolchain. At the hardware level, the company points to its GCU-CARE compute unit and GCU-LARE high-speed chip-to-chip interconnect as the pieces meant to rival Tensor Core and NVLink — on paper, with higher efficiency for inference-heavy large-model workloads. ## The numbers behind the debut Enflame's prospectus tells a familiar story for China's frontier chip firms: fast revenue growth, persistent losses, and heavy R&D. - Revenue reached RMB 990 million (≈ US$139m / HK$1.09bn) in 2025, up from RMB 722 million (≈ US$102m / HK$794m) in 2024 and RMB 301 million (≈ US$42m / HK$331m) in 2023. - The company remained unprofitable, posting a net loss of roughly RMB 1.2 billion (≈ US$169m / HK$1.32bn) in 2025. - R&D spending in 2024 was about RMB 1.31 billion (≈ US$185m / HK$1.44bn) — larger than that year's revenue. - By the end of 2025, Enflame held 313 domestic invention patents and had joined the drafting of 58 national or industry standards for AI chips and intelligent computing systems. Its prospectus set out around RMB 6 billion (≈ US$845m / HK$6.6bn) in planned fundraising, earmarked for fifth- and sixth-generation chip series and advanced software–hardware co-innovation. ## Who owns it, and who buys it Tencent (腾讯) and its affiliates are the largest external shareholder, holding about 20.26% of the company. The relationship is both a strength and a concentration risk: Enflame has supplied AI compute to internet majors since 2019, and the prospectus acknowledges that sales to Tencent-linked customers will likely stay high for some time. On the demand side, Enflame shipped about 66,000 AI accelerator cards and modules in 2025, giving it roughly 1.7% of China's AI accelerator card market. For context, the prospectus cites IDC figures estimating China's total AI accelerator card shipments at around 4 million units in 2025, with Nvidia accounting for roughly 2.2 million — about 55%. ## The crowded domestic field Enflame sits in a dense pack. The prospectus names Huawei HiSilicon and Cambricon (寒武纪) as earlier-moving DSA peers, and Moore Threads, MetaX, Iluvatar, and Biren as GPGPU peers. None yet matches Nvidia's scale in China. The practical question is which firms can move from pilot deployments to repeat, large-volume orders from cloud providers and state-backed compute (算力) centers. ## Why a loss-making firm still listed The logic is strategic, not accounting. China's cloud AI-chip segment is still early, and the prospectus argues that as large-model inference demand rises, non-GPGPU architectures with better efficiency and lower power draw gain an edge. A public listing funds the multi-year chip cadence that fabrication, packaging, and software Ecosystem (生态) all require. According to Li An, chief scientist at BrainNet (脑机网), the real test for domestic accelerator firms is not the first product but whether they can keep a two-year design rhythm while their software stack matures alongside customers. ## Honest limitations This article relies mainly on Enflame's own prospectus disclosures and Chinese financial-media summaries of them; we did not independently audit the figures. The market-share and shipment numbers come from third-party consultancy data cited in the filing, not from a regulator. We have not assessed Enflame's actual chip performance with independent benchmarks, and we have not covered its supply-chain exposure to advanced fabrication or export controls. The Tencent concentration and the path to profitability are flagged by the company itself but not resolved here. ## What readers can do now - Track Enflame's post-listing quarterly reports for the ratio of R&D spend to revenue and for any decline in net loss — that, not the listing day, signals traction. - Watch whether its 驭算 (TopsRider) software gains outside-Tencent customers; a single large buyer is a structural risk worth monitoring. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-chips-enflame-ai-accelerator-ipo Free to quote with attribution and a link to the original.