Enflame Technology is a Chinese cloud AI computing company that develops accelerator chips, cards, modules, servers, clusters, and a unified software platform for model training and inference. It is often grouped with MetaX, Biren Technology, and Moore Threads as one of China’s “four AI GPU unicorns,” although Enflame’s architecture is more precisely described as a domain-specific AI accelerator rather than a conventional general-purpose GPU.
This guide examines Enflame’s technology stack, its move from inference-led commercialization toward integrated training and inference systems, Tencent’s importance, and how eligible users can take long or short positions through the ENFLAME contract without owning Enflame shares.
Founded in Shanghai in 2018,
Enflame Technology builds computing infrastructure for cloud data centers and large AI workloads. The company operates under a fabless model: it designs chips and related systems while relying on external foundries, packaging providers, memory suppliers, and other partners for manufacturing. Its stated goal is to provide an independently developed computing foundation for artificial intelligence.
Enflame’s stack connects proprietary processors with CloudBlazer accelerator products, computing systems and clusters, and the TopsRider programming platform. Customers evaluate more than peak chip performance: model compatibility, distributed communication, debugging, server stability, and predictable multi-device performance all affect adoption.
The company’s GCU-CARA design belongs to the domain-specific architecture, or DSA, camp of cloud AI chips. A DSA retains programmability across AI workloads while optimizing for machine-learning computation. Calling Enflame a “GPU company” is useful shorthand, but it obscures differences from a general-purpose GPU platform.
Enflame is therefore best understood as an AI infrastructure platform. Its value depends on whether customers can move real workloads onto the stack, scale them, and adopt later chip generations without costly migration.
Earlier CloudBlazer products separated training and inference, while the third-generation S60 focused on inference. The fourth-generation L600 integrates both uses for generative and multimodal models. Inference remains the more immediate market, but large models increasingly need high-bandwidth interconnects and system capabilities once associated mainly with training.
TopsRider connects Enflame hardware with AI frameworks through programming, operator, compilation, communication, debugging, and performance tools. A chip can meet headline specifications yet fail commercially if customers cannot migrate models efficiently or achieve stable utilization.
Enflame also packages its accelerators into servers, intelligent computing systems, and large clusters. At the 2026 World Artificial Intelligence Conference, it introduced the CloudBlazer ESL64-O supernode with ZTE. The system uses an orthogonal, backplane-free design intended to remove internal cabling, reduce interconnect complexity, and support denser deployment. Enflame also displayed the ESL64-C cable-tray design, giving customers a choice between a newer orthogonal architecture and a more established implementation path.
The event also highlighted a CoPoS, or Chip-on-Panel-on-Substrate, advanced-packaging sample. Memory, interconnect density, power, thermals, and manufacturing yield increasingly determine usable performance. The supernode and packaging work signals competition at rack and cluster level, not only on individual cards.
Enflame earns revenue mainly from accelerator products, computing systems, and related services sold to internet companies, operators, and computing projects. TopsRider primarily supports hardware adoption rather than generating separate licensing revenue.
Revenue increased from about RMB 301 million in 2023 to RMB 722 million in 2024 and RMB 990 million in 2025, according to the company’s
Shanghai Stock Exchange filing. That growth shows increasing commercialization, but Enflame remained deeply loss-making. Its net loss was approximately RMB 1.66 billion in 2023, RMB 1.51 billion in 2024, and RMB 1.16 billion in 2025.
Research and development expenditure exceeded annual revenue in each of those years, reaching about 115% of revenue in 2025. This is the central financial tension in the story. Enflame needs repeated chip generations, software development, model adaptation, supply commitments, and customer validation before sales can scale. Faster revenue growth does not by itself establish a path to positive cash flow.
Enflame’s sales mix has leaned heavily toward inference: training and integrated products represented only about 1% of accelerator-product revenue in 2025. L600 and the supernode strategy may broaden the mix, but validation does not guarantee high-volume orders.
Chinese cloud providers, internet platforms, operators, and public projects are seeking domestic AI capacity. Supply uncertainty strengthens the case for local accelerators, but vendors must still compete on performance, software maturity, efficiency, cost, and delivery.
Tencent is both a major shareholder and Enflame’s most important commercial partner. The companies began working together in 2019 and progressed from small-scale validation to deployment across multiple AI workloads. This relationship gives Enflame access to demanding production scenarios and a reference customer capable of testing products at scale.
It also creates concentration risk. Tencent-linked sales represented a substantial majority of 2025 revenue, so procurement or supplier changes could materially affect Enflame. Progress with other internet customers, operators, and computing centers is critical.
AI systems increasingly make many accelerators operate as one computing domain. ESL64-O and cluster experience may increase value per deployment, but stability, networking, cooling, orchestration, supply, and service all become part of the product.
Enflame’s application for a Shanghai Stock Exchange STAR Market listing was accepted in January 2026, and its registration took effect in July 2026. The
official SSE project page shows planned fundraising of RMB 6 billion. Registration is a major milestone, but it is not the same as a completed offering, assigned stock code, final offer price, or first trading day.
The proposed raise reflects the cost of competing in AI infrastructure. Investors must judge whether it creates scalable products and diversified demand or extends high research costs and negative cash flow.
Useful operating evidence includes volume deployment of L600 products, repeat orders from non-Tencent customers, stable supernode operation, higher software compatibility, improving gross margins, and a declining ratio of research spending to revenue. Announcements and benchmark results can attract attention, but durable value depends on customers moving from testing to recurring production use.
ENFLAME is a stock futures contract, not a share. It provides no ownership rights, IPO allocation, or guaranteed conversion. Its market price and the eventual offering’s code, share count, or price may differ from pre-listing assumptions.
For current availability and product specifications, consult the
ENFLAMESTOCK_USDT contract page. Do not infer contract terms from unofficial tables or third-party estimates.
Traders expecting the contract price to rise may open a long position:
Step 1: Sign in to MEXC and complete required checks.
Step 2: Transfer supported collateral to the futures account.
Step 4: Select long and choose an order type.
Step 5: Enter the size, check the details, and submit.
Traders expecting the contract price to fall may open a short position:
Step 1: Sign in, confirm eligibility, and complete verification.
Step 2: Add supported collateral to the futures account.
Step 4: Select short and choose an order type.
Step 5: Set the size, check the details, and submit.
No equity ownership: The contract does not represent a share and cannot be assumed to convert into stock.
Listing risk: Final offering terms may differ from assumptions in the futures price.
Commercial concentration: Enflame depends heavily on Tencent-linked demand, while diversification remains a work in progress.
Technology risk: Delays, compatibility problems, poor yield, or slow validation could limit adoption.
Financial risk: Enflame remains unprofitable and research-intensive.
Supply-chain risk: Foundries, packaging, memory, controls, and supplier terms can affect delivery.
Trading risk: Limited liquidity, wide spreads, and rapid repricing can cause losses.
What is Enflame Technology?
Enflame develops cloud AI accelerators, systems, clusters, and the TopsRider software platform.
Is Enflame a GPU company?
It is grouped with AI GPU peers, but its filings classify GCU-CARA as a domain-specific architecture.
Is Enflame publicly listed?
Its STAR Market registration is effective, but that does not establish a completed listing or final price.
Does the ENFLAME futures contract represent company shares?
No. It is a derivative and provides no ownership, voting rights, dividends, or guaranteed conversion into Enflame stock.
Where can I find current contract specifications?
What is TopsRider?
TopsRider connects Enflame hardware with model frameworks, operators, communication tools, and deployment workflows.
What is the CloudBlazer ESL64-O?
It is a 64-accelerator supernode developed with ZTE using an orthogonal, backplane-free design.
Why is Tencent important to Enflame?
Tencent is both a major shareholder and the dominant customer, giving Enflame production-scale validation while also creating material customer-concentration risk.
Is Enflame profitable?
No. In 2025, revenue was about RMB 990 million and net loss roughly RMB 1.16 billion.
Enflame must scale integrated products, diversify beyond Tencent, and convert research spending into repeatable revenue.
Its futures remain derivatives; concentration, losses, supply constraints, and volatility make careful sizing essential.
Risk Disclaimer: This article is for educational purposes only and does not constitute financial advice. Stock futures can involve limited liquidity, wide bid-ask spreads, uncertain price discovery, and rapid losses. Product availability may vary by jurisdiction. Review the
ENFLAMESTOCK_USDT contract page and assess your risk tolerance before trading.