Enflame’s $908m IPO draws 6,109x demand despite 84% Tencent dependence

Shanghai Enflame Technology Co., Ltd. has priced its Shanghai STAR Market initial public offering at RMB142.18 per share, selling approximately 43.04 million new shares to generate gross proceeds of roughly RMB6.12 billion, equivalent to about $908 million at recent exchange rates. The issue represents approximately 10% of Enflame’s enlarged share capital and implies an offer-price equity valuation of about RMB61.2 billion, or close to $9 billion. Investor demand has been extraordinary: the online portion was subscribed thousands of times over, forcing shares to be reallocated from the institutional tranche and leaving online applicants with a final winning rate of only about 0.0246%. Yet the same prospectus behind the frenzy shows that Enflame generated just RMB990 million of revenue in 2025, lost roughly RMB1.2 billion and derived 83.8% of revenue from Tencent-related business.

The transaction gives Enflame ticker 688801 on the STAR Market and provides capital for fifth-generation and sixth-generation artificial intelligence chips, hardware-software co-development and commercialization. Tencent Holdings Limited is not merely a financial backer: it owns roughly one-fifth of Enflame and is overwhelmingly its largest customer. A Tencent entity also received shares in the strategic placement, further deepening a relationship that has helped Enflame establish commercial scale while simultaneously creating its clearest concentration risk.

Why did Enflame’s online IPO demand reach more than 6,000 times available supply?

China’s artificial intelligence chip market sits at the intersection of extraordinary computing demand and strategic pressure to reduce dependence on foreign suppliers. U.S. restrictions on advanced semiconductor exports have increased the urgency for Chinese cloud companies, internet platforms and state-linked customers to develop domestic alternatives to NVIDIA accelerators. Enflame is one of the relatively small group of Chinese companies with products already deployed in data-centre AI workloads, giving public investors exposure to a market that remains capacity constrained.

Investor enthusiasm has also been reinforced by the performance of other Chinese AI chip listings. Moore Threads, MetaX Integrated Circuits and Biren Technology have attracted enormous interest as investors seek domestic semiconductor champions with exposure to generative AI. That creates a scarcity premium because public investors have only a limited number of listed companies through which they can express a view on China’s accelerator market.

Enflame’s subscription figures demonstrate how intense that demand has become. More than 7 million online accounts submitted valid applications for approximately 42.07 billion shares, while the final online allocation after the clawback was only about 10.33 million shares. The resulting winning rate was approximately 0.0246%, meaning fewer than three shares out of every 10,000 applied for were effectively available through the final allocation mechanism.

That demand is a market signal, not evidence that the valuation is inexpensive. IPO subscription ratios measure competition for a limited amount of stock and can become particularly extreme when speculative enthusiasm and strategic narratives collide. Enflame still needs to prove that its technology can generate enough revenue and profit to justify an equity value exceeding RMB60 billion.

How aggressive is Enflame’s roughly 61.8-times sales valuation?

The offering values Enflame at about 61.8 times 2025 sales, based on the RMB61.2 billion post-offer market capitalization and approximately RMB990 million of 2025 revenue. That multiple is unusually high for a hardware company, particularly one that has not yet generated annual profit, although it is below the even richer sales multiples recently attached to some Chinese AI-chip peers.

A price-to-earnings ratio is not meaningful because Enflame lost approximately RMB1.2 billion in 2025. Revenue increased from roughly RMB301 million in 2023 to RMB990 million in 2025, showing rapid commercial expansion, but accumulated losses remained substantial. First-quarter 2026 revenue reached approximately RMB287 million and increased more than 1,400% year over year, while the company still recorded a net loss of approximately RMB444 million.

This creates a valuation structure heavily dependent on future scale. Investors are not paying for current earnings power but for the possibility that Enflame becomes one of China’s leading domestic alternatives in a market where spending on AI computing is expanding rapidly and advanced foreign chips remain constrained.

The upside scenario requires revenue growth to continue for several years while gross margins improve and research expenditure becomes a smaller proportion of sales. The downside scenario is that increased competition among Chinese accelerator suppliers forces prices lower before Enflame reaches sufficient scale to absorb its development costs.

The distinction matters because semiconductor startups can experience strong revenue growth and still destroy capital if each generation requires enormous design expenditure without producing durable pricing power. Enflame’s IPO gives it additional resources to compete, but public shareholders are entering at a valuation that already assumes substantial future success.

Why is Tencent responsible for 83.8% of Enflame’s revenue and why does that matter?

Tencent’s role has been both a commercial accelerator and a concentration risk. The technology group and related arrangements accounted for approximately 83.8% of Enflame’s 2025 revenue, up from roughly 38% the previous year. Tencent also owns approximately 20% of the company, making the relationship unusually deep across both ownership and customer economics.

For a young semiconductor company, such a customer can be enormously valuable. Tencent operates large cloud and internet platforms capable of deploying AI accelerators in meaningful volumes, giving Enflame access to workloads that help refine hardware, software and system performance. Management has indicated that Tencent demand has exceeded Enflame’s ability to supply accelerator cards, suggesting concentration currently reflects scarcity and customer pull rather than a struggle to find demand.

The risk emerges if that relationship changes. Tencent can alter procurement priorities, develop more internal technology, increase purchases from another domestic supplier or regain access to more competitive foreign hardware. Even a modest shift in Tencent ordering could materially affect Enflame while more than four-fifths of revenue remains connected to the same corporate ecosystem.

Customer concentration also affects bargaining power. A supplier depending on one customer for the overwhelming majority of sales may have less flexibility around pricing, roadmap priorities and delivery terms than a supplier serving several hyperscale buyers.

Enflame says it is seeking additional internet and commercial customers, with trial orders and negotiations underway. Diversification will become one of the most important post-IPO metrics because rapid revenue growth carries much greater valuation quality when it comes from several independent buyers rather than one strategically aligned shareholder-customer.

Can Enflame realistically challenge NVIDIA in China’s AI accelerator market?

The market-share gap remains enormous. Industry estimates cited around the IPO put NVIDIA at roughly 55% of Chinese AI accelerator shipments in 2025, while Enflame was estimated at only about 1.7%. That still places Enflame among the more important domestic suppliers, but it illustrates how far the company remains from matching NVIDIA’s installed base, software ecosystem or production scale.

Enflame does not need to replace NVIDIA nationally for the IPO economics to work. China’s AI infrastructure market is large enough that a domestic supplier capturing a mid-single-digit share could potentially grow revenue several times from today’s base, particularly while demand outstrips domestic supply.

The harder competition may come from other Chinese companies. Huawei’s Ascend platform carries substantial strategic importance, while Moore Threads, MetaX, Biren and other suppliers are raising capital and developing successive chip generations. Domestic substitution therefore creates demand while simultaneously encouraging more competitors to enter the same opportunity.

Software remains another barrier. AI accelerators are valuable only when developers can efficiently run frameworks, models and applications on them. NVIDIA’s CUDA ecosystem has accumulated years of libraries, optimization and developer familiarity. Chinese hardware suppliers must invest continuously in compilers, software stacks and model compatibility if customers are to treat their chips as more than emergency substitutes created by export controls.

Enflame’s planned use of proceeds reflects that reality. The company is allocating capital not only toward fifth-generation and sixth-generation processors but also toward hardware-software integration and platform development. The commercial race is therefore as much about software maturity as transistor performance.

Why could Enflame’s heavy inference exposure create both opportunity and margin pressure?

More than 80% of Enflame’s revenue is associated with inference-oriented products, according to reporting based on its prospectus. Inference is the stage where trained models process new inputs and produce results, and the market could eventually become much larger than training as AI features spread through search, advertising, productivity software, commerce and consumer applications.

That creates a potentially enormous volume opportunity. Every enterprise or consumer application using generative AI needs inference capacity each time a user requests a response, creating recurring computational demand rather than a one-time training event.

Inference buyers can also be much more price sensitive than frontier model laboratories. A company serving millions of user requests needs low cost per token, low power consumption and predictable latency, meaning suppliers may face stronger pricing competition as hardware alternatives multiply.

Enflame therefore needs more than shipment growth. It needs improving economics per device and enough software efficiency that customers view its total cost of ownership as competitive against Huawei, NVIDIA and other domestic suppliers.

The IPO valuation implicitly assumes that scale eventually converts into profitability. Inference could help create that scale, but it can also expose the company to a more competitive market where customers aggressively compare price and power efficiency.

What should investors watch once Enflame begins trading on the STAR Market?

The first market test will be whether Enflame repeats the extraordinary first-day performance seen in some Chinese AI chip IPOs or whether the RMB142.18 issue price already captures too much optimism. A huge subscription ratio creates scarcity at allocation but does not guarantee sustained post-listing returns once more investors can buy and sell freely.

The second test is customer diversification. Reducing Tencent-related revenue from 83.8% toward a materially lower level while continuing to grow total sales would improve the quality of the investment story considerably.

Third, investors should watch losses and operating cash flow. Enflame has narrowed losses, but a company valued at more than 60 times historical revenue will eventually need a visible path toward profitability rather than indefinite reliance on capital markets.

Fourth, the fifth-generation and sixth-generation chip programmes must progress on schedule. Semiconductor roadmaps lose competitive value quickly when development delays allow rival products to reach customers first.

The IPO gives Enflame roughly RMB6.1 billion of gross new capital and intense public-market attention. What it does not provide is immunity from semiconductor economics. The company still needs competitive silicon, a strong software stack, broader customers and improving margins.

The 6,109-times demand figure tells investors that Chinese AI chips are scarce in public portfolios. The 83.8% Tencent concentration tells them why scarcity alone is not enough.