Who else is willing to pay for Enflame Technology besides Tencent?

On September 11, 2026, Enflame Technology was listed on the Sci-Tech Innovation Board (STAR Market), with an opening market value exceeding 170 billion yuan.

How can a company founded eight and a half years ago that posted revenue of only 990 million yuan and a net loss of over 1 billion yuan in 2025 justify such a high valuation?

The answer is written in its prospectus: Tencent is both its largest shareholder and its top customer, contributing 83.79% of its revenue in 2025.

The story of Enflame is essentially a "Tencent story".

But what the capital market pays for is never the past, but the future.

Then, apart from Tencent, who else is willing to pay for Enflame?

I. Can Tencent's "model room" be replicated?

The partnership between Enflame and Tencent can be described as a long-standing close bond since Enflame's early days.

Tencent has been investing in Enflame since its establishment, with total investment exceeding 2 billion yuan. After that, Tencent participated in follow-on financing in every round except for one round that was solely led by the China Integrated Circuit Industry Investment Fund. Starting from its second-generation products, Enflame's chips have been widely deployed in Tencent's scenarios including enterprise-level online conferencing, social applications, gaming AI, and search ad recommendation.

In this regard, Zhao Lidong, founder of Enflame Technology, calls this "setting a successful benchmark".

However, this "model room" has an inherent flaw: it only has one single room.

According to the prospectus, Enflame's direct sales to Tencent and sales under the AVAP model accounted for 33.34% of its total revenue in 2023, 37.77% in 2024, and surged sharply to 83.79% in 2025.

This curve does not indicate successful customer expansion, but that the growth rate of Tencent's procurement far outpaces the speed of developing other customers. More notably, after excluding the 830 million yuan of revenue contributed by Tencent, Enflame's actual effective revenue in 2025 was only about 160 million yuan — this reflects its real market-oriented operation capability.

Meanwhile, the proportion of revenue from the top five customers of the company has reached as high as 96.5%, 92.6% and 96.89% respectively for three consecutive years. Its customer structure is extremely concentrated, and the top four customers other than Tencent change frequently every year, with no stable repurchase relationship established.

Zhao Lidong disclosed during the roadshow that the fourth-generation products have passed hardware and model matching verification with three potential internet customers, are currently in gray-scale testing, and are expected to be delivered on a small scale in 2026 and in large batches in 2027. This is the key window for Enflame to prove that its "model room is replicable".

However, between "passing matching verification" and "batch procurement", there are multiple challenges including software adaptation, cluster stability, and delivery capacity. No one will know whether Enflame can overcome these hurdles before 2027.

II. Who else can become Enflame's customer?

Veteran technology industry observer Zhang Dongwei believes that from a fundamental perspective, Alibaba, ByteDance, Huawei and Baidu will not become large-scale customers of Enflame.

Alibaba's T-Head self-developed PPU "Zhenwu 810E" has been deployed in 10,000-chip clusters on Alibaba Cloud; ByteDance's self-developed inference chip SeedChip is scheduled for mass production in 2026, all of which will be used for its own scenarios and will not be sold externally; Baidu's Kunlunxin has operated independently, and its P800 chip undertakes most of the inference tasks within Baidu and is sold to hundreds of external customers; Huawei's Ascend shipped about 812,000 chips in 2025, accounting for nearly half of the total shipments of domestic manufacturers.

These four largest purchasers of AI computing power are not only not potential customers of Enflame, but also its most direct competitors in the domestic market.

Enflame's potential customers are only divided into two categories:

The first category refers to internet companies without self-developed chip capabilities, such as Meituan, Kuaishou and NetEase; the second category refers to intelligent computing center projects in government and enterprise industries such as operators, finance and energy.

However, the total computing power demand of these customers is not at the same magnitude as that of Tencent, Alibaba and ByteDance.

Bernstein predicts that by 2028, the AI capital expenditure of ByteDance, Alibaba and Tencent will account for nearly 50% of China's total AI capital expenditure. The remaining 50% of the market will still be divided by Huawei, Kunlunxin, Cambricon and other players.

The market share Enflame can capture is rather limited.

More critically, the DSA architecture (non-GPGPU with CUDA compatibility) that Enflame adheres to further raises the migration threshold for non-Tencent customers.

For mid-sized customers that do not have self-developed chips and strong R&D teams, re-adapting the software stack and operator library for Enflame will lead to extremely high R&D costs, which forms the second invisible barrier in addition to the market size constraint.

A noteworthy detail is that the total shipment of AI acceleration cards in China reached about 4 million units in 2025, and Enflame did not rank among the top domestic manufacturers in terms of shipment volume counted by IDC.

Huawei shipped 812,000 units, T-Head shipped 256,000 units, and both Kunlunxin and Cambricon shipped 118,000 units — Enflame is not on the list.

But it needs to be clarified that this does not mean Enflame has no shipments at all.

Enflame's revenue from AI acceleration cards and modules reached 856 million yuan in 2025, which was driven by the increased sales of relatively high-priced full boards/modules. The actual number of shipped chips still has a magnitude gap compared with manufacturers like Huawei.

Taking the high-value, small-batch route may help increase revenue, but it is difficult to build a large-scale ecosystem and solid customer base.

III. Is it a supply chain manufacturing enterprise or a chip design company?

If we compare Enflame to "Foxconn in the AI era", although this analogy is logically reasonable, it cannot be directly applied in a simple way.

Domestic computing power chip companies generally face the dilemma of "lacking self-owned scenarios and relying on a small number of large customers": the top five customers of Cambricon account for 88.66% of its revenue, and the top five customers of Moore Threads even accounted for 98.29% of its revenue in the first half of 2025.

—— This is a typical feature of supply chain manufacturing.

But chip design is not Foxconn-style assembly manufacturing.

In the short term, Enflame has technological barriers and independent control over its architecture. Even if the gross margin of its AI acceleration cards dropped from 40.78% to 32.71% in 2025, it is still far higher than that of traditional manufacturing industries.

However, the essence of the gross margin decline is the loss of pricing power.

Tencent is not only the largest shareholder holding more than 20% of the shares, but also the top customer contributing more than 83% of the revenue. This dual identity determines that Enflame has almost no bargaining chips in pricing negotiations.

As Tencent's procurement scale expands, unit price concessions and declining gross margins are inevitable trends.

This is not an accidental result of cost fluctuations, but a manifestation of the structural deprivation of bargaining power under the reliance on large customers.

In the long run, Enflame's product iteration pace, pricing space, and growth boundary are almost entirely defined by Tencent alone. Without an ecosystem, without general-purpose customers, and having its pace locked by a single client — its final profit margin, pricing power and growth space will all converge to the level of the manufacturing industry.

This is a "high-tech barrier OEM model", positioned between NVIDIA and Foxconn.

However, Enflame's valuation is priced according to the logic of NVIDIA.

Its issuance valuation is about 61.2 billion yuan, corresponding to 990 million yuan of revenue in 2025, with a price-to-sales ratio of about 62 times, and the company is still operating at a loss. This valuation does not price its current performance, but the expectation that "it can successfully replicate its customer base outside Tencent".

If its promise of large-scale delivery in 2027 is fulfilled, the valuation will be supported; if it fails to expand other customers, the 62x price-to-sales ratio will be unsustainable.

There is also an easily overlooked reverse risk that needs to be supplemented: Tencent itself has its own self-developed chip layout.

Chips including Canghai, Zixiao and Xuanling developed by Tencent's Penglai Laboratory are already under development. Enflame's role in Tencent's system is essentially an external alternative backup, not an irreplaceable one.

This means that even if Enflame retains Tencent as its customer, there is still an implicit risk of being gradually replaced by Tencent's self-developed chips in the long run.

Dong's Insight

Enflame has no problem with its short-term survival.

Tencent's continuous procurement provides cash flow and product verification scenarios. Its fourth-generation integrated training and inference product has returned from the chip fabrication factory, and large-scale delivery on the training side will be its next growth point.

But its long-term value depends on a fundamental question: Apart from Tencent, who else is willing to pay for Enflame?

The answer to this question does not lie in Enflame's technical route, but in the structure of China's AI computing power market.

When Huawei, Alibaba, Baidu and ByteDance have all built their own high walls of self-developed chips, and when the total demand of remaining customers is not enough to support the scale effect of an independent chip company, the most likely destination for Enflame is to become a dedicated supplier deeply bound to Tencent's AI infrastructure map but with limited bargaining power.

This is not Foxconn-style manufacturing, but it is far from being a NVIDIA-style platform company.

The answer to this question will determine whether Enflame is a chip company with a market value of 170 billion yuan, or a "Tencent concept stock" with a market value of 170 billion yuan.

This article is from the WeChat public account "Zhang Dongwei", author: Zhang Dongwei, and is published with authorization from 36Kr.