Without Tencent, what is left of Enflame?

If Tencent stops purchasing from Enflame tomorrow, how much is Enflame still worth?

This is the question that runs through Enflame's IPO prospectus, though no one explicitly states it.

In 2025, 74.9% of Enflame's 990 million yuan in revenue came from Tencent.

According to another statistical method, this figure even exceeded 80%.

In other words, for this upcoming AI chip star company on the Science and Technology Innovation Board, nearly 8 yuan out of every 10 yuan earned comes from the same customer.

On the 15th, news came from the Shanghai Stock Exchange that Enflame Technology's IPO on the Science and Technology Innovation Board was officially approved by the listing committee. This means it has successfully completed the process, and the "Four Little Dragons" of domestic GPUs have completed their final encirclement in the capital market.

In December last year, Moore Threads and Muxi Semiconductor successively entered the Science and Technology Innovation Board and have now grown into giants with a market value close to 300 billion. In January this year, Biren Technology also successfully listed on the Hong Kong stock market.

Although Enflame was the last to arrive, its successful passing of the review sends an extremely clear signal of migration in China's AI chip industry to the outside world: The computing power industry has completely bid farewell to the financing - driven "PPT era" and has fully entered an extremely cruel order - delivery cycle.

1. Why was Enflame the last?

Enflame may be the least - like startup among the Four Little Dragons.

Moore Threads has targeted the general - purpose GPU route of NVIDIA from the beginning, telling the story of China's GeForce. Biren focuses on high - performance training chips, following the route of hardcore parameters. Muxi focuses on the training card direction, emphasizing technological catch - up.

The rhythm of the three companies is to rush into the capital market first, inflate the valuation first, and tell a good story first.

Enflame has taken a different path. Since its establishment, it has been aiming at one thing: delivering to large customers. It doesn't tell stories first; it does business first.

The figures speak for themselves.

The revenue was 300 million yuan in 2023, 720 million yuan in 2024, and 990 million yuan in 2025. The three - year compound growth rate was 81%. An even more amazing explosion occurred in the first quarter of 2026, with a single - quarter revenue of up to 287 million yuan, a year - on - year surge of 1,474.85%. The revenue in the first half of 2026 is expected to directly exceed 1.06 billion yuan, surpassing the total of the previous entire year in just half a year.

This extremely abnormal and steep growth rate is not due to the natural spread of the industry but the concentrated and targeted release of computing power orders from super large customers.

Enflame has reaped the benefits of China's computing power construction cycle. It has exchanged its delivery capabilities for orders, orders for revenue, and revenue for the confidence to go public.

The fact that it was the last to go public actually illustrates a truth: In the AI chip industry, it's not about who goes public first now, but who has real revenue in hand.

2. The industrial truth behind 83.79%

There is a set of very prominent figures in the prospectus. In 2025, the proportion of Enflame's sales revenue from Tencent was 74.9%. According to the caliber of the second - round inquiry response (including the AVAP model), this figure was 83.79%.

Looking at the structure intuitively, there is an overly high customer concentration, related - party transactions, and the company would be in trouble if Tencent reduces its purchases.

That's right. These are all facts.

But from an industrial perspective, a more worthy question to ask is: Under the technological monopoly of NVIDIA and the competition from domestic counterparts, why is Tencent so determined to buy so many products from Enflame?

You know, for large companies like Tencent and ByteDance, their previous procurement logic was very simple: buy NVIDIA products if possible and never use domestic ones; choose the first - tier products and never compromise on the second - tier ones.

But now Tencent has started to purchase from Enflame on a large scale. In 2023, Tencent's procurement amount from Enflame was only 100 million yuan; in 2024, it climbed to 270 million yuan; in 2025, it directly switched to large - scale direct sales, pulling the order amount up to 768 million yuan at once. In three years, it has almost increased eight - fold.

Behind this is not that Tencent has suddenly become a philanthropist. It's that Tencent's computing power demand has become so large that it can no longer rely solely on NVIDIA.

Yuanbao, Hunyuan, WeChat AI, Enterprise WeChat, Tencent Meeting, CodeBuddy, WorkBuddy, all these products require a large amount of GPU computing power. Moreover, the demand is continuous and increasing.

Tencent's Q1 2026 financial report shows that the new AI products dragged down the operating profit by about 8.8 billion yuan in a single quarter; the capital expenditure reached 31.9 billion yuan, a significant year - on - year increase, and the management expects the investment in new AI businesses to double this year.

Under this demand pressure, what Tencent needs is no longer a more powerful card but a controllable, stable, and unconstrained computing power supply system.

Especially, more than 80% of Enflame Technology's revenue from accelerator cards and modules comes from inference products, which are most urgently needed for the implementation of large models. What Enflame sells is not just silicon wafers but the sense of security that Tencent's large models urgently need.

3. The dual identity of shareholder and customer

Many people are used to interpreting the 83.79% as a one - way blood - transfusing dependence. But looking from the other side, it's Tencent that is actively nurturing Enflame.

Why? Because Tencent clearly knows one thing: if domestic AI chip companies don't have enough orders, they will never grow up. Without mass - production scale, there will be no process iteration, and without iteration, they will never catch up with NVIDIA.

Why is NVIDIA so strong? It's not because it was strong from the very beginning. It's because Microsoft, Meta, and Amazon have continuously given orders, provided scenarios, and given feedback. It was nurtured by large customers one after another.

What Tencent is doing to Enflame today is essentially the same thing.

This is very similar to when Apple poured all its resources into supporting TSMC to fight against industry monopoly. Their relationship has long gone beyond the ordinary sales contract. Instead, with definite orders, it helps the supplier polish the underlying process to the extreme and then becomes an irreplaceable part of its supply chain.

Tencent holds a total of 20.26% of Enflame's shares through technology - related parties and their affiliates, firmly sitting in the position of the largest shareholder and also being the largest customer at the same time. The dual identity of shareholder and customer is deeply bound. In the eyes of traditional PE, this is a risk of related - party transactions, but in the industrial logic, this is supply - chain cultivation.

Enflame was quite straightforward in its response to the first - round inquiry: Tencent lacks the motivation and practical conditions to replace the company's products on a large scale through other suppliers at this stage.

Simply put, Tencent has deployed Enflame's chips in a large number of real - world business scenarios, possibly including the inference cluster of the Hunyuan large model, the daily conversations of Yuanbao, and Tencent Cloud's AI computing power services for enterprise customers.

To replace them is not just a matter of replacing a card. The entire software stack, drivers, and scheduling systems need to be rebuilt. The cost is extremely high.

This also explains why although there are a small number of domestic competitors entering Tencent's supply chain, and Enflame itself admits this, Tencent still reserves the main procurement position for it. What Tencent needs is not to buy from multiple suppliers to disperse risks but to let one company run through the entire process and then replicate it on a large scale.

The 74.9% sales concentration has, at this moment, turned from a weakness into an ecological moat for Enflame.

4. The ecological division in the third layer of AI chips

If we look at today's AI chip industry in isolation, the pattern is already very clear.

Layer 1: NVIDIA. The rule - maker. It controls the global ecosystem, CUDA, and production capacity.

Layer 2: Huawei Ascend. A national - level player. It has its own ecosystem, is bound to operators and central state - owned enterprises, and does not fully follow the market - oriented route.

Layer 3: Enflame, Moore Threads, Muxi, and Biren. Commercial players that rely on market orders to survive.

In the past, people thought that these companies in the third layer were in a competitive relationship, as they are all making domestic GPUs and competing for the same group of customers.

But now it's more and more like another pattern, where each is bound to a different ecosystem.

Moore Threads follows the general - purpose GPU route, targeting inference and PC scenarios, with the widest product line. Muxi focuses on the training scenario, competing closely with Moore Threads. Biren went to the Hong Kong stock market and focuses on high - performance training.

What about Enflame? It's increasingly taking on a role: The computing power foundation of the Tencent ecosystem.

This IPO aims to raise 6 billion yuan, and the investment direction is clear: R & D and industrialization of the fifth - generation AI chips, R & D and industrialization of the sixth - generation AI chips, and advanced software - hardware collaborative innovation.

Note that it's not using the money to make general - purpose GPUs to compete in the consumer market but continuing to dig deeper along the cloud - based AI training and inference line. The product roadmap highly coincides with Tencent's demand roadmap. Obviously, this is not a coincidence but the result of collaborative design.

It's also worth taking a look at Enflame's shareholder list. The National Integrated Circuit Industry Investment Fund Phase II, Tencent, Wuyuefeng Capital, Shanghai Guofang, and Shanghai Industrial Investment. This is a standard combination of state - owned capital and industrial capital. The national team provides endorsement, and the industrial side provides orders. In the chip industry, this structure means that it has a foothold in both policy and the market.

Looking at the entire industry pattern, this is a bit like how AWS has its self - developed chip Graviton, Google has TPU, and Meta has its self - developed ASIC. In the future, it may not be that everyone buys the same kind of GPU. Instead, each super cloud provider will have its own computing power system.

If this trend holds, Enflame's biggest competitor may not be Moore Threads or Muxi. Instead, it's another question: Will Tencent continue to support it in the future, or will it enter the chip - making business itself?

5. The 80/20 rule in China's AI industry

In the past five years, the market has evaluated AI chip companies based on the amount of financing, valuation, technical parameters, computing power indicators on PPT, and architectural innovation.

In the next five years, the industry will look at another set of indicators: order volume, delivery capacity, customer structure, and the depth of ecological binding.

What Enflame Technology deserves the whole industry's attention for is no longer the financial pain of losing 4.3 billion yuan in three years, nor when its fifth - generation chips will come off the production line. It's why Tencent is willing to place more than 80% of its computing power orders on this company's table.

Behind this question lies the next stage of China's AI industry. In the past, the competition was about who could make chips, and in the future, it will be about who can enter the supply chain of super customers.

Enflame is the first domestic chip company to truly enter the deep - water area of this supply chain.

Moore Threads, Muxi, and Biren have all gone public, and their prices have been set in the capital market. Enflame is the last to submit its answer, but what it holds in its hand is different from the previous three. It has a three - year procurement order from China's largest Internet company, with the amount doubling year after year.

In the current market environment, the value of this procurement order can help people better understand the value of this company than any top - level technical white paper.

Words "Beyond the Page":

For many years, the story of China's AI chips has always been about how to catch up with NVIDIA.

But perhaps the end of the chase is not to make a chip as powerful as NVIDIA's.

It's to find a customer who is willing to continuously buy your chips.

NVIDIA is what it is today not just because of its strong technology. It's also because the world's largest companies place billions of dollars of orders with it every quarter.

Technology can be caught up with. But trust, scenarios, and orders need to be earned step by step.

This may be the only long - term strategy for domestic chips in this long - distance computing power race.

This article is from the WeChat public account "Beyond the Page", author: Huahua. Republished by 36Kr with permission.