Enflame IPO: 84% of revenue came from one customer

Enflame closed its Shanghai debut on September 11 up 179%, worth about 170.9 billion yuan — roughly $25.5 billion — on 990.2 million yuan of 2025 revenue. That is a price-to-sales ratio north of 170. The number worth writing down is a different one: 84% of that revenue came from Tencent, which is also Enflame's largest shareholder. Read that as a lesson in how to evaluate any vendor you are about to depend on.

What actually happened

Enflame priced at 142.18 yuan, raised about 6.12 billion yuan ($911 million), opened at 410 yuan and closed at 397, touching 475 intraday. Per the South China Morning Post, the retail tranche was oversubscribed 4,073 times by about seven million applicants, with an allocation rate of 0.025%. It is the last of China's "four little dragons" of domestic AI silicon to list, after Moore Threads, Biren and MetaX.

Now the fundamentals. Jon Peddie Research puts Enflame's 2025 net loss at $178.6 million, narrowed from $223.2 million, on revenue growing north of 80% compounded since 2023. And it tracks the concentration: Tencent's share of Enflame revenue went 33% in 2023, 38% in 2024, then 84% in 2025. Enflame holds roughly 1.7% of China's AI accelerator shipments against Nvidia's 55%. The company builds domain-specific inference accelerators rather than general-purpose GPUs.

A market cap of $25.5 billion on 1.7% share and one customer is a bet on Beijing's import substitution policy, not on the current business. That can be a perfectly reasonable bet. It is just not the same bet as "this is a healthy company."

Why customer concentration matters for your business

You are not buying Enflame stock. You are, however, going to sign with a vendor this quarter whose pitch deck has a logo wall, and the same arithmetic applies.

When one customer is 84% of revenue, that customer sets the roadmap. Features you need get scheduled behind features they need. Pricing you negotiate is a rounding error against their contract. And if they leave or insource, the company you depend on has a survival problem, not a growth problem — which is when support quality goes first and the API deprecations start arriving with 30 days' notice.

The tell is usually available if you look. Public companies disclose concentration in their filings; ask a private vendor directly what share of revenue their top customer represents, and treat a refusal as an answer. Ask how many customers are at your size, not their biggest. Ask what happens to your data and your integration if the company is acquired — we have watched that question go unasked more than once.

Then build for the answer you get. Keep your data exportable in a format you can actually reload somewhere else. Put an abstraction layer between your application and any single model, chip or API provider, so switching is a config change and a week of testing rather than a rewrite. Write the exit into the contract: data export format, notice period, and what happens to your rate card at renewal. None of that is pessimism. It is what you do when a vendor's fate is tied to a customer that is not you.

Sources: South China Morning Post, Jon Peddie Research.