--- title: "Enflame Priced Its STAR Market IPO at RMB 142.18. The Harder Exam Is Whether Developers Will Use It" date: 2026-09-02 category: Chips & Compute site: NeuroAI canonical: https://neuroai.site/a/na-enflame-star-market-ipo language: en --- # Enflame Priced Its STAR Market IPO at RMB 142.18. The Harder Exam Is Whether Developers Will Use It > Enflame Technology raised RMB 6.119 billion at RMB 142.18 per share. The listing is a beginning, not an achievement: a high-end AI chip costs hundreds of millions to tape out, and the real moat is software ecosystem, not silicon. ## Key takeaways - On the evening of **31 August 2026**, Enflame Technology's STAR Market IPO was priced: **RMB 142.18 per share**, **43.0352 million shares**, gross proceeds of **RMB 6.119 billion**, net proceeds of approximately **RMB 5.81 billion**. - Context on the demand side: in January–July 2026, profits at above-scale industrial enterprises grew 17.6% year on year, with the **electronics sector up 110%** and the **integrated circuit sector up 1,850%**. High-tech manufacturing overall grew 50.1%. - Evidence that domestic silicon is actually being used: **Zhipu's API revenue rose 2,736% year on year in the first half, while its per-token inference cost fell 80%** — a cost curve driven substantially by scaling inference on domestic chips. - Three genuine risks: ecosystem migration, customer concentration, and valuation. ## An IPO is where the funding starts, not where the work ends For a chip company, going public is not a finish line. It is the beginning of staying alive. A high-end AI chip costs hundreds of millions of yuan from design to tape-out. One failed tape-out burns tens of millions of dollars. Talent costs more: for a mature chip design team, personnel typically accounts for more than half of R&D spend. RMB 5.81 billion sounds like a great deal of money. In the GPU business it funds roughly two or three generations of product iteration. Which is why the market's question for Enflame was never "can it list." It is "can it keep shipping competitive products after listing." ## The tailwind is real this time Enflame has good timing, and the numbers show it. January to July 2026: profits at above-scale industrial enterprises up 17.6% year on year; the electronics industry up 110%; integrated circuits up 1,850%; high-tech manufacturing up 50.1% overall. That is not paper prosperity. Inference demand from Chinese model developers is genuinely expanding — Zhipu's first-half API revenue grew 2,736% year on year while its per-token inference cost fell 80%, and scaled deployment of domestic chips is one of the core reasons that cost curve is possible at all. Demand is releasing. Supply is filling in. Enflame sits at that junction. ## Three realities worth stating **Ecosystem is harder than silicon.** Nvidia's real moat is not the card. It is CUDA — fifteen-plus years of accumulated developers, tools and libraries. For a domestic chip to catch up, the difficult part is persuading developers to migrate. **Customer concentration.** Current orders for domestic AI chips depend heavily on a small number of large internet companies and state-backed compute platforms. If one of those customers starts designing its own silicon or changes direction, revenue swings hard. **Valuation.** On 1 September, semiconductor and compute-hardware names on China's A-share market pulled back together. The external environment is not friendly to high-multiple chip stocks, and a listing does not guarantee a repeat of Cambricon's trajectory. ## The observation window: from "can it be built" to "is it good to use" To judge real progress in domestic AI chips, ignore conference parameters and watch three plainer indicators. **Do model developers use it?** Whether Chinese model companies are willing to move their primary inference load onto domestic silicon is the hardest vote available. **Has unit cost fallen?** Zhipu's 80% drop in per-token inference cost in the first half came substantially from scaled domestic inference. Cost data does not lie. **Is there a developer community?** Whether anyone in open-source communities is voluntarily porting, writing tutorials and building tooling determines whether the platform is still alive in three years. ## What to do with this If you are considering subscribing to the offering: look at the gap between the issue price-earnings multiple and comparable companies. Chip stocks are extremely volatile in early trading, and being allotted shares is not the same as free money. If you want exposure to AI compute: prioritise demand certainty — whose models are actually running, whose orders are actually landing — over whose story is more elegant. If you want to enter the industry: the chip talent shortage is structural. Verification, back-end design and driver development have lower barriers to entry than algorithm work and are just as scarce. Building a good chip is a passing grade. Getting ten million developers to want to use it is the actual exam. ## Honest limitations The 1,850% integrated-circuit profit growth figure is measured against a very low base and is not a growth rate anyone should extrapolate. This is not investment advice, and I hold no position in any company mentioned. IPO pricing data is as announced; post-listing performance is unknowable in advance. *Sources: Shanghai Stock Exchange announcement (31 August 2026); National Bureau of Statistics industrial profit data; Zhipu first-half disclosures. Information only — not investment advice.* --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-enflame-star-market-ipo Free to quote with attribution and a link to the original.