NextFin News - Retail investors bid 5.98 trillion yuan ($890 billion) for a sliver of Shanghai Enflame Technology's initial public offering, a 4,073-times subscription that underscores how far China's domestic AI-chip mania has traveled since the first of the country's "four little dragons" reached the public market. The frenzy is not just about Enflame. It is about what the order book says investors are willing to pay for any company that can claim to be a homegrown answer to Nvidia — even one that is not yet profitable and sells 84% of what it makes to a single customer that also happens to be its largest shareholder. Data as of September 2, 2026.
The Deal: 6.12 Billion Yuan for the Last Dragon
The initial public offering of Shanghai Enflame Technology Co., a Tencent Holdings-backed designer of artificial-intelligence accelerator chips, drew demand from individual investors that dwarfed every recent Chinese debut except one. The retail tranche was subscribed 4,073 times, with roughly 7 million orders placed for about 5.98 trillion yuan of shares, according to an exchange filing released Wednesday. The only comparable book this year was memory-chip maker CXMT's blockbuster July listing, which drew a 7.1-trillion-yuan retail order book and stands as the second-biggest China debut on record.
Enflame priced its offering at 142.18 yuan a share on Shanghai's STAR Market, selling about 43 million shares — roughly 10% of its enlarged capital — and raising approximately 6.12 billion yuan ($911 million), slightly above its original 6-billion-yuan target. The China Securities Regulatory Commission approved the listing in July, clearing the last of the so-called "four little dragons" of Chinese AI-chip makers to enter public markets. Moore Threads Technology and MetaX Integrated Circuits listed on the STAR Market in December; Biren Technology debuted on the Hong Kong exchange in January. Enflame, founded in 2018, was the earliest of the four to be established but the last to file.
The valuation math is where the tension begins. At the offer price, Enflame trades at 61.8 times its 2025 sales, the company said in its filing. That is a steep discount to the more than 160 times sales commanded by listed peers Moore Threads and MetaX, but a large premium to Nvidia's 25.4 times. Enflame's revenue reached 990 million yuan in 2025, growing at a compound annual rate of about 81% from 2023, yet the company posted a net loss of 1.2 billion yuan in 2025, narrowed from 1.5 billion yuan a year earlier, and expects to lose about 600 million yuan in the first half of 2026. Since 2022, cumulative losses have exceeded 5 billion yuan.
The customer concentration is the second tension. Tencent accounted for 83.79% of Enflame's 2025 revenue, up from 37.77% in 2024 and 33.34% in 2023, while holding a 20.26% stake in the company together with its concert party. The company disclosed the relationship plainly in its prospectus:
Tencent's demand for AI accelerator cards has far exceeded the company's supply capacity.
The proceeds will fund production of its fifth- and sixth-generation AI chips and advanced software and hardware projects.
What the 4,073-Times Number Really Buys
The subscription multiple is a measure of access, not of conviction about Enflame's standalone fundamentals. China's retail investors cannot buy Nvidia. They cannot easily buy the unlisted AI-chip champions either. What they can buy is a listed vehicle with a domestic-AI narrative, a recognizable anchor shareholder, and a STAR Market ticker that carries no daily price limit during its first five trading days. The 4,073-times figure tells you how starved retail capital is for that specific combination.
The precedent supports that reading. Moore Threads rose 425% on its December debut; MetaX nearly septupled on its first day; Biren's grey-market shares jumped 80.41% the day before its Hong Kong listing. Each of those moves had little to do with near-term earnings and everything to do with scarcity value. Enflame is the final piece of a collectible set — the last of the four dragons to list — and retail order books price scarcity before they price cash flow.
That mechanism has a limit, however. The comparison with CXMT is instructive. CXMT's 7.1-trillion-yuan retail book in July came with a cleaner strategic story: memory chips are a commodity with visible demand, and CXMT is the national champion in a category where import substitution is a matter of policy. Enflame's addressable market is narrower. Nvidia held 55% of China's AI-accelerator shipments in 2025; Enflame held an estimated 1.7%. A 1.7% share does not command a 61.8-times sales multiple on fundamentals. It commands it on the option value of the remaining 98.3%.
The Captive-Supplier Model Investors Are Actually Buying
Enflame's structure inverts the usual IPO red flag. Concentrated revenue from a single customer normally depresses valuations because it signals weak bargaining power. Here, the customer is also the largest external shareholder, and the concentration is the business model rather than a defect. Tencent is not merely a buyer; it is a guaranteed offtaker with a strategic incentive to keep Enflame alive, funded, and technically improving.
Read literally, the prospectus disclosure reframes the entire investment case. The risk is not whether Tencent will buy. The risk is whether Enflame can build enough working silicon to sell. For a retail investor, that converts a demand-side question into a supply-side one — a much more forgiving framing for a company that has burned through more than 5 billion yuan since 2022.
But the model carries its own ceiling. A captive anchor customer caps the price Enflame can charge and limits the evidence that its chips compete in an open market. Until Enflame proves it can win orders from cloud providers, model developers, or government computing projects outside the Tencent ecosystem, the revenue line is a transfer within a related-party orbit rather than proof of product-market fit. The prospectus itself acknowledges the dependency even as it frames it as strength.
Why the STAR Market Lets Loss-Makers Price at 62 Times Sales
The institutional enabler of this deal is the STAR Market itself. Launched in 2019 as Shanghai's answer to Nasdaq, the board was built to channel capital toward strategic technology sectors, with listing rules that accommodate unprofitable companies in ways China's traditional exchanges do not. Enflame's ability to list despite significant losses is a direct product of those more flexible standards — and of a policy environment that treats domestic semiconductor capacity as a national-security priority rather than a purely commercial one.
That policy backdrop is the reason the multiple survives scrutiny at all. Washington has progressively tightened export controls on advanced chips and chipmaking equipment to China, pushing Beijing to accelerate domestic alternatives. Every restriction that keeps Nvidia's most advanced products out of Chinese data centers enlarges the protected home market for Enflame and its peers. The capital markets are not pricing Enflame's current earnings; they are pricing the stream of revenue that a closed border around China's AI compute demand could deliver over the next decade.
The arithmetic of that protection is visible in the market-share gap. Nvidia's 55% share of China's AI-accelerator shipments in 2025 represents the incumbent position that policy is trying to dislodge. Enflame's 1.7% is the starting point. The spread between those two numbers is the entire investment thesis — and the reason a company with a 1.12-billion-yuan first-half revenue run can raise 6.12 billion yuan in a single offering.
Cyclical Mania or a Structural Regime?
This is where the judgment has to be made. Is the 4,073-times subscription a cyclical peak in IPO sentiment that will revert, or the first evidence of a structural regime in which China's retail market permanently re-rates domestic semiconductor assets?
The cyclical case is strong. China's IPO fever has a well-documented history of overshooting and mean-reverting. The first wave of STAR Market listings after 2019 produced spectacular first-day pops and mediocre long-term returns. The current cycle is being driven by a specific, time-bound catalyst — US export controls that make any credible domestic AI-chip name a scarce asset — and by a retail order book that has already absorbed three comparable debuts in twelve months. Once the fourth dragon has listed, the scarcity premium that powered the first three has nowhere to go.
The structural case rests on policy durability. US restrictions on advanced chips and chipmaking equipment are not a cycle; they are a regime. As long as Chinese cloud giants and state computing projects are barred from Nvidia's most advanced products, domestic suppliers have a protected home market that did not exist five years ago. The capital markets are simply catching up to that reality, and the re-rating could persist as long as the restrictions do.
The right read is a hybrid: the policy backdrop is structural, but the valuation is cyclical. The protected home market is real and durable. A 61.8-times sales multiple for a company with a 1.7% market share and an 84% single-customer concentration is not. When the cycle turns, the structural thesis survives but the multiple does not.
The Second-Order Question Nobody in the Order Book Is Asking
The first-order effect of Enflame's debut is obvious: another Chinese AI-chip name becomes tradable, and retail capital piles in. The second-order effect is what happens to Tencent.
Enflame is the cheapest way for Tencent to monetize its AI-chip strategy without selling down its own stake. By taking a portfolio company public at 61.8 times sales, Tencent converts years of private-market funding into a marked-to-market asset on its own balance sheet. If Enflame's shares trade up on debut — as the peers' first-day performances suggest they will — Tencent's 20.26% holding is revalued upward, improving the parent's reported equity value without a single share sale. That is the quiet arithmetic behind the anchor-shareholder structure: the captive customer relationship doubles as a valuation engine for the customer itself.
The third-order implication runs through China's broader AI supply chain. Enflame's successful pricing gives regulators and sponsors confidence to keep the IPO gate open for the next tier of domestic chip designers, model developers, and computing-cluster operators. The financing push now spans the entire supply chain, from chipmakers to model developers. A 4,073-times retail book is not just a vote for Enflame; it is a green light for the next ten offerings.
The Counter-Thesis: This Is Not Mania, It Is Rational Optionality
The strongest argument against the mania reading is that the market is pricing a real option, not current earnings. Enflame's L600 chip, launched at the World Artificial Intelligence Conference in July, is a fourth-generation training-and-inference processor with 144GB of memory, 3.6TB/s of bandwidth, and FP8 support. Its S60 inference chip has shipped 70,000 units and powers a 10,016-card cluster in Gansu. First-half 2026 revenue reached 1.12 billion yuan, up 279% year on year — nearly quadruple the comparable period. These are not the metrics of a shell company riding a narrative.
A rational investor can argue that paying 61.8 times sales for a company operating in a market where the incumbent is legally barred from selling its best products is not irrational at all. The option value of capturing even a fraction of the 98.3% of the market that Nvidia still serves justifies a loss-making balance sheet today. If Enflame converts Tencent's captive demand into a broader customer base, today's multiple will look cheap in hindsight.
That argument holds only if Enflame can break out of the Tencent orbit. The falsifying signal is specific: if, by the end of 2027, Tencent's share of Enflame's revenue remains above 70% and no top-five Chinese cloud provider has placed a material order for Enflame silicon, the option-value thesis fails and the multiple has no support beyond policy sentiment. Customer diversification is not a nice-to-have metric here; it is the entire case.
What to Watch Next
The beneficiaries of this IPO are clear. Tencent holds a marked-to-market asset that appreciates if the debut trades up. Existing shareholders — including the founders, who together control 28.14% of voting rights — gain liquidity and a public currency for future fundraising. The exposed parties are the retail investors bidding 4,073 times subscription into a stock with no daily price limit in its first five trading days and no earnings to anchor it, and the peer group: if Enflame's debut disappoints, the scarcity premium that lifted Moore Threads and MetaX could compress across the sector.
Short term, the stock will trade on sentiment and scarcity, and the first-day move will likely follow the peers' pattern of a large pop. Medium term, the stock will be anchored by two numbers: the pace at which Enflame narrows its losses, and the rate at which non-Tencent revenue grows. Long term, the structural thesis survives only if US restrictions remain in place and Enflame's technology stays within striking distance of the foreign products they keep out.
Base case: Enflame debuts strongly, the multiple holds near current levels for a quarter, and revenue growth continues to outrun loss reduction. Upside case: a material order from a non-Tencent cloud provider or government computing project re-rates the stock as a genuine competitor rather than a captive supplier. Downside case: the debut fades, revenue growth slows as Tencent's build-out matures, and the 61.8-times multiple compresses toward the peer median.
Watch three signals: the first-day closing price and turnover, full-year 2026 revenue against the company's growth trajectory, and the customer-concentration disclosure in the next annual report. If concentration stays above 70% through 2027, the structural-shift story is an option trade, not a company.
The 4,073-times subscription is not a vote for Enflame's earnings. It is a vote for the one thing China's retail market cannot get enough of: a listed claim on the domestic AI-chip dream, priced as if the dream were already profitable.
Explore more exclusive insights at nextfin.ai.
