The Shanghai AI chip startup is seeking about $830 million in a listing that lays bare a central tension: Tencent is simultaneously Enflame's biggest backer, its largest shareholder, and the source of nearly three quarters of its revenue.
When Enflame Technology's STAR Market IPO application was accepted in January 2026, the Shanghai startup did not just set a fundraising target. It opened its books on one of the most revealing relationships in China's technology sector: a chip company that owes 71.84% of its revenue to a single shareholder and customer, Tencent, which also holds 20.26% of its equity. That is not a partnership. That is a dependency, and the institutional investors who price this listing will have to decide whether Tencent's grip makes Enflame bulletproof or captive.
The growth story is real. Revenue climbed from RMB 90.1 million in 2022 to RMB 722.4 million in 2024, roughly an eightfold rise in two years. The fourth-generation L600 chip, unveiled at the World AI Conference in Shanghai last July, packs 144GB of on-chip memory and 3.6 terabytes per second of memory bandwidth with FP8 support. Its predecessor, the S60, had already shipped approximately 70,000 units by mid-2025. These are not vaporware numbers, and Bloomberg has reported fresh movement on the listing process.
The losses are just as real. Net losses hit RMB 1.51 billion in 2024, and Enflame burned through another RMB 887.8 million in the first nine months of 2025 alone. Cumulative losses across three years reach RMB 4.29 billion, or about $600 million, according to the company's prospectus. Costs are outpacing revenue, not the other way around. That is the classic profile of a company racing to become indispensable before the economics close in, and whether it gets there depends almost entirely on whether Tencent keeps ordering.
Captive suppliers are not inherently weak businesses. Tencent deploying Enflame chips inside its own data centers while holding a fifth of the company's equity is a fundamentally different arrangement from a startup hunting for its next sale. Tencent pre-funds Enflame's roadmap through purchasing commitments, which is why a chip with 144GB of on-chip memory exists at all. You do not build that without a guaranteed buyer on the other end, and you do not guarantee a buyer without giving them a stake in the outcome.
But the model has a visible cost. Tencent's volume has already forced a 6.5% compression in Enflame's average selling price, down to roughly $1,836 per module, according to the company's filings. That is what monopoly customer power looks like from the inside. If Tencent's AI infrastructure spending slows, or its priorities shift, Enflame has no meaningful fallback. The prospectus discloses this plainly.
What complicates the timing is that the export control landscape has not moved in a straight line. The Trump administration approved limited Nvidia H200 sales to China under strict conditions, including a 25% revenue share to the U.S. government, but the Financial Times reported in January that Chinese customs had blocked H200 shipments even after Washington cleared them. Reuters could not immediately verify the report, but the uncertainty is the point. Tencent may want access to Nvidia hardware, Beijing may want domestic substitution, and Enflame is being asked to build a business inside that gap.
The most credible case for Enflame is long-term hedging. Tencent cannot risk another blackout if Washington reverses course, and China's push for semiconductor self-sufficiency has not slowed because one export rule changed. The incentive to develop domestic alternatives is structural, not situational. Enflame still needs Tencent to stay committed across years, not quarters, and that kind of commitment is harder to guarantee when policy can move faster than a chip roadmap.
One giant, a crowded field
Even with Tencent's backing, Enflame enters the public market in a domestic landscape it does not lead. Reuters reported in April, citing IDC figures, that Chinese chipmakers shipped 1.65 million AI GPUs in 2025 and took 41% of China's domestic AI server market, with Huawei accounting for about 812,000 units. Cambricon has also moved from chronic losses to a profitable 2025 and reported RMB 2.89 billion in first-quarter 2026 revenue. A company deriving nearly three quarters of its revenue from one customer is not well positioned to challenge those installed bases in any near-term sense.
That context reframes what the RMB 6 billion is actually for. Pandaily's reporting on the prospectus shows the proceeds are earmarked for fifth- and sixth-generation chip R&D and a hardware-software co-innovation project, not customer diversification. The underlying bet is that Enflame's technology improves fast enough to stay essential inside Tencent's infrastructure and eventually attract a second anchor buyer. Co-founder Zhao Lidong, who worked at AMD before helping start Enflame in 2018, is making a deliberate long-cycle wager on that outcome. It is also worth noting that Chinese regulators eased listing rules for unprofitable hardware companies in June 2025, which is partly why Enflame can pursue this listing at all despite its loss profile. Moore Threads, MetaX, and Biren have already gone public and are trading well above their IPO prices, giving Enflame a receptive market backdrop.
China's AI chip sector needs a credible public company to anchor it, one with real silicon, real shipments, and a genuine technology roadmap. Enflame is one of the strongest independent candidates for that role right now. Whether it holds that position while carrying about $600 million in cumulative losses and a customer list of essentially one is the harder question the IPO actually asks.
Also read: Legal AI's two-platform race is hardening and Legora just planted its flag in Europe • Kazakhstan bets on Nvidia-backed Firebird to become Central Asia's AI hub • FINQ's AI-run ETFs beat the S&P 500 in their debut and the industry is taking notes
