Tencent and Xiaomi both appear on Suiyuan Technology's IPO list, which could easily be read as a story of tech giants collectively placing orders for domestic GPUs. But the two companies play different roles: Tencent is both a major shareholder and a core customer, while Xiaomi is more accurately a strategic placement participant. Mixing equity, customers, and orders makes for a lively narrative but risks distorting the analysis.
What concerns me more is the shift in China's domestic computing power industry behind this listing: the focus has moved from fundraising, specs, and the promise of import substitution to real competition over customers, delivery, and payment collection. The first large order proves the product works; the second independent customer proves the business is replicable.
Capital Has Already Voted
Business Outcome Still Pending
Suiyuan's offering price is 142.18 yuan, with expected proceeds of about 6.119 billion yuan, implying a listing market value of roughly 61.187 billion yuan. For a chip company with 2025 revenue of about 990 million yuan and no profits yet, this is clearly not a price based solely on current earnings.
The capital market's willingness to assign such a valuation largely stems from its industrial shareholders. The prospectus discloses that Tencent and related entities hold about 20.26% combined before the offering, making it Suiyuan's largest external shareholder. Xiaomi, GigaDevice, and other industrial investors also joined the list through strategic placements.
These companies are not betting merely on a chip's peak computing power. What they truly value is whether Suiyuan can combine chips, software stacks, cluster deployment, and customer service into a computing power product that can be delivered consistently in cloud training and inference scenarios.
But a line needs to be drawn: industrial capital taking stakes only means investors are willing to take a seat at the table early; it does not mean they have become major customers. Capital can vote for the future in advance, but the business model still has to be proven order by order.
Tencent: Both the Answer and the Problem
Tencent plays a dual role with Suiyuan. It provides capital and industrial resources, as well as real intelligent computing centers and cloud business scenarios. In 2025, Suiyuan's revenue from Tencent-related customers was about 830 million yuan, accounting for 83.79% of total revenue.
That concentration looks high, but it shouldn't be read simply as negative. For an AI chip company, what's often most lacking in the early stage is not attention at product launches, but an anchor customer willing to put the chip into production environments and subject it to sustained, high-intensity validation.
Once chips enter real clusters, customers look beyond benchmark scores. Whether model migration goes smoothly, whether the cluster can run stably over long periods, whether faults can be handled quickly, and whether the software stack is compatible with existing operations — these issues only surface in large-scale use. Tencent provides exactly such a testing ground.
Thus, Tencent helped Enflame clear the first hurdle—moving from simply manufacturing chips to running them in real business operations. But flip that around and an interesting picture emerges: when a single customer simultaneously shoulders product validation, primary revenue, and growth expectations, the company's fortunes become deeply intertwined with that client. Tencent resolved the question of whether the chips could work, but did not fully resolve whether they could be replicated.
The Next Tencent
The Real Turning Point
The true commercial inflection point for domestic AI chips is usually not an IPO, nor landing the first big order, but the emergence of a second and third independent major customer. Only then can the market confirm that earlier success came from product capability, not just shareholder relationships or special procurement arrangements.
This step is far harder than imagined. Different customers have different model frameworks, server architectures, operations and maintenance systems, and procurement standards. Switching chips means re-adapting software, validating cluster stability, and bearing the time cost and business risk during migration. Similar peak computing power does not mean customers are willing to switch immediately.
Enflame's next-generation products have entered hardware adaptation, model matching, and gray testing with some potential customers, showing it is trying to replicate the capabilities developed in Tencent's environment. But between "entering testing" and "forming scale orders" lie three hurdles: delivery, repeat purchases, and payment collection.
So what is really worth watching next is not which institution appears on the shareholder list, but whether the revenue share from non-Tencent customers can rise, whether third-party customers make repeat purchases after their first order, and whether new orders can be delivered smoothly while maintaining gross margins. The moment a second Tencent appears will mark the beginning of Enflame's transition from a project-based company to a platform-based company.
Valuation Has Run Ahead of Profits
Based on a market capitalization of about 61.187 billion yuan and 2025 revenue of 990 million yuan, Enflame's price-to-sales ratio is about 61.8 times. This means the capital market is pricing not the money it has already earned, but the possibility of rapid expansion in future domestic computing-power orders.
But AI chips are a very capital-intensive business. From 2023 to 2025, Enflame's cumulative R&D investment was about 3.676 billion yuan, exceeding its cumulative revenue of about 2.013 billion yuan over the same period. Chips need continuous iteration, and the software ecosystem requires long-term maintenance. Even as the previous generation starts to ramp up, R&D spending on the next generation is hard to stop.
Revenue reached about 1.12 billion yuan in 1H 2026, up roughly 279% year-on-year, but the company still lost around 632 million yuan, with losses barely narrowing from a year earlier. While revenue growth is important, it also shows that scale expansion has yet to translate into profits.
The company expects to turn profitable on a consolidated basis in 2026 or 2027, but that timeline depends on gross margin, product yield, supply stability and purchase volumes from new customers. A gross margin of around 30% is only a starting point; whether it can cover rising R&D and operating expenses will determine if the so-called profit inflection point truly arrives.
My assessment of Suiyuan won't stop at 'domestic substitution.' The market is really repricing whether domestic computing power can move from single-customer validation to multi-customer replication. Going forward, I'll watch non-Tencent revenue share, gross margin and operating cash flow more than revenue growth. Would you rather wait for profits to emerge, or for a second major customer to sign on?
#SuiyuanTech #DomesticComputingPower #AIChips #Semiconductors #TechInvesting