Without Tencent, What's Left for Suiyuan?

marsbitPublicado em 2026-06-15Última atualização em 2026-06-15

Resumo

The article centers on the crucial question posed in the title: what is Seyond Technology really worth if its dominant customer, Tencent, were to stop purchasing its AI chips? As the last of China's "Four AI Chip Dragons" to secure approval for a public listing, Seyond's IPO filing reveals a profound and controversial dependency. In 2025, 74.9% to over 80% of its revenue came from Tencent. The piece argues that this extreme customer concentration is not merely a vulnerability but a strategic outcome of China's AI industry evolution. It contrasts Seyond's path with its peers (Moore Thread, Biren Technology, and MetaX), noting that while others raced to market with ambitious stories, Seyond focused first on securing and delivering for a major client. Its explosive revenue growth—with Q1 2026 up 1474.85% year-on-year—is driven by concentrated orders from Tencent, which itself faces massive, escalating AI compute demands for products like its Yuanbao and Hunyuan models. The relationship is framed as a deliberate, symbiotic cultivation of a supply chain. As both a major shareholder (20.26%) and primary client, Tencent is actively fostering Seyond to build a controllable, stable alternative to NVIDIA, similar to how global tech giants historically nurtured key suppliers. The high switching costs—involving software stacks and deployed systems—create a deep "ecological moat" for Seyond within Tencent's ecosystem. The analysis positions the AI chip landscape in three tiers: NVIDIA ...

If Tencent stopped procuring tomorrow, how much would Suiyuan be worth?

This is the question that runs through Suiyuan's IPO prospectus, unspoken by anyone, yet ever-present.

In 2025, 74.9% of Suiyuan's 9.9 billion yuan in revenue came from Tencent.

By another statistical measure, that figure even exceeds 80%.

In other words, for this soon-to-list AI chip star company on the STAR Market, nearly 8 out of every 10 yuan earned comes from the same customer.

On the 15th, news came from the Shanghai Stock Exchange that Suiyuan Technology's IPO application for the STAR Market was officially approved by the listing committee. This means it has successfully submitted its paper, completing the final encirclement of the domestic GPU "Four Little Dragons" in the capital market.

Last December, Moore Thread and Muxi Stock successively made landings on the STAR Market, both having grown into behemoths close to 300 billion yuan in market value; this January, Biren Technology also successfully listed on the Hong Kong stock market.

Although Suiyuan arrived last, its successful passing signals an extremely clear migration trend in China's AI chip industry to the outside world: The computing power industry has completely bid farewell to PPT-style financing-driven models and has fully entered an extremely brutal cycle of order delivery.

I. Why Was Suiyuan the Latest?

Suiyuan may be the least startup-like company among the Four Little Dragons.

Moore Thread aimed at a general-purpose GPU route from the start, telling the story of a Chinese GeForce; Biren focused on high-performance training chips, taking a hardcore parameter route; Muxi pushed forward in the training card direction, emphasizing technological catch-up.

The rhythm of these three was to first charge into the capital market, first maximize valuation, first amplify the story.

Suiyuan took a somewhat different path. From its inception, it anchored itself to one thing: big customer delivery. It's not about telling stories first, it's about doing business first.

The data makes it clear.

Revenue of 3 billion yuan in 2023, 7.2 billion in 2024, 9.9 billion in 2025. A three-year compound annual growth rate of 81%. An even more astonishing explosion occurred in Q1 2026, with single-quarter revenue soaring to 2.87 billion yuan, a staggering year-on-year increase of 1474.85%. Revenue for the first half of 2026 is projected to directly skyrocket past 10.6 billion yuan, surpassing the total of the entire previous year in just six months.

This extremely abnormal steep growth rate is by no means a natural, evenly distributed industry bounty, but rather the concentrated, targeted release of computing power orders from a super-large customer.

Suiyuan has reaped the dividends of China's computing power construction cycle. It traded delivery capability for orders, orders for revenue, and revenue for the confidence to go public.

The fact that it is the latest to list itself illustrates a principle: in the AI chip industry, what matters now is not who lists first, but who has real revenue in hand.

II. The Industrial Truth Behind 83.79%

There is a conspicuous set of numbers in the prospectus. In 2025, sales revenue from Tencent accounted for 74.9% of Suiyuan's total. According to the response to the second round of inquiries (including the AVAP model), this figure is 83.79%.

Viewed superficially in terms of structure: excessively high customer concentration, related-party transactions, doomed if Tencent reduces procurement.

Correct. These are all facts.

But viewed from an industrial perspective, the more worthwhile question is: Under Nvidia's technological monopoly and the squeeze from domestic competitors, why is Tencent so determined to buy so much from Suiyuan?

You have to understand, for major manufacturers like Tencent and ByteDance, the procurement logic used to be simple: buy Nvidia if possible, never use domestic; buy from the first tier if possible, never compromise with the second tier.

But today, Tencent is starting to procure from Suiyuan on a large scale. In 2023, Tencent's procurement from Suiyuan was only 1 billion yuan; in 2024, it climbed to 2.7 billion; by 2025, it switched to large-scale direct sales, pulling orders up to 7.68 billion yuan in one go. In three years, it increased nearly eightfold.

Behind this is not Tencent suddenly becoming a philanthropist; it's that Tencent's computing power demand has grown too large to rely solely on Nvidia anymore.

Yuanbao, Hunyuan, WeChat AI, Enterprise WeChat, Tencent Meeting, CodeBuddy, WorkBuddy – all these products burn GPUs in the background. And it's continuous burning, burning more and more.

Tencent's Q1 2026 financial report shows that new AI products alone dragged down operating profit by approximately 8.8 billion yuan for the quarter; capital expenditure reached 31.9 billion yuan, a significant year-on-year increase, with management expecting full-year investment in new AI businesses to double.

Under this demand pressure, what Tencent needs is no longer just a more powerful card, but a controllable, stable computing power supply system that won't be choked by supply chain issues.

Especially since over 80% of Suiyuan Technology's accelerator card and module revenue currently comes from inference products, which are most urgently needed for large model deployment. What Suiyuan sells is never just silicon wafers; it's the sense of security urgently needed by Tencent's large models.

III. The Dual Identity of Shareholder and Customer

Many are accustomed to interpreting the 83.79% as a one-way, transfusion-like dependency. But looked at the other way, it's Tencent actively nurturing Suiyuan.

Why? Because Tencent clearly understands one thing: if domestic AI chip companies don't have enough orders, they will never grow up. Without production scale, there's no process iteration; without iteration, they'll never catch up to Nvidia.

Why is Nvidia strong? Not because it was strong from day one. It's because Microsoft, Meta, Amazon continuously gave orders, continuously provided scenarios, continuously gave feedback. It was nurtured by one big customer after another.

What Tencent is doing for Suiyuan today is essentially the same thing.

This strongly resembles when Apple, to combat industry monopoly, poured resources into supporting TSMC. The relationship between the two had long transcended ordinary buyer-seller contracts; it was about using definitive orders to help the supplier refine the underlying process to the extreme, then making it an irreplaceable part of my supply chain.

Tencent, through its technology affiliate and related parties, collectively holds 20.26% of Suiyuan's shares, firmly sitting as the largest shareholder, while also being the largest customer. The deep binding of shareholder + customer dual identity. This structure is seen as a related-party transaction risk in traditional PE eyes, but in industrial logic, this is supply chain cultivation.

Suiyuan's response in the first round of inquiries was blunt enough: Tencent lacks both the motivation and the practical conditions to replace the company's products on a large scale with other suppliers at this stage.

Simply put, Tencent has already deployed Suiyuan's chips into numerous real business scenarios, possibly including the inference clusters for the Hunyuan large model, the daily dialogues of Yuanbao, and Tencent Cloud's AI computing power services for enterprise customers.

To switch isn't just about swapping a card; it's about redoing the entire software stack, drivers, scheduling system. The cost is extremely high.

This also explains why, although a small number of domestic competitors have entered Tencent's supply chain, as Suiyuan itself acknowledged, Tencent's main procurement position is still reserved for it. What Tencent needs isn't to buy from multiple suppliers to diversify risk; it's to get one company to run through the entire process, then replicate it on a large scale.

At this moment, the 74.9% sales concentration transforms from a weakness into Suiyuan's ecosystem moat.

IV. The Ecosystem Fragmentation of the Third Tier in AI Chips

If you look at the AI chip industry today in isolation, the landscape is already very clear.

First Tier: Nvidia. The rule-maker. Global ecosystem, CUDA lock-in, dictates capacity.

Second Tier: Huawei Ascend. A national-level player. Own ecosystem, tied to operators and state-owned enterprises, not entirely market-driven.

Third Tier: Suiyuan, Moore Thread, Muxi, Biren. Commercial players, survive on market orders.

In the past, people thought these companies in the third tier were competitors, all making domestic GPUs, all competing for the same batch of customers.

But today, it looks more like another pattern: each is binding to different ecosystems.

Moore Thread takes the general-purpose GPU route, targeting inference and PC scenarios, with the broadest product line. Muxi focuses on the training scenario, competing head-to-head with Moore Thread. Biren went to Hong Kong, leaning towards high-performance training.

And Suiyuan? It's increasingly taking on one identity: the computing power foundation of the Tencent ecosystem.

This IPO aims to raise 6 billion yuan, with clear investment directions: R&D and industrialization of fifth-generation AI chips, R&D of sixth-generation AI chips, and advanced software-hardware collaborative innovation.

Note, it's not taking money to make general-purpose GPUs to grab the consumer market, but continuing to drill deeper down the line of cloud AI training and inference. The product roadmap and Tencent's demand roadmap are highly aligned, clearly not a coincidence, but the result of co-design.

Suiyuan's shareholder list is also worth a glance. National Integrated Circuit Industry Investment Fund Phase II, Tencent, Summitview Capital, Shanghai Guofang, Shanghai ChanTou. This is a standard combination of state capital + industrial capital. The national team provides endorsement, the industrial side provides orders. This structure in the chip industry means it stands firm on both policy and market legs.

Looking at the entire industry landscape, this is somewhat similar to AWS having its own Graviton chips, Google having TPUs, Meta having its own ASICs. The future may not be everyone buying the same kind of GPU, but each super cloud vendor having its own computing power system.

If this trend holds, Suiyuan's biggest competitor may not be Moore Thread or Muxi. It's another question: will Tencent continue to support it in the future, will Tencent venture into chip-making itself.

V. The 80/20 Rule of China's AI Industry

Over the past five years, the market evaluated AI chip companies based on financing amounts, valuations, technical parameters, computing power metrics on PPTs, and architectural innovations.

In the next five years, the industry will look at another set of metrics: order volume, delivery capability, customer structure, depth of ecosystem binding.

What makes Suiyuan Technology most worthy of attention across the industry is no longer its financial pain of losing 4.3 billion yuan over three years, nor when its fifth-generation chip will roll off the production line. It's why Tencent is willing to place over eighty percent of its massive computing power pie entirely on this company's table.

Behind this question lies the next stage of China's AI industry. The past was about who could make the chip; the future is about who can enter the supply chain of super customers.

Suiyuan is the first among domestic chip companies to truly step into this deep-water zone of the supply chain.

Moore Thread, Muxi, and Biren have all already gone public, and have valuations in the capital market. Suiyuan is the last to submit its paper, but what it holds in its hand is different from the first three: it has a procurement order from China's largest internet company, written for three years with the amount doubling year after year.

In the current market environment, the substance of this purchase order speaks more clearly to the company's value than any top-tier technical white paper.

Words from 【Beyond the Layout】:

For years, the story told about China's AI chips has been about how to catch up with Nvidia.

But perhaps the endpoint of catching up is not making a chip as strong as Nvidia's.

It's finding a customer willing to keep buying your chips.

Nvidia is Nvidia not just because its technology is strong. It's because the world's largest companies place orders worth tens of billions of dollars with it every quarter.

Technology can be chased. But trust, scenarios, and orders must be earned, one by one.

This, perhaps, is the only long-termism for domestic chips in this computing power Long March.

This article is from the WeChat public account "Beyond the Layout," author: Huahua

Perguntas relacionadas

QWhat is the central question raised by the article about Enflame Technology's IPO prospectus, and what is the core evidence supporting it?

AThe central question is how much Enflame Technology would be worth if its major customer, Tencent, were to stop its purchases. The core evidence is that in 2025, 74.9% (or over 80% by another accounting method) of Enflame's 9.9 billion yuan in revenue came from Tencent, indicating a critical dependency.

QAccording to the article, why did Enflame Technology take the latest among the 'Four Little Dragons' to get listed, and what does this timing signify about the industry shift?

AEnflame was the latest to list because it prioritized securing and fulfilling orders from major clients, particularly Tencent, over rapidly pursuing capital market hype. This signifies that the AI chip industry in China has shifted from a phase driven by storytelling and fundraising to one dominated by the brutal reality of securing and delivering on actual orders.

QBeyond simple dependency, how does the article reinterpret the meaning of Tencent accounting for over 83% of Enflame's sales?

AThe article reinterprets this high concentration not just as a risk, but as a strategic 'ecological moat' or supply chain cultivation. Tencent, as both a major shareholder and the largest customer, is actively nurturing Enflame with guaranteed orders to build a controllable, stable, and non-interruptible domestic compute supply system, similar to how major U.S. tech firms fostered their suppliers. The high switching costs due to deep software and system integration further solidify this partnership.

QHow does the article categorize the current landscape of the AI chip industry, and what specific role does it assign to Enflame Technology within this structure?

AThe article categorizes the industry into three layers: 1) Nvidia (the global rule-maker), 2) Huawei Ascend (a national-level player with its own ecosystem), and 3) Commercial players like Enflame, Moore Thread, Biren, and MetaX. Within the third layer, Enflame is increasingly seen as 'the compute foundation for the Tencent ecosystem,' specializing in cloud AI training and inference chips that are co-designed with Tencent's roadmap, rather than a direct competitor for the general GPU market.

QWhat fundamental shift in the evaluation criteria for Chinese AI chip companies does the article predict for the next five years?

AThe article predicts a shift from evaluating companies based on fundraising amounts, valuations, technical specs, and PPT promises to judging them by tangible metrics like order volume, delivery capability, customer structure, and the depth of their ecosystem partnerships. Enflame's value is exemplified by its multi-year, rapidly scaling purchase order from Tencent, which is portrayed as more telling than any technical whitepaper in the current market.

Leituras Relacionadas

As the US and Japan Hike Interest Rates, Which Asset Class is Most at Risk?

This week, global markets face two major events: the Bank of Japan's likely interest rate hike and the US Federal Reserve's FOMC meeting. For risk assets, it is a pivotal and volatile week. In the US, expectations for rate cuts have faded dramatically. May's higher-than-expected CPI and resilient jobs data have shifted the Fed's focus from potential cuts to the possibility of future hikes. New Fed Chair Wash is unlikely to raise rates at this meeting, but any hawkish shift in communication, the dot plot, or the policy statement could lead markets to price in tighter policy, pushing up short-term Treasury yields and strengthening the dollar. High-valuation growth stocks, AI-related assets, and small-cap stocks reliant on cheap funding are most vulnerable to rising rates. In Japan, a 25 basis point hike is almost fully priced in (98.3% probability), which would bring the policy rate to 1%, its highest since 1995. The concern is not the hike itself, but its potential to unwind the massive "carry trade," where investors borrowed low-yielding yen to invest globally. Historically, Japan's rate hikes have coincided with global market stress (2000, 2007, 2024). While this well-telegraphed hike may be digested smoothly, two key factors increase uncertainty: 1) Governor Ueda's absence due to illness, putting communication in the hands of less-familiar deputies, and 2) the Fed meeting occurring just days later, creating potential for a compounded market reaction if both central banks sound hawkish. Asset implications: * **Bonds:** US short-term yields sensitive to Fed signals. Japan's rate hike could pressure its massive US Treasury holdings. * **Currencies:** Dollar likely supported by Fed; Yen's reaction hinges on BoJ's forward guidance. * **Equities:** US growth stocks, small-caps most at risk. Japanese stocks face pressure from a stronger yen. * **Crypto:** Assets like Bitcoin face headwinds from higher rates and tighter liquidity; high-beta altcoins are even more vulnerable. The convergence of these two central bank meetings amplifies market volatility risks, with potential spillovers across asset classes globally.

marsbitHá 8m

As the US and Japan Hike Interest Rates, Which Asset Class is Most at Risk?

marsbitHá 8m

Data Decrypts the BTC Cycle: Three Major Bottom Signals Illuminate Simultaneously, Q4 Could Be a Crucial Turning Point Window?

"Decoding the Bitcoin Cycle: Three Bottom Signals Flash Simultaneously, Is Q4 the Key Turning Point?" The article analyzes Bitcoin's current market position, comparing it to historical cycles. BTC has corrected over 52% from its October 2025 peak of $126,198 to around $59,100 in June 2026. While significant, this drawdown is milder than the 77-86% declines seen in past bear markets. The analysis is framed within Bitcoin's four-year halving cycle. Past cycles show a pattern: prices peak 12-18 months post-halving, bottom 12-14 months after the peak, with lows typically occurring roughly 17 months before the next halving. Following the April 2024 halving and the October 2025 peak, this pattern suggests a potential bottoming window around Q4 2026, ahead of the expected 2028 halving. Three key on-chain metrics are signaling undervaluation: The MVRV Z-Score has dropped near 0.27, approaching historic bottom zones. The market price is only about 9% above the network's average realized price of ~$53,600, a rare low premium. Bitcoin's price recently touched its 200-week moving average (~$62,200), a level that aligned with bottoms in 2015, 2018, and 2020. While US spot Bitcoin ETFs saw record outflows in May/June 2026, indicating retail panic, whale addresses (holding 100+ BTC) reached a yearly high. Entities like MicroStrategy resumed buying, and long-term holders control a near-record 78% of the supply, suggesting accumulation. A major macro overhang was partially removed with a US-Iran ceasefire agreement in mid-June 2026, which eased oil prices and triggered a sharp BTC rally. However, persistent inflation means high-interest rates remain a constraint. The conclusion notes that genuine investment opportunities often arise when confidence is lowest, amidst narratives that "this time is different." While not guaranteeing an immediate bottom, the confluence of cycle timing, undervaluation signals, and shifting macro risks suggests late 2026 may be a critical period for reassessing risk/reward and patient accumulation for long-term believers.

marsbitHá 8m

Data Decrypts the BTC Cycle: Three Major Bottom Signals Illuminate Simultaneously, Q4 Could Be a Crucial Turning Point Window?

marsbitHá 8m

NVIDIA's $20 Billion Bond Issuance Adds Fuel to the Story of Bitcoin Miners Transitioning to AI

Nvidia is reportedly joining the AI bond issuance wave, planning to raise at least $20 billion through a multi-tranche bond offering. This financing move underscores sustained high market demand for AI infrastructure and data centers. It also highlights a growing trend among Bitcoin miners, who are pivoting to repurpose their high-power facilities and existing energy agreements into high-performance computing and AI hosting sites. Companies like HIVE Digital, TeraWulf, Hut 8, and CleanSpark are shifting their business models from solely relying on Bitcoin mining revenue to becoming providers of data center computing power. The primary driver for this transition is the increasingly challenging economics of Bitcoin mining, especially following the April 2024 halving event. The combination of the halving, persistently high mining difficulty, and operational costs has severely squeezed profit margins, leading miners to sell BTC reserves and seek alternative revenue streams. Analysts note that major mining firms are expected to increasingly transform into AI infrastructure suppliers, capitalizing on the ongoing AI construction boom to secure new growth opportunities.

marsbitHá 38m

NVIDIA's $20 Billion Bond Issuance Adds Fuel to the Story of Bitcoin Miners Transitioning to AI

marsbitHá 38m

The Shutdown of Claude Mythos Revealed the True Cost of Renting AI to Me

The sudden shutdown of Claude Mythos this week starkly highlights a critical, often overlooked risk for founders: when your core capability relies entirely on someone else's platform, your fate is not in your own hands. The key question becomes: who truly owns the intelligence your product depends on? For years, the debate around open-source models focused on cost. Now, the evidence is clear: fine-tuned open-source models can achieve frontier-level quality for specific, mission-critical tasks at a fraction of the cost. However, the deeper issue is control. Relying on a third-party API is like renting; it works until the landlord changes the rules, raises the rent, or asks you to leave—as Mythos experienced. The lesson is not to stop using frontier models—they are incredible infrastructure. The goal is ownership. Ownership means starting with a powerful open-source model and shaping it around what makes your company unique: your data, workflows, domain expertise, and definition of "good." Over time, the model becomes less generic and more reflective of your business, creating durable value. The optimistic conclusion is that AI's future doesn't hinge on one superior model. There is no single frontier. The frontier includes proprietary models, models fine-tuned on company-specific knowledge, specialized models for narrow problems, and intelligent routers orchestrating model ensembles. The most interesting development is not models getting smarter, but intelligence becoming increasingly customizable. The winning companies will be those that transform intelligence into a unique, owned asset. Looking ahead, the vision is not one model dominating all, but many teams owning the part of the frontier that matters most to them.

marsbitHá 56m

The Shutdown of Claude Mythos Revealed the True Cost of Renting AI to Me

marsbitHá 56m

Tiger Research: U.S. Strategic Bitcoin Reserve - Should the Market Be Happy or Disappointed?

Tiger Research analyzes the evolution of U.S. legislative efforts regarding a strategic Bitcoin reserve, concluding the market impact is limited in the short term but potentially positive long-term. The core event was a March 2025 executive order by former President Trump, which designated confiscated Bitcoin as a strategic reserve and promised not to sell existing holdings (approx. 190k BTC). As it contained no mandate to purchase new Bitcoin, the market reacted negatively, with prices dropping 5.7%. Legislative history shows a significant retreat from initial ambitions. The 2024 "BITCOIN Act" proposed mandatory purchases of 1 million BTC over five years. Reintroduced in 2025, it stalled due to high fiscal costs, concerns over dollar hegemony, and opposition from the Treasury Secretary. The current frontrunner, the 2026 "American Retirement and Monetary Advancement (ARMA) Act," is a compromise. It lacks any purchase requirement, instead focusing on consolidating existing government-held Bitcoin and legally prohibiting its sale for at least 20 years. While ARMA has higher passage odds due to bipartisan support and no purchase mandate, its immediate market effect is neutral. It eliminates potential government selling pressure but creates no new demand. The long-term significance is that formally establishing Bitcoin as a national reserve asset in law could later reignite debates on mandatory purchases. Therefore, the path to a government buyer is longer than initially priced by the market, but the directional narrative remains intact.

marsbitHá 58m