Over the past two decades, China's internet has been doing one thing: packing as many capabilities as possible into a super company.
Tencent does social networking, then does payments, gaming, cloud computing; Alibaba does e-commerce, and also logistics, finance, and big entertainment on the side; Baidu does search, while expanding into maps, autonomous driving, and scaling large models.
The logic of that era was to claim the mountain and become king. Whoever had the most portals and users was the winner.
But recently, a few seemingly isolated news items, when put together, indicate a complete change in the wind direction.
Baidu plans to spin off Kunlunxin for a Hong Kong IPO, targeting a valuation of about $50 billion.
What does that mean? Baidu's current total market capitalization is only around $36 billion. The subsidiary is nearly 40% more valuable than the parent company.
Around the same time, Alibaba's Pingtouge also reportedly has plans for an independent listing.
More intriguing is a report from The Information showing that Tencent has become a significant customer of Kunlunxin.
Meanwhile, according to reports, Kunlunxin has proposed an extremely tough condition during its roadshow: to subscribe to IPO shares? First commit to purchasing chips, with an amount 3 to 7 times the subscription value. "Be a customer first if you want to be a shareholder."
Many see this as merely a capital game. But in my view, this points to a change in the underlying logic of China's internet.
In the past, giants built moats through closure and monopoly; today, giants must scale up through openness and specialization.
I. Chips: From a Money-Burning Black Hole to a Cash Cow
In the past, why did giants make chips?
The answer is simple: to save money.
Baidu needed to train search algorithms and large models; self-developed chips were cheaper than buying Nvidia's. Alibaba needed to support its massive cloud infrastructure; making its own chips could lower hardware costs.
Back then, the chip department, managed within the group's R&D, was a pure cost center. It spent money but didn't directly generate revenue. Keeping it internal for self-sufficiency made perfect sense.
Today, that calculation has been redone.
The emergence of Agents has made the whole industry see a harsh reality: the most terrifying consumption in AI isn't in large model R&D, but in high-frequency inference.
Every AI response, every Agent task execution, every piece of code generation is voraciously consuming tokens in the background. The foundation of tokens is GPUs, inference chips, networks, and data centers.
When user volume on the application side crosses a tipping point, API calls turn into real cash flow. So, a hardware R&D department that was once tucked away internally and inconspicuous suddenly possesses an independent and compelling business model: the chip itself becomes a highly profitable business.
According to public information, Kunlunxin P800 has completed large-scale validation, has delivered multiple 10,000-card clusters since 2025, and completed the training of Wenxin 5.1 on fully domestic clusters. Its customer list has rapidly expanded from Baidu's own use to include China Mobile, Geely, China Southern Power Grid, China Merchants Bank, and Tencent.
When chips transform from a cost-incurring department into a revenue-generating business, spinning it off as an independent entity changes from a financial tactic to a strategic necessity.
II. Tencent Buying Baidu Chips is More Significant Than the IPO Itself
In The Information's report, the appearance of Tencent is the most dramatic detail.
Over the past two decades, China's internet giants have almost had zero interaction at the infrastructure level. Alibaba's cloud would never sell to Tencent, and Tencent's technology would never use Baidu's underlying infrastructure. Every company was expensively reinventing its own wheel.
Today, Tencent has started procuring Baidu's Kunlunxin chips. This marks the first time China's AI industry is beginning to form a truly mature division of labor.
The infrastructure of the AI era is too expensive. Chip R&D cycles are long, capital investment is heavy, and talent barriers are high. If a giant makes chips only for its own ecosystem, economies of scale can never be achieved, and the cost per chip cannot be amortized.
In the end, they found that trying to make everything themselves meant nothing was made well.
Thus, the industry is beginning to shift. Tencent's procurement of Kunlunxin means top players are starting to accept one thing: the best AI ecosystem does not require every component to be self-contained.
This is similar to Apple and Samsung in the mobile phone industry. While they fight tooth and nail in the end-market, the core OLED screens for iPhones still rely on Samsung's factories.
Procurement by a competitor is the highest level of endorsement. When Tencent is willing to base part of its computing power infrastructure on Kunlunxin, it shows that domestic AI chips have passed the most rigorous practical tests, and even rivals find them adequate and secure.
III. The Capital Market is Re-pricing Computing Power
Kunlunxin was founded as early as 2011. Why is it accelerating its IPO push specifically in 2026?
The answer isn't really about Baidu; it's about the shift in capital market targets.
Five years ago, AI chips were just R&D expenses in giants' financial reports. Without a large enough commercial throughput, the capital market would never give hardware companies high valuations.
Today, it's completely different. Nvidia, Samsung, and SK have shattered market cap myths one after another, setting a new anchor point for the global capital market: the biggest winners in the AI era are often those who sell the shovels.
Especially with the explosion of Agent and multimodal applications this year, inference demand is growing exponentially. The entire investment community is starting to re-evaluate the value of AI companies. In the past, people talked about parameters and benchmark scores; today, they only calculate token cost, inference efficiency, and data center utilization.
AI infrastructure, for the first time, possesses an extremely clear and viable business return model.
Kunlunxin didn't just decide to go public today; it's that only today has the capital market finally understood and become willing to assign a sufficiently high price to domestic AI chips.
In fact, this has become a collective capital expedition. Alibaba's Pingtouge has initiated an independent listing; Cambricon has completed capital verification in the A-share market; Biren, Moore Threads, and Moore Threads have all gone public, and companies like Birentech are also entering new IPO phases.
Over the past decade, the toughest challenge for domestic chip companies was proving they could make chips and whether anyone would use them.
Today, that stage has passed. What they need to prove is another thing: who, besides Nvidia, can become the domestic chip foundation for China's AI era.
IV. The Global Giants' War of Certainty
Pulling the perspective to the global level, you'll find everyone is doing exactly the same thing.
Why is OpenAI making its own chips? Essentially, the monthly active user call volume is too large, and every call burns real money. Even if self-developed chips only improve performance per watt by 20%, it could save billions of dollars a year. More importantly, they understand their fate cannot be tied to a single supplier like Nvidia.
Look at other giants. Google's TPU has iterated to the 8th generation, forming the world's most mature self-developed chip system; Amazon has Trainium and Graviton; Microsoft has Maia; Meta has MTIA.
All the world's leading AI players have extended their hands to the most fundamental hardware.
The reasoning is extremely direct. Inference cost is the largest single expense for AI companies. Only those who can reduce hardware costs can establish a viable business model. Moreover, when you control both the large model architecture and the chip architecture, the synergistic optimization of software and hardware you can achieve is a barrier that buying external, general-purpose GPUs can never reach.
Kunlunxin's IPO is just a signal, declaring that China's AI infrastructure is officially moving from the giants' backyards to the public market.
V. The Dimension of Competition Sinks to the Deep End
Over the past two years, everyone has been discussing GPT, Claude, DeepSeek. People habitually believed that large model parameters and benchmark scores were the entirety of AI competition.
But now, models are becoming more like operating systems. They are crucial, but they are no longer the sole determinant of victory or defeat.
What truly determines how much money an AI company can make in the future and how long it can survive are those more fundamental, more mundane hard metrics:
Who can minimize the cost per token to the extreme? Whose inference cluster is the most efficient? Who can possess a continuous, uninterrupted supply of computing power?
2023 was the war of models, where everyone compared who was smarter; 2024 is the war of applications, where everyone compares who is more useful.
By 2025-2026, the war has sunk to the deepest layer. Global AI competition has officially entered the attrition warfare of infrastructure.
Words [Off the Page]:
In the past, we always tended to view China's internet giants through the lens of empires and monopolies.
They were like black holes that devoured everything, extending their tentacles into every corner, trying to complete the entire commercial loop within their high walls. Those were two decades where the giants grew larger and their ecosystems became increasingly closed.
But the spin-offs of Kunlunxin and Pingtouge, coupled with Tencent's order, have personally drawn a period on that era.
Behind this is not the decline of Baidu or Alibaba, but an inevitable destiny: the AI industry chain is simply too vast, too vast for any single super company to swallow the complete infrastructure alone.
Models iterate daily, applications reshuffle every three months, but the underlying chips, networks, and data centers, once built, become the industrial foundation for a decade.
On the surface today, it seems like Baidu and Alibaba are "breaking up the house," but in reality, it's China's internet undergoing a decoupling process that has lasted twenty years.
In the internet era, giants got bigger and bigger. In the AI era, giants are actually getting smaller and smaller.
They are beginning to release their capabilities to form a larger industry.
This article is from WeChat public account: Off the Page , Author: Huahua
