Domestic Chips Reach a 'Golden Turning Point'.

Over the past six months, the capitalization process of domestic chip manufacturers has accelerated. Recently, in addition to native chip manufacturers, chip subsidiaries of internet giants have also been rumored to be spinning off and going public independently.

In January this year, Baidu's Kunlunxin submitted an application to the Hong Kong Stock Exchange for a mainboard listing in a confidential manner, planning to spin off and go public independently in Hong Kong. Just a few days later, Alibaba's Pangu was also reported to be preparing for an independent listing, a move that Alibaba did not comment on. On the same day, Enflame Technology, which has deep cooperation with Tencent, also had its IPO application accepted by the Shanghai Stock Exchange.

Over the years, the 'chip-building movement' of internet giants has been quite low-key. Why are there suddenly rumors of independent listings at this time? After the chip business moves toward independent commercialization, will internet giants engage in a new round of head-on competition in the chip field?

01. Big Players Forge Ahead on Diverse Paths

With the explosion of AI business, quite a few domestic internet giants are now developing their own chips. Over the past year, JD.com and Meituan, which have been relatively conservative in AI investment, have also started laying out their own AI chip development. However, before the AI wave swept across the industry, only four internet giants—Alibaba, Baidu, ByteDance, and Tencent—truly elevated chip self-development to a core group strategy and continued to increase resource investment.

The chip-building efforts of these four giants can be traced back to 2018 when international trade frictions escalated, and high-end chips and core computing power became bottlenecks, sounding an alarm for domestic internet companies that relied on imported chips to support massive computing power.

On the other hand, cost was also a direct driving factor. Li Yanhong once recalled Baidu's original intention for developing chips: 'When we first started doing search, buying other people's chips was too expensive, costing $10,000 per chip. By making our own, we could do it for 20,000 yuan. So we pushed ourselves to develop the Kunlun chip.'

Baidu was perhaps the fastest among the four giants to take action. Traditional search engines required AI to enhance the efficiency of ad distribution, and during this period, Baidu accumulated chip architecture design capabilities. Baidu's Intelligent Chip and Architecture Department became the precursor to Kunlunxin.

In 2018, when the international trade situation changed, Baidu had already released its first-generation AI chip, 'Kunlun.' This chip, a dedicated AI acceleration chip (ASIC/NPU nature), primarily provided support within Baidu and has been deployed in over 10,000 units across Baidu Search, Xiaodu, and other businesses over the years.

Around 2018, the explosion of cloud computing brought about a massive demand for server chips. From 2018 to 2021, Alibaba Cloud's revenue growth rate once reached as high as 101%, while Tencent Cloud and Huawei Cloud also experienced triple-digit growth rates. However, at the time, Intel and AMD's general-purpose CPUs were not designed for cloud scenarios, resulting in high procurement costs and low computing efficiency. Therefore, Alibaba and Tencent both expressed their intentions to develop their own chips that year to improve computing power adaptability and usage efficiency, thereby reducing the computing power costs of their cloud businesses from the root.

Against this backdrop, Alibaba acquired C-Sky Microsystems, a domestic veteran embedded CPU manufacturer, and integrated it with its in-house chip development team from the Academy for Discovery, Adventure, Momentum and Outlook (DAMO) to establish the semiconductor company 'Pangu.'

From its inception, Pangu established a full-stack and platform-based strategy, not limiting its research and development to a single type of chip. In addition to launching the AI inference acceleration chip 'Hanguang 800,' it also developed the server CPU 'Yitian 710,' which competes with Intel's offerings.

Recently, Alibaba just released its high-end AI chip PPU (Zhenwu 810E), applicable to AI training, AI inference, and autonomous driving. The Zhenwu 810 series chips have long been referred to as 'PPU' in the market, primarily because their technological route is closer to that of NVIDIA's GPUs (Graphics Processing Units). This time, Alibaba did not disclose key information such as the process technology and foundry partner for the Zhenwu 810E. However, according to Alibaba's internal evaluation, the performance of the Zhenwu 810E is already comparable to that of NVIDIA's H20. (The H20 is the main chip used by domestic cloud providers to build AI computing clusters, supporting large-scale model inference and medium-to-small-scale training.)

Targeting the domestic computing power gap, Baidu's chip development in recent years has also shifted to large-scale model training and inference scenarios. In the first half of 2025, Baidu lit up a 30,000-card cluster of Kunlunxin P800, capable of simultaneously supporting the full training of multiple large models with hundreds of billions of parameters like DeepSeek. In 2026 and 2027, Kunlunxin also plans to release the M100 and M300, focusing on inference scenarios and multimodal training, with an emphasis on ultimate cost-effectiveness.

In comparison, Tencent and ByteDance have chosen a more reserved approach of 'scenario-specific + self-sufficiency' for their self-developed chips.

Tencent officially established the Penglai Laboratory in 2020, focusing on hardware acceleration for cloud computing and audio-video scenarios. It has successively launched three major product lines: Canghai (video transcoding), Zixiao (AI inference), and Xuanling (SmartNIC/DPU), all of which have been delivered for internal use at Tencent.

ByteDance's chip layout is even more focused on the vertical field of video. In 2023, Volcano Engine Video Cloud officially announced that its self-developed video codec chip had started internal application, primarily serving video businesses such as Douyin and Xigua Video, and was also made available to enterprise customers through Volcano Engine Video Cloud. Over the past two years, media reports have mentioned that ByteDance is also advancing its self-developed AI chips and server chips, but ByteDance has remained very low-key about this. Since there are no plans for external sales, it has not proactively disclosed the progress of its self-developed chips.

02. What Supports Independent Listings?

From their product routes, it can be felt that Alibaba and Baidu hope that their self-developed chips will not only be used internally but also enable their chip teams to become industry suppliers through B2B sales.

This is indeed the case from a business operation perspective. According to media reports, around the 2021 Cloud Town Conference, Alibaba considered spinning off Pangu for independent financing to allow it to obtain more development funds and a more independent market position as a chip supplier. Also in that year, Baidu began operating Kunlunxin as an independent entity, with Baidu's Chief Chip Architect Ouyang Jian serving as CEO and completing its first round of financing.

Data from Tianyancha shows that as of July 2025, Kunlunxin has completed seven rounds of financing, with investors including BYD, Merchant Ventures, Zhongguancun Science City, Industrial and Commercial Bank of China Investments, and the Beijing Artificial Intelligence Industry Investment Fund. Its post-investment valuation is approximately $2.97 billion (about 21 billion yuan). Baidu currently holds a 59.45% stake in Kunlunxin and will maintain control after the spin-off. Multiple institutions, including China International Capital Corporation and Haitong International, have given positive assessments from different perspectives, pushing Kunlunxin's potential valuation expectations to the $50 billion range.

In comparison, on January 23, J.P. Morgan's China Securities Research Team stated in a research report that, referencing peer valuations, Pangu's potential valuation may range between $25 billion and $62 billion, accounting for approximately 6% to 14% of Alibaba's current market value.

As independent entities participating in market competition, Kunlunxin and Pangu face significant challenges, the first of which is the direct competition and ecological barriers from giants such as NVIDIA and Huawei.

Data from international market research firm IDC shows that in the first half of 2025, the Chinese accelerated (i.e., AI chip) server market reached $16 billion in size, with over 1.9 million units shipped. Among them, NVIDIA accounted for approximately 62% of the market share, while domestic AI chips accounted for about 35%. According to a survey by Bernstein Data, in the domestic chip market, Huawei's Ascend is expected to expand its market share to 28% in 2025, making it the absolute domestic leader.

Competing with giants, emerging manufacturers must not only present mature chip products but also offer sufficiently user-friendly software stacks and toolchains. NVIDIA's dominance in the chip market over the years has relied on developers' default language, tools, and ecological habits.

Neil Shah, Vice President of Counterpoint Research, a technology market research firm, once mentioned that in addition to hardware, NVIDIA's dominance lies in its powerful ecosystem, especially CUDA (a complete software stack from underlying drivers to upper-layer applications). The time, code, and project experience that developers have invested in CUDA are sunk costs that cannot be migrated. Switching to another ecosystem means these investments may become worthless, so developers tend to continue staying within the CUDA system.

Furthermore, the stability and scale of the supply chain also determine customer experience. Long-term cooperation with enterprises requires ensuring chip mass production yields, stable supply from wafer foundries, and establishing a complete upstream-downstream chain including packaging, testing, and hardware support.

Judging from the currently disclosed customer base of Kunlunxin and Pangu, both companies have achieved a certain scale of commercialization. Kunlunxin's customers include China Mobile, China Southern Power Grid, China Merchants Bank, and other government, enterprise, and large state-owned enterprise clients. In August 2025, Kunlunxin also won a billion-dollar order in China Mobile's centralized procurement project. Pangu's recently updated Zhenwu PPU chip is rumored to have shipped in the hundreds of thousands of units, serving over 400 customers.

Media reports have mentioned that Alibaba recently proposed the concept of 'Tongyun Ge,' referring to the trinity development of Tongyi Laboratory, Alibaba Cloud, and Pangu, i.e., large models, cloud computing, and chips. It seems that Pangu's future market advantage lies in the synergy of 'cloud + chip + AI,' forming a complete AI infrastructure system and reducing customer computing power deployment costs and migration barriers. In comparison, Kunlunxin leverages its differentiated advantage of in-depth cultivation in large-scale model training and inference to continuously expand its government and enterprise customer base. The competition between the two in the B2B sector has also become their core competitiveness for impact (chōngjī, meaning ' impact ' or 'making a push for') listings.

03. IPO Boom Among Chip Companies

The surge in demand for large-scale model training and inference has made computing power an increasingly essential resource in recent years. The continuous tightening of overseas chip export controls has exactly (qiàhǎo, meaning 'coincidentally' or 'just at the right time') created a window of opportunity for domestic GPUs to replace foreign counterparts, drawing increasing attention to domestic chip manufacturers.

'Quijie Business' has noticed that over the past year, not only Kunlunxin and Pangu have been preparing for IPOs; multiple domestic chip companies are also rushing to the capital market. To date, more than ten domestic chip companies are in the process of going public or have already completed their listings. Examples include Moore Threads, MetaX, Biren Technology, and Iluvatar CoreX, known as the 'Four Little Dragons of GPUs.' Recently, Enflame Technology, which has deep cooperation with Tencent, was also reported to be undergoing IPO counseling.

Currently, listed chip companies have performed outstandingly in terms of stock prices. MetaX and Moore Threads successively listed on the STAR Market in December 2025. Moore Threads' stock price soared by over 468% on its debut, with a market capitalization exceeding HK$250 billion. MetaX's stock price surged by 568% on its first trading day, and at one point, investors holding one lot stood to gain nearly 400,000 yuan in floating profits, directly setting a new record for single-lot profits on the first trading day of A-shares in the past decade. Biren Technology, which just listed on the Hong Kong Stock Exchange in January this year, saw its stock price jump by 76% on its debut, with a market capitalization also exceeding HK$100 billion.

Although none of these chip companies have yet to turn a profit, it has not seemed to affect the capital market's enthusiasm for them. The explosion of the AI industry has opened up a hundred-billion-yuan market space, while policy easing for hard-tech company listings and industrial support have also filled the market with anticipation for these companies' technological breakthroughs and profit realization.

To date, the highest-valued chip design company in China is Cambricon, which went public in 2020, with a total market capitalization of 457.296 billion yuan. Cambricon achieved its first profit in 2025, with expected revenue ranging from 6 billion to 7 billion yuan, representing a year-on-year increase of 410.87% to 496.02%. However, in terms of revenue scale, it still has a considerable gap compared to global giants like NVIDIA (NVIDIA's revenue in Q3 2025 was $57.006 billion).

Bo Wenxi, Chief Economist of the China Enterprise Capital Alliance in China, believes that 2025 can be considered the first year of domestic GPU IPOs, with the core catalyst being the implementation of the '1+6' reform policies on the STAR Market, which explicitly allow unprofitable hard-tech companies to go public. This has opened up a capital channel for chip companies with high R&D investment but no profits yet. Moore Threads took only 88 days from acceptance to approval, while MetaX took 116 days, demonstrating accelerated A-share review efficiency. In 2025, the Hong Kong Stock Exchange and the Securities and Futures Commission of Hong Kong opened a dedicated service channel for eligible technology companies, allowing specialized technology companies to submit applications confidentially, providing convenience for chip manufacturers. This chip listing boom is expected to continue for the next one to two years.

However, going public is just a new starting point. The chip industry lacks a 'one-size-fits-all' moat. As the capital market presses the accelerator for 'China's chips,' what we anticipate is not just the fluctuations in the market capitalization of two companies but the true formation of an autonomous, controllable, and diversified chip industry ecosystem.