HomeArticle

Alibaba's Pingtouge has launched a listing plan and has deployed full-stack AI chips.

邱晓芬2026-01-22 18:37
The self-developed PPU by T-Head has become one of the main chips in China's newly added AI computing power market.

On January 22nd, people close to the market said that Alibaba Group has decided to support the future independent listing of its chip company, T-Head. T-Head is a wholly-owned subsidiary of Alibaba, established in 2018.

Since its establishment eight years ago, the company has launched a number of chips with top-notch performance in the industry in fields such as computing, storage, and networking. It is reported that T-Head's self-developed PPU has become one of the main chips in China's newly added AI computing power market.

I. An Early Explorer of Self-Developed Chips among Internet Companies

Chips have always been considered one of the toughest nuts to crack in the technology field. Before the establishment of T-Head, Alibaba's "resume" in hardware R & D was almost blank, and there was no precedent in China for an internet company to make chips. The difficulty of T-Head's chip R & D was no less than Alibaba's investment in cloud computing.

Although the reality was harsh, in this game of technology and capital, Alibaba's strategic determination was no less than its long - term investment in Alibaba Cloud.

After the company was established, the internal chip R & D route and model were very clear, that is, to focus on a series of product lines around the cloud data center scenario. This was quite different from the model of independent chip companies developing standard - product chips at that time, but this model also made T-Head the biggest dark horse in the chip industry chain.

Traditional semiconductor manufacturers need to first conduct customer surveys and collect requirements before entering the long - term chip definition and design stage, which consumes a lot of time in the early stage of chip R & D. Backed by Alibaba Cloud, T-Head can better understand customer scenarios and the demand for cloud computing power, which greatly shortens the cycle from chip design to application implementation.

In 2019, in its first year of establishment, T-Head made a tentative attempt and launched its first chip, the AI inference chip, Hanguang 800. This was a chip deeply customized for scenarios, using a self - developed architecture. Its inference performance reached 78,563 IPS, capable of processing 78,000 pictures per second, creating two records in performance and energy - efficiency ratio in the same category of chips at that time. This chip was gradually applied in the main search scenario of Taobao during the Double 11 shopping festival.

Meanwhile, T-Head also launched the R & D of a more difficult general - purpose CPU chip. At the 2021 Yunqi Conference, T-Head released Alibaba's first general - purpose server chip, Yitian 710, whose performance exceeded the industry benchmark of the same period by 20% and the energy - efficiency ratio increased by more than 50%.

It is worth noting that the general - purpose CPU is the brain of the computing system, and its design is one of the most difficult mountains to cross in the semiconductor industry. Even today, there are only a few enterprises that master this technical ability. Alibaba is the first Chinese internet company in the industry with the ability to R & D and design CPUs.

Currently, Yitian 710 has been widely applied in fields such as video encoding and decoding, high - performance computing, and gaming through Alibaba Cloud services. Across the ocean, Amazon's EC2 instances based on its self - developed Graviton chips have also become the main CPU computing power of AWS. Alibaba and Amazon almost simultaneously proved that the chip - making model of in - depth collaboration between self - developed chips and cloud service scenarios is feasible.

Today, this model has gradually been emulated by other large domestic internet companies.

II. Betting on the AI Wave and Having a Full - Stack of Chip Products

Around 2020, during the period when the previous round of AI deep - learning bubble burst and before the current AI wave, domestic GPU companies such as Brion Technologies, Moore Threads, and Muxi Co., Ltd. were successively established, setting off a wave of domestic self - developed chips. Few people know that T-Head also secretly launched the R & D of a general - purpose GPU chip at the same time, which is the so - called "PPU" in the rumors.

Parameters of T-Head's PPU

It is reported that as early as the end of 2022 and the beginning of 2023, the PPU completed R & D and scenario verification. However, different from the high - profile "Four Little Dragons of AI Chips", T-Head has never publicly disclosed the progress of this chip.

In September 2025, the US media, The Information, reported that the performance of T-Head's first - generation PPU could rival that of NVIDIA's best - selling H20, and the upgraded PPU was even more powerful than NVIDIA's A100. According to The Information, T-Head's PPU has been used in some small - scale large - model trainings of Alibaba, leading other domestic chip companies.

In September 2025, two frames showing the specific performance indicators of T-Head's PPU flashed in the news broadcast of China Central Television. The strength of the PPU gradually emerged: the PPU has a 96G HBM2e video memory, an inter - chip interconnection bandwidth of up to 700GB/s, a PCIe 5.0×16 interface, and a power consumption of only 400W. In these key parameters, it has completely exceeded NVIDIA's A800 and mainstream domestic GPUs, and its overall performance is comparable to that of the H20.

If the parameter specifications prove its strong theoretical strength, then the ease of use of T-Head is the key weapon for it to be widely accepted by the market. Due to its excellent and stable performance and outstanding ease of use, T-Head's PPU chip has a good reputation in the industry.

In the past year, this chip has been in short supply. According to industry insiders, in 2025, T-Head's PPU has become one of the domestic self - developed GPUs with the highest shipment volume.

In addition to the spotlighted computing power chips, T-Head is also constantly expanding its product line. In 2023, T-Head launched its first storage chip, the SSD controller chip, Zhenyue 510, which can be compared with the flagship chip of Samsung, the industry benchmark, and can meet the requirements of low latency and high bandwidth for AI training and inference.

It is understood that T-Head will also launch a new network chip this year, and it will become one of the few domestic companies covering the full - stack of data center chips. It is worth noting that in the field of edge - side chips, T-Head has also launched the Yuzhen IoT chip, with a shipment volume of hundreds of millions.

T-Head has always been known for its low - key style in the industry. Its full - stack chip capabilities are the most underestimated part of Alibaba in the current AI wave. Previously, the infrastructure of Alibaba Cloud and the most powerful open - source model, Qianwen, were already well - known. Now, with the mysterious veil of T-Head lifted, it officially announces that the last piece of the puzzle in Alibaba's AI full - stack layout has been completed.

III. The Listing Wave of Domestic GPU Companies: What's the Valuation of T-Head?

In the past year, the demand for AI computing power from large models has continued to grow, and the market share of domestic AI chips has also increased significantly. As the market pattern has been subtly changing, domestic AI chip companies have also witnessed a wave of listings.

At the end of August 2025, as an extremely rare AI chip company in the A - share market, Cambricon, which had a revenue of only 4.607 billion yuan in the first half of 2025, saw its intraday stock price exceed that of Kweichow Moutai, becoming the stock with the highest price in the A - share market, attracting wide attention from the market.

Currently, Cambricon's total market value is stable at around 600 billion yuan.

The positive feedback from the capital market has opened up the imagination space for domestic AI chips. From the end of 2025 to the beginning of 2026, many domestic GPU companies were listed intensively, adding fuel to the already hot domestic GPU chip industry.

Among them, Moore Threads set the fastest record for passing the review of the STAR Market IPO in 88 days and was listed on December 5th, 2025. Its current market value is around 300 billion yuan. Public information shows that in the first three quarters of 2025, Moore Threads had a revenue of 780 million yuan.

On December 17th, 2025, Muxi Co., Ltd. was listed on the STAR Market, with a nearly seven - fold increase on the first day. Its current market value exceeds 240 billion yuan. Muxi Co., Ltd. had a revenue of 1.236 billion yuan in the first three quarters of 2025.

On January 2nd, 2026, Brion Technologies was listed on the Hong Kong Stock Exchange, becoming the "first domestic GPU stock in the Hong Kong stock market". Its current market value exceeds 70 billion yuan. It is worth noting that Brion Technologies only had a revenue of 58.903 million yuan in the first half of 2025.

After the listing of Brion Technologies, Daysci Technologies also followed suit and was listed on the Hong Kong Stock Exchange on January 8th, 2026. Its current market value is about 40 billion yuan. The company had a revenue of 324 million yuan in the first half of 2025.

Enflame Technology, one of the "Four Little Dragons of AI Chip Companies", has also completed IPO counseling and is about to sprint for the STAR Market. According to the prospectus, the company had a revenue of 2.83 billion yuan in the first half of 2025.

Overall, the total market value of the five domestic AI chip companies, Cambricon, Moore Threads, Muxi Co., Ltd., Brion Technologies, and Daysci Technologies, is about 1.25 trillion yuan, which is about 4% of NVIDIA's market value.

Compared with the above - mentioned companies, T-Head has a more comprehensive chip product family, which can provide enterprises with richer solutions and meet more diverse computing power requirements. According to industry insiders, the market share of T-Head's PPU has far exceeded that of the chips of the above - mentioned companies. If T-Head is successfully listed in the future, it will surely trigger a re - evaluation of the value of domestic chip companies by the capital market.