Jiliu Technology Accelerates Toward Hong Kong Stock Market

Author | Yezi

Click to listen to the podcast of this article

China’s leading AI computing infrastructure provider is set to go public.

Recently, Jiliu Technology formally submitted its prospectus to the Hong Kong Stock Exchange, positioning itself as Hong Kong’s pioneering “AI infrastructure stock.”

Founded just three years ago, the company boasts an impressive profile: a unicorn valued at over RMB 9 billion, a founding team of academic stars from Tsinghua and Peking universities, and RMB 2.2 billion raised across three years of financing.

Yet, beneath this high-tech façade lies a business engaged in the demanding, labor-intensive infrastructure work of the AI sector. Rather than manufacturing AI chips or developing large-scale models, Jiliu Technology specializes in assembling tens of thousands of high-cost GPU cards into a seamless “supercomputer” for large model companies to utilize in training.

Its influential backers are equally remarkable: Zhipu AI is not only an early customer but also its largest external shareholder, while SenseTime, though not holding direct shares, supports half of Jiliu Technology’s business through its substantial orders as a core customer.

Amid the global AI boom, how did this company, led by three post-90s entrepreneurs, carve out a niche in the fiercely competitive computing power market dominated by industry giants?

Tsinghua-Peking Academic Stars Pioneer Computing Infrastructure: Three Post-90s Entrepreneurs Build a RMB 9 Billion Valuation

Jiliu Technology’s journey began with a reunion of two former classmates from Hengshui High School in late 2022.

At that time, ChatGPT emerged, igniting a large model startup frenzy across Zhongguancun. While the focus was on who could create the next ChatGPT, Hu Xiaohe and Xie Wenqi identified a more subtle yet critical challenge: the computing power bottleneck.

The industry faced an awkward reality: procuring GPUs did not guarantee full utilization of computing power. Due to communication limitations, the computing speed of clusters could not infinitely scale with the number of GPUs.

To illustrate, imagine one person building a house in 10 days. Bringing in 100 people doesn’t mean it will be done in a fraction of the time if 90 can’t even fit on the construction site—this is akin to network congestion.

Without precise layout and intelligent scheduling, tens of thousands of expensive graphics cards could become little more than piles of overheated scrap metal, with actual computing power utilization possibly falling below 30%.

Recognizing this challenge, the seed for their entrepreneurial venture was planted. In February 2023, Jiliu Technology officially launched operations from a complimentary workspace in Yuandian Community, Dongsheng Town, Haidian District, Beijing, and later relocated to Shanghai in 2026. They positioned themselves as a full-stack, autonomous computing infrastructure builder.

The core executive team, all post-90s, boasts a strong technical background from Tsinghua and Peking universities.

Hu Xiaohe, the 33-year-old founder and CEO, is a quintessential tech enthusiast. He holds a bachelor’s degree and a Ph.D. from Tsinghua’s Department of Automation, was a visiting scholar at UC Berkeley under SDN (Software-Defined Networking) pioneer Academician Scott Shenker, and has over a decade of experience at Tsinghua’s Cybersecurity Lab.

Xie Wenqi, the 33-year-old co-founder and co-CEO, graduated from Peking University with a bachelor’s degree and later earned a master’s in finance from Tsinghua. His previous experience in PE investments provides him with a deep understanding of both capital and industry dynamics. Wang Xuyang, the 35-year-old co-chairman, rounds out the leadership team, with over 60% of the Tsinghua-educated group holding master’s or doctoral degrees.

Jiliu Technology can be likened to the “China State Construction Third Engineering Bureau of the AI world.” They don’t manufacture bricks or purchase land but serve as the general contractor for computing power clusters. Their core product is the self-developed Galaxy computing power cluster system: at the hardware level, it features the self-developed Mercury high-performance integrated AI network system to alleviate network congestion; at the software level, it incorporates the Venus computing power operating system for cluster scheduling, fault tolerance, and monitoring to ensure no computing power goes to waste.

Facing integrated giants like NVIDIA and Huawei, Jiliu Technology differentiates itself through independence and decoupling. By separating hardware and software, it can flexibly adapt to chips from various vendors, preventing customers from being locked into a single supplier.

Leveraging this expertise, Jiliu Technology has boosted large models’ computing power utilization by 30% while maintaining extremely low hardware failure rates.

According to Frost & Sullivan data, by 2025 revenue, Jiliu Technology ranks as China’s ninth-largest AI computing power cluster provider, trailing behind giants like Alibaba Cloud, Huawei Cloud, and Tencent Cloud; but among independent AI computing power cluster providers, it holds the top spot.

Clients Become Shareholders: Zhipu and SenseTime Provide Vital Support

Jiliu Technology’s rapid ascent over just three years is closely tied to its strategic capital approach.

In its early stages, many investors were skeptical about the commercial value of communication optimization, believing software solutions were valuable but difficult to price. The first to show confidence was MiraclePlus, which invested RMB 2.055 million in the seed round.

However, it was Zhipu AI that truly propelled Jiliu Technology into the industry spotlight.

In the latter half of 2023, domestic large models entered a competitive phase focused on thousand-card and ten-thousand-card training. Jiliu Technology delivered its first 2,000-card cluster project for Zhipu AI using its self-developed networking solution.

Xie Wenqi recalled that a five-person team camped in the client’s data center for over a month, debugging around the clock and sleeping in the server room when exhausted. This industrial-strength, all-out delivery completely won over Zhipu.

In December 2023, Zhipu, along with MiraclePlus and other institutions, injected RMB 22.7 million in angel round funding into Jiliu Technology. This deep binding of “client as shareholder” provided not just financial support but the strongest industry endorsement.

Subsequently, Zhipu transferred its shares to its large model ecosystem fund, Starlink Capital. Starlink not only took over the existing shares but also independently participated in four subsequent financing rounds. Before the IPO, Starlink Capital held a 7.7% stake in Jiliu Technology, firmly establishing itself as the largest external shareholder, with a paper value of RMB 705 million.

Another major supporter is SenseTime. During a GPU cluster project delivery for SenseTime, a minor incident occurred. When equipment failed, Hu Xiaohe immediately replaced nearly a hundred switches on the spot and vowed, “Solve the problem in half an hour, or refund the full amount for every minute of delay!”

The project was ultimately accepted smoothly. With such hardcore capabilities, orders from major clients like SenseTime continued to pour in.

The prospectus reveals that in 2023 and 2024, revenue from the top two clients once accounted for a staggering 96.5% and 75.1%; even by 2025, the top five clients still contributed 56.6% of revenue.

With orders and endorsements from these giant clients, Jiliu Technology’s financing accelerated. Over three years, the company completed 11 financing rounds, raising over RMB 2.2 billion in total, averaging one round every three months.

Top-tier VCs like Lightspeed China Partners, Guofang Innovation, Shanghai Guotou Futeng Capital, and Zhangjiang Haocheng, along with state-owned capital from multiple regions, flocked to invest. Just eight days before submitting its Hong Kong Stock Exchange application, the company completed a RMB 1.16 billion Series D round, with its valuation soaring to RMB 9.16 billion.

Revenue Surges 16-Fold in Three Years, But Heavy Asset Risks Loom

As a “shovel seller” in the AI era, Jiliu Technology’s revenue growth is impressive.

The prospectus shows that in 2023, revenue was just RMB 3.18 million; by 2024, it skyrocketed to RMB 325 million; and in 2025, it further surged to RMB 520 million.

Revenue multiplied 16-fold in three years, with a compound annual growth rate of 304.5%.

More impressively, calculated under non-IFRS standards, the company has been profitable for two consecutive years, with adjusted net profits of RMB 20.29 million and RMB 31.12 million in 2024 and 2025, respectively.

However, behind this glossy financial report lie the hardships and risks of working for industry giants.

Jiliu Technology’s current core revenue comes from AI computing power cluster products, meaning designing, networking, and delivering computing power centers for clients. This is essentially an engineering general contracting model. While business booms during industry upturns, the project-based nature of the business means thin profit margins.

In 2024 and 2025, the gross margins for this contracting business were only 18.2% and 16.8%, respectively. After deducting high sales, administrative, and R&D expenses of 13.2%, the net profit margin landing in pockets was just around 6%.

This is a common dilemma for many AI infrastructure companies: standing at the forefront of technology, yet earning money like a high-tech construction crew.

To break free from thin margins, Jiliu Technology began transforming, aiming to upgrade from a contractor to a property manager, or even a real estate developer.

On one hand, the company launched AI computing power cluster operation services. After project completion, it continues to handle maintenance and scheduling for clients. This property management business had gross margins of 56.1% and 47.7% in 2024 and 2025, far exceeding delivery services.

On the other hand, Jiliu Technology has grander ambitions: it started purchasing equipment and building computing power clusters with its own money, offering computing power rental services. While this computing power developer model offers more imagination, it also brings heavy asset pressures.

The cost of transformation is evident on the balance sheet. To support the expansion of computing power services, from 2024 to 2025, the company’s machinery and equipment soared from RMB 26 million to RMB 173 million. Consequently, debt skyrocketed from RMB 210 million to RMB 974 million.

Expansion also strained cash flow. In 2025, inventory turnover days rose to 131.7, while accounts receivable turnover days lengthened to 64.1. With longer project cycles and slower payments, the company relied heavily on massive financing cash flows to sustain operations. From 2024 to 2025, both operating and investing cash flows were net outflows.

Moreover, Jiliu Technology’s touted independence faces ultimate commercial logic questions after transforming into a computing power operator. Truly high-quality large model platforms, like Zhipu and SenseTime, consider computing power core infrastructure and would never rely on external leases long-term; those willing to rent computing power continuously are often price-sensitive, demand-scattered SMEs.

References

“China’s Largest AI Contractor Goes Public,” Jinjiao Finance;

“Zhipu Invests in Its First IPO,” Daily IPO;

“Zhipu to Get Its First IPO! AI Computing Power Cluster Provider Jiliu Technology Valued Over RMB 9 Billion,” Direct IPO;

“Tsinghua Postdoc Aims to Be ‘Computing Power Contractor’: Shanghai Jiliu Technology Strikes Hong Kong ‘AI Infrastructure First Stock,’” Shanghai Observer;

“AI Computing Power Dark Horse Rushes to IPO Just 3 Years After Founding: Jiliu Technology’s 3 Executive Directors All Post-90s,” Ruichi Finance;

“Another AI Unicorn Rushes to IPO! Led by Tsinghua-Peking Academic Stars,” Enterprise Listing Law & Business Research;

“Zhipu’s First IPO Investment Arrives: 3 Post-90s Build ‘AI Computing Power Architecture’ Business, Valued Over RMB 9 Billion,” Cyzone;

“Jiliu Technology Submits Hong Kong Stock Exchange Application: The Light and Shadow of an Independent AI Computing Power ‘Little Giant,’” 21st Century Business Herald;

THE END

Copyright and Disclaimer

1. Content Copyright: Except for quoted public data, policies, and cases, this article is original. Professional data comes from authorized databases and government websites; cases are compiled from real events.

2. Image Licensing: Some images are proprietary or officially licensed, while others are AI-generated; for web images with unclear copyright, rights remain with the original authors, and infringement claims will be removed.

3. Republication Guidelines: Unauthorized republication is prohibited; republication must retain full attribution and author.

4. Liability Disclaimer: This article is a commercial figure observation and industry commentary based on public information, for reference only, not professional advice. Risks from use are borne by the user. Xinzhikongneng reserves final interpretation rights.