{"text":[[{"start":7.6,"text":"This article only represents the author's own views."}],[{"start":10.8,"text":"The rapid rise of AI agents has made computing power infrastructure one of the hottest investment themes in capital markets lately. That tide is washing some names from China’s golden internet era back to center stage for encore performances."}],[{"start":26.3,"text":"One of those is none other than Yao Xin, founder of PPTV, which was once at the cutting edge of China’s online video revolution. Now, Yao has now returned for his encore with a Hong Kong IPO application for his similarly named PPLabs Technology Ltd. From his earliest days building a content delivery network two decades ago, to his latest foray integrating decentralized GPU resources to build an AI computing power network, Yao Xin has always focused on the infrastructure at the base of the internet world."}],[{"start":59.650000000000006,"text":"PPLabs does not train models, nor does it directly provide AI applications. Instead, it integrates decentralized computing power resources across the globe to provide edge cloud and AI cloud services to enterprises and developers. The company's computing power network covered over 1,340 cities and counties globally at the end of last year, with more than 4,600 compute nodes."}],[{"start":84.25,"text":"The platform had over 574,000 registered developers at the end of April this year. According to third-party market data in its preliminary prospectus, the company is the largest independent edge cloud computing service provider in China, and is also the country’s largest independent AI cloud computing service provider."}],[{"start":104.55,"text":"Yao Xin founded his original PPTV online video platform in 2005, placing him in the generation of China's earliest internet entrepreneurs. After leaving PPTV, he became a partner at Lanchi Ventures before founding PPLabs in 2018. If PPTV’s mission was tackling problems surrounding video transmission in the early internet era, then PPLabs is now tackling a new problem set involving computing power scheduling in the new AI era."}],[{"start":135.05,"text":"According to its listing application, the company's revenue rose from 358 million yuan ($53 million) in 2023 to 770 million yuan last year, averaging 46.6% annual growth over that time. The most eye-catching piece of its pie is AI cloud computing services, whose revenue surged from just 265,000 yuan in 2023 to 10.39 million yuan in 2024, and rose more than 10-fold to 119 million yuan in 2025. In that process, the segment’s share of total revenue jumped from 1.9% to 15.5%. The platform's average daily token consumption jumped from 271 billion at the end of 2025 to 1.028 trillion this April, and revenue from its AI cloud business also continued to grow rapidly over those four months."}],[{"start":189.05,"text":"Rather than simply renting out GPU computing resources, PPLabs hopes to position itself as a computing power scheduling platform. Through its self-developed systems, the company integrates heterogeneous computing resources from different suppliers, while simultaneously emphasizing its inference optimization capabilities. According to the listing application, its platform can reduce first-token latency by 95% and increase total throughput by 300% when compared to the same model using the SGLang inference engine."}],[{"start":222.25,"text":"In other words, the company can get more mileage out of the same amount of GPUs than rival products. Management has also put forward the concept of “agent infrastructure,” with an aim of providing underlying services to AI agents in the future."}],[{"start":235.6,"text":"Still in the red"}],[{"start":237.15,"text":"Despite all the possibilities, PPLabs has yet to find profits. The company's overall gross margin dropped from 17.7% in 2023 to 12.3% in 2024, and fell further still to 9.4% last year. Its AI cloud computing business is even deeper in the red, with a gross profit margin of negative 95.1% in 2024. And while that improved to a negative 10.7% in 2025, the business is still far from profitable. The company recorded a loss of 223 million yuan last year, narrowing from a loss of 293 million yuan in 2024."}],[{"start":279.9,"text":"Obtaining and managing computing power is an extremely high-cost business, and PPLabs’ primary costs come from procuring computing resources from third parties. In 2025, the AI cloud business incurred costs of 132 million yuan, outstripping its 119 million yuan in revenue for the year. In other words, the company is currently still building market share by selling its services at a loss, hoping to build a scale advantage and developer ecosystem while the industry is still in an early stage of rapid growth."}],[{"start":312.54999999999995,"text":"The company’s financial data reflects this pressure. At the end of 2025, PPLabs’ net liabilities totaled 889 million yuan, giving it a current ratio of only 0.4 times – far less than the ratio of 1 or higher that is generally considered healthy. Not only has it recorded net operating cash outflow in each of the last three years, but the scale of that outflow has also been growing. The new listing will help the company replenish its funds, but it will still need to pour much of that back into new investments in computing power and platform construction to keep its AI cloud business growing."}],[{"start":350.54999999999995,"text":"One of PPLabs' greatest advantages is its asset-light business model, since it doesn’t need to build its own data centers or hold large-scale GPU assets. Instead, it focuses on computing power procurement, scheduling and optimization. But that also makes the company highly dependent on external computing power suppliers, requiring ongoing investment to ensure it can procure the computing power it needs."}],[{"start":375.09999999999997,"text":"At the same time, PPLabs also faces competition from much larger cloud service providers such as Alibaba Cloud, Tencent Cloud, Baidu Cloud and Kingsoft Cloud (3896.HK). Most of those possess their own data centers and GPU resources, giving them better control of their computing assets. Hong Kong investors have becoming quite fond of AI-related stocks lately, but PPLabs’ valuation will still depend on whether investors believe it can establish itself in the computer power industry chain using its asset-light model and ability to use computing power resources more efficiently than its peers."}],[{"start":411.2,"text":"In both PPTV and now PPLabs, Yao Xin has bet on the infrastructure business. Two decades ago, it was video transmission. Now, it’s computing power scheduling. For investors, the key lies not in the company's scale, but in whether it can establish technological skills that make its services more effective than its peers, and create a developer ecosystem around its product. If the market sees the company as an AI infrastructure platform, its valuation upside could be far higher than if it’s simply viewed as just another computing power service provider."}],[{"start":453.35,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1781775316_5123.mp3"}