In the long term, GPU cloud (compute power leasing) provides a flexible and cost-effective solution, and its penetration rate is expected to continue increasing.
According to the Zhito Finance APP, Guotai Junan Securities has released a research report stating that the prosperity of AI computing power continues. In the short term, GPU cloud (computing power leasing) may be the core solution to address the mismatch in supply and demand for high-end computing power; in the long term, GPU cloud (computing power leasing) offers a flexible and low-cost solution, with penetration rates expected to continue to rise. It is recommended to pay attention to domestic GPU cloud-related companies, particularly Runjian Co., Ltd. (002929.SZ) and related industry firms.
Key points from Guosen Securities are as follows:
The global computing power prosperity continues. With NVIDIA’s GB series high-density computing cabinets accelerating shipments, the global high-end computing power prosperity further improves.
Amid the current AIGC wave, global server shipments continue to grow. Consulting firm IDC predicts that by 2028, the global artificial intelligence server market size is expected to reach USD 222.7 billion, with generative AI servers’ share increasing from 29.6% in 2025 to 37.7% in 2028. On the demand side, the training demand remains high amid accelerated model iterations, while inference-side demand gradually increases with application penetration. On the supply side, high-performance computing power chips represented by NVIDIA’s B/Rubin and AMD MI series continue to iterate, with the GB300 expected to accelerate delivery in the second half of 2025. According to CSP manufacturers’ Capex guidance, it is expected that in 2025, the total Capex of overseas manufacturers Amazon, Google, Microsoft, and Meta will increase to USD 361 billion, a year-on-year increase of over 58%; domestic companies such as ByteDance, Tencent, and Alibaba are expected to exceed CNY 360 billion in Capex, maintaining a high prosperity level in AI development.
GPU cloud (computing power leasing) may address the current global shortage of high-end AI chips, as the GPU cloud (computing power leasing) market rapidly develops.
In the context of the large model arms race, major companies are accelerating the construction of clusters with tens of thousands or even hundreds of thousands of nodes. AI giants such as Meta, Microsoft & OpenAI, and xAI have successively announced or completed the construction of 100,000-node clusters, while domestic telecommunications operators, leading internet companies, and large AI research and development enterprises are also vigorously developing layouts for clusters exceeding ten thousand nodes. However, in the context of a global shortage of high-end AI chips, the business model of leasing instead of purchasing has emerged, as leasing models are more adaptable and cost-effective. The cloud computing market, after going through traditional cloud and hybrid cloud phases, is now welcoming a third wave of differentiation—the AI Intelligent Cloud NeoCloud, namely GPU cloud (computing power leasing), with the global GPU cloud (computing power leasing) market expected to grow to USD 12.8 billion by 2033 (according to Verified Market Research).
AI chip giants are strategically expanding their global market presence through the GPU cloud (compute power leasing) business model, and the domestic GPU cloud market development is promising.
(1) In the global market, NVIDIA is supporting the development of GPU cloud providers (such as CoreWeave, NBIS, Omniva) through equity or partnership arrangements, consolidating its global leadership position in the high-end chip sector. Although the three GPU cloud providers vary in coverage and development scale, they all benefit from the robust demand in the GPU cloud market, experiencing rapid growth. By Q2 2025, CoreWeave and NBIS are projected to achieve revenue growth rates of 207% and 625%, respectively. (2) Domestically, domestic AI chips currently mainly support inference tasks, while NVIDIA’s high-end AI chips perform better in certain training scenarios. There are differences in compute policies between domestic and international markets, and utilizing OPEX leasing for compute power may offer a higher cost-performance ratio for training operations, presenting a development opportunity for domestic compute leasing companies. Currently, the rental returns for domestic compute leasing enterprises are quite promising, with net profit margins potentially reaching 15%, sharing some similarities with the business model and development prospects of overseas GPU cloud (compute power leasing).
Risk Warning: The development and investment in AI may fall short of expectations, industry competition may intensify, global geopolitical risks may arise, and new technological developments may trigger changes in the industry chain.