AI Computing Power Industry May Face Upward Price Trend

[baoshi.rao]

With the continuous development of new technologies like AI, the market demand for computing power is enormous. Against the backdrop of supply shortages, the computing power market has entered a strong 'price hike' mode. Many institutions believe that due to factors such as the accelerated iteration of AI applications, the demand for computing power will continue to grow, potentially triggering a medium-to-long-term price surge. The performance boost from the computing power sector will be reflected quarter by quarter, and the localization of AI computing power will also accelerate.

AI Computing Power Industry May See Upward Price Trend

Recently, the computing power industry has been witnessing consecutive price hikes. On November 16, a disclosure by ZBCOM about signing a computing power service framework contract revealed that the company recently entered into a technical service framework agreement with Beijing Zhongke Xinyuan Technology Co., Ltd. The company will provide a total of 1920P AI computing power technical services, with the contract amounting to 345.6 million yuan and a unit price of 180,000 yuan/P/year. Notably, a computing power service contract disclosed by ZBCOM on September 7 showed a unit price of 120,000 yuan/P/year. Compared to this, the computing power service price in the November contract has risen by 50%.

This price hike was foreshadowed. On November 1, a ZBCOM representative stated during an institutional survey that due to tight server supply, the price of computing power servers has risen significantly recently. The company will also increase the price of computing power leasing services for customers, with the extent of the hike under negotiation.

The recent 'cluster' of price hikes in the computing power industry is no coincidence. It stems from the continuous rise in computing power demand driven by the booming development of AI applications and the increasing tightness of computing power supply.

Computing power is the productivity of AI development. This year, AI has advanced rapidly, with tech giants engaging in a 'hundred-model battle.' Data shows that as of October this year, 238 AI large models have been released domestically. As these models continue to iterate, the demand for computing power keeps expanding. According to a Zhejiang Securities research report, OpenAI's ChatGPT-3 training used 128 NVIDIA A100 servers over 34 days, corresponding to 640P computing power, while the GPT-4 model training used 3,125 NVIDIA A100 servers over 90 to 100 days, corresponding to 15,625P computing power. From GPT-3 to GPT-4, the model parameter scale increased about 10-fold, but the number of GPUs used for training rose nearly 24-fold.

In the short term, considering the training needs of 15 leading domestic large-model manufacturers benchmarking GPT-3, an estimated 1,920 A100/A800 servers are required, corresponding to 15,360 GPUs. In the long term, considering the training needs of five leading domestic large-model manufacturers benchmarking GPT-4, an additional 13,705 A100/A800 servers are estimated to be needed, corresponding to nearly 110,000 GPUs.

2023 Cloud Computing Service Structure Models and Product Competition Survey

JD Cloud has initiated China's first public price war in the cloud market, announcing that all its core products will participate in price comparisons, promising a 'price match guarantee.' This includes computing, storage, networking, databases, middleware, and video services, making JD Cloud the vendor with the most extensive product line in this round of price wars, addressing the full range of industrial customers' cloud migration and usage needs.

Earlier, Tencent Cloud announced on May 16th that it would reduce prices for multiple core cloud products, including cloud networking and databases, with some product lines seeing price cuts of up to 40%. These price reductions will officially take effect on June 1st.

This round of price wars began on April 26th when Alibaba Cloud announced its largest-ever price reduction, slashing prices for core cloud products by 15% to 50%, with storage products seeing the highest cuts of 50%.

The fundamental goal of cloud service providers actively lowering prices is to exchange lower prices for higher volume, reducing the barrier to cloud adoption, attracting more users to the cloud computing sector, expanding the user base and scale, and thereby uncovering market demand and increasing the market penetration of cloud computing. In the short term, price reductions will inevitably release technological dividends, driving more users into the cloud computing space and expanding market share.

A significant factor enabling substantial price reductions for cloud products is technological cost reduction. As Qiu Yuepeng, Senior Vice President of Tencent Group, COO of the Cloud and Smart Industries Group, and President of Tencent Cloud, stated: 'Tencent Cloud continuously achieves the best cost-performance ratio for core products through technological innovation and supply chain integration, and will further release technological dividends to users through price adjustments.'

The head of JD Cloud stated that extreme cost-performance is an essential feature of next-generation digital infrastructure. JD.com's large-scale business practices naturally drive JD Cloud to pursue technological cost reductions. Scenarios like online retail subject JD Cloud to extremely rigorous tests, prompting continuous investment in self-developed technologies to improve resource utilization and maximize cost reductions.

Intelligent Cloud Becomes a New Growth Driver for the Market

A recent report by market research firm IDC, 'China Public Cloud Services Market (Second Half of 2022) Tracker,' shows that the overall market size of China's public cloud services (IaaS/PaaS/SaaS) reached $18.84 billion in the second half of 2022, growing by 24.5%. IDC believes that the three-year pandemic led public cloud customers to cut spending and cloud computing vendors to reduce investment budgets, resulting in growth below the previously expected 30%.

In the second half of 2022, the top five public cloud vendors remained Alibaba, Tencent, Huawei, China Telecom, and Amazon.

Among Alibaba Cloud's price reductions, the 7th generation elastic computing instances and Yitian instances see up to a 20% price drop. OSS deep cold archive storage is now 50% cheaper than its previous lowest tier, while SLB load balancing and NAT gateway services are reduced by 15%. The Yitian edition of RDS database services gets up to 40% price cuts, with video cloud and CDN services decreasing by up to 20%. Web Application Firewall security services become 30% more affordable. These price adjustments will take effect on May 7th.

ChatGPT's AI models require large-scale computing power scheduling to support massive training, reduce training costs, and accelerate efficiency. After migrating to cloud computing, ChatGPT's overall costs dropped by approximately 80%, from $12 million to $1.4 million. Cloud-native technology plays a crucial role in ChatGPT by connecting applications with underlying computing resources. This operating system can support various AI and big data applications while managing different computing systems, making cloud-native a core competitive advantage for many high-tech companies.

As the foundational architecture for new infrastructure including big data centers, AI, and industrial internet, cloud computing is receiving significant national investment and accelerating deployment across provinces. With carbon neutrality initiatives, cloud computing is evolving toward energy-efficient, green, and intelligent development, forming new strategic layouts for data centers in China. The Chinese cloud computing industry has maintained over 30% annual growth in recent years, making it one of the fastest-growing global markets. The COVID-19 pandemic particularly boosted demand for remote work, online education, and video conferencing, further accelerating cloud market expansion.

The cloud computing industry is projected to maintain a 22% CAGR in coming years, reaching 386.86 billion RMB by 2025 in China. One key trend is AI-powered cloud platforms enabling access to massive datasets. Using machine learning, these platforms can optimize critical capabilities. AI also helps address various cloud computing challenges, making cloud services the only viable way to scale AI systems. Conversely, cloud computing enables faster AI computations and better resource management.

China Telecom has launched a "self-controllable, intelligent, secure, green, and application-driven" state-owned enterprise cloud resource pool to ensure secure cloud migration for SOEs. Based on fully domestic architecture using Tianyi Cloud 4.0 foundation and multi-chip solutions, this "SOE Cloud" meets Level 3 security certification standards with 200,000 vCPU capacity planned. A second node will be built to achieve nationwide coverage, providing a solid digital foundation for SOE digital transformation. The resource pool already supports SOE regulatory cloud and clients like China Resources, China Coal Energy, and China Information Technology, ensuring interconnected regulatory data and cloud applications.

In 2022, the State-owned Assets Supervision and Administration Commission (SASAC) organized a selection process for public cloud projects among central enterprises. After multiple rounds of evaluation, the first batch of 40 outstanding industry cloud projects was chosen, covering 14 sectors and involving numerous industry leaders. Notably, China Telecom spearheaded or jointly submitted 24 selected projects.

Defense and Energy Sectors
China Telecom collaborated with six industry giants—including China Aerospace Science and Industry Corporation, China Coal Technology and Engineering Group, and State Grid—to build industry clouds. These platforms aim to enhance specialization integration and strengthen supply chain leadership in defense and energy fields.

Metallurgy, Manufacturing, and Construction Sectors
Partnering with five key enterprises such as Chinalco Group and China Minmetals Corporation, China Telecom established industry clouds to improve efficiency across upstream and downstream supply chains.

Healthcare, Logistics, Culture/Tourism, and Finance Sectors
China Telecom's "eSurfing Financial Group Cloud" was successfully selected. The company also joined forces with five leaders including China Resources Group and China Merchants Group to build industry clouds, leveraging state-owned enterprise capabilities to drive digital upgrades.

Automotive, Environmental Protection, Agriculture, Chemicals, and Service Sectors
Two China Telecom-led projects—"Sailwing Smart Connected Vehicle Industry Cloud" and "eSurfing Video Network Industry Cloud Based on Cloud-Network Integration"—were chosen. Collaborations with Dongfeng Motor Corporation and others aim to foster cross-industry convergence.

Under SASAC guidance, China Telecom developed the "State-Owned Enterprise SaaS Application Service Sharing Platform," hosted on eSurfing Cloud. The platform has deployed over 20 SaaS solutions—from office software to security platforms—creating a secure, one-stop cloud service ecosystem for state-owned enterprises.