Overview of China's Computing Power Market Supply in 2026

I. Overview of China's Computing Power Market Supply in 2026

In 2026, China's computing power industry continues its strong growth momentum. By the end of 2025, China's computing power market size had reached RMB 835.1 billion, a year-on-year increase of over 30%. Intelligent computing power exceeded 1590 EFLOPS, and both total computing power and intelligent computing power ranked second globally. A diversified computing pattern integrating general-purpose, intelligent, and supercomputing has taken shape. According to industry estimates, investment in computing power infrastructure in 2026 will be approximately RMB 450 billion, with the share of intelligent computing power continuing to rise. By the end of 2026, the total national computing power is expected to exceed 300 EFLOPS (FP32), a year-on-year increase of 65%, of which intelligent computing power will account for 58%.

On the policy front, the 2026 Government Work Report explicitly called for "implementing new infrastructure projects such as ultra-large-scale intelligent computing clusters and computing-power‑electricity synergy, strengthening the national integrated computing power monitoring and dispatching system, and supporting the development of public clouds." The 15th Five-Year Plan proposal, for the first time, incorporated the "National Integrated Computing Power Network" into the national infrastructure system, calling for moderately advanced construction. Donghai Securities predicts that China's computing power infrastructure scale will continue to maintain a growth rate of about 30% in 2026, with domestic 100,000‑card clusters likely to emerge and the share of domestic computing power further increasing.

On the chip supply side, import substitution has achieved breakthrough progress. According to IDC, NVIDIA's share of China's AI accelerator card market has plummeted from about 95% before sanctions to 55%, while domestic manufacturers have captured 41% of the market. Bernstein Research predicts that in 2026, Huawei will hold 50% of China's AI chip market, becoming the absolute leader. The Zhengzhou node of the National Supercomputing Internet has deployed a 30,000‑card heterogeneous domestic computing power pool, Baidu Intelligent Cloud has built a 30,000‑card Kunlun core self‑developed cluster, and China Mobile's Harbin Intelligent Computing Center has built an 18,000‑card purely domestic computing power cluster, supporting full‑process training of trillion‑parameter models. Industry reports indicate that in 2026, inference scenarios have achieved full substitution of imported products, and low‑to‑mid‑end training scenarios are basically self‑sufficient.

On the demand side, the large‑scale rollout of AI agents is driving exponential growth in token consumption. Huawei noted that demand for AI computing power infrastructure in 2026 is expected to exceed USD 1.3 trillion, a year-on-year increase of 41%. Computing power is no longer used only for model training but also supports massive intelligent decisions occurring anytime, anywhere. Guojin Securities predicts that the computing power industry chain will enter a "full‑chain inflation" cycle in 2026, with industry prosperity spreading from core chips to AIDC, cloud and computing power services, supporting power equipment, servers, and other links. Guoxin Securities pointed out that the current price increase is the result of three factors: demand explosion, supply constraints, and business model upgrades, with inference computing power replacing training computing power as the core engine.

From a global perspective, TSMC's 2025 revenue reached USD 122 billion, a year-on-year increase of 35.9%; SK Hynix's Q4 2025 operating profit increased 137% year-on-year to RMB 96.89 billion, with an operating margin of 58% reaching an all‑time high. Samsung Electronics' DS division recorded a 465.5% year-on-year increase in operating profit during the same period. HBM, high‑density DDR5, and enterprise SSDs have become key growth drivers. Samsung stated that all HBM production capacity for 2026 has already been booked, and the supply tightness is unlikely to ease in the short term.

II. Price Analysis of Complete Server Systems

In 2026, the AI server market experienced significant price increases, with cost pressures running through the entire industry chain. Storage chip prices soared: the unit price of DDR4 64G server memory reached RMB 3,100, doubling from RMB 1,500 two months earlier; DDR5 64G consumer memory modules rose to RMB 7,000; second‑hand high‑frequency DDR5 server memory modules exceeded RMB 13,000, while new products were quoted at over RMB 18,000. U.2 interface enterprise SSDs are "priceless and out of stock" in Huaqiangbei. GPU prices also surged simultaneously, with a single RTX 5090 exceeding RMB 30,000; the latest quote on January 28 was RMB 33,000, with orders requiring queuing.

In terms of complete server quotes, a server equipped with eight NVIDIA RTX 5090 GPUs in Shenzhen's Huaqiangbei electronics market was quoted at RMB 400,000, an increase of RMB 100,000 from the end of December 2025. Merchants said component prices change "every half day," quotes are valid only for the day, and buyers must pay a 50% deposit when placing an order, with full payment due before shipment. An 8‑card A100/A800 server (including dual Xeon 8462Y + 1TB memory) is mainly quoted in the RMB 580,000–650,000 range. One major internet company purchased 30 units through centralized procurement and managed to push the final transaction price down to RMB 558,000 per unit. It is important to note that systems equipped with HGX H100 8‑GPU still require a 3–5 week wait. There are a small number of Hong Kong‑source units in the Huaqiangbei spot market, with a single card starting price of RMB 93,000, 23% higher than the official price. NVIDIA's newly released Vera Rubin rack‑scale AI system is even more expensive, with a single rack estimated to cost between USD 3 million and USD 7 million.

On the domestic computing power front, as advanced process nodes are gradually achieved, the market share of domestic high‑end AI chips has significantly increased in 2026. However, due to capacity scheduling constraints, spot prices have risen about 15%–25% from the beginning of the year. On the cost side, the mismatch between supply and demand for HBM memory and DDR5 has pushed up chip manufacturing costs. Combined with global shortages in advanced packaging capacity, the deployment lead time for new Blackwell clusters has now been extended to June–July 2026. Overall, the AI server market in 2026 exhibits a pattern of "both volume and price rising, supply falling short of demand," significantly raising the threshold for small and medium‑sized enterprises to build their own computing power.

III. Price Analysis of Computing Power Leasing Market

The computing power leasing market in 2026 ushered in a landmark price‑hike cycle, ending the nearly 20‑year "only down, never up" pricing tradition in the cloud computing industry. According to the latest report from research firm SemiAnalysis, the one‑year lease price for an H100 GPU rose sharply from a low of USD 1.70 per hour in October 2025 to USD 2.35 per hour in March 2026, an increase of nearly 40%. On‑demand computing power is sold out across all GPU models.

In the domestic market, industry monitoring shows that surging AI computing power demand at the beginning of 2026 has driven a price‑hike cycle in the computing power leasing market. As of the end of February, rental prices for high‑end GPUs such as NVIDIA H200 and H100 rose 15%–30% month‑on‑month. The hourly rental rate for H200 reached RMB 7.5–8.0 per card‑hour, with monthly rent ranging from RMB 60,000 to 66,000, an increase of 25%–30%. Monthly rent for H100 rose to RMB 55,000–60,000, an increase of 15%–20%. Delivery lead times have been extended to Q2 2027 (H200) and Q1 2026 (H100). Leading players such as UCloud and Senhua Yiteng have officially announced a 20%–30% price increase across the board in March. In the Huaqiangbei market, B200 is basically not available in stock, and H100 and H200 are also in short supply.

On the domestic computing power leasing side, leading domestic computing power service providers have officially announced across‑the‑board price increases for computing power services, extending from high‑end categories to all service categories. Delivery lead times for high‑end GPU computing resources have significantly lengthened, and supply‑side gaps continue to stand out. The market has clearly entered a "seller's market."

IDC predicts that the global computing power leasing market will exceed USD 80 billion by 2026. The Ministry of Industry and Information Technology has promoted exploration of innovative service models such as "computing power banks" and "computing power supermarkets," supporting SMEs in obtaining computing resources through flexible methods such as "card‑hour" or "core‑hour" billing. Overall, as AI agents are rolled out at scale and multimodal applications become widespread, inference‑side computing power consumption is growing exponentially. Computing power leasing prices are expected to remain high for an extended period.

IV. Comprehensive Expansion of Computing Power Application Scenarios

In 2026, computing power has deeply penetrated thousands of industries, with application scenarios extending from traditional cloud‑based training to all‑round intelligent decision‑making. Huawei noted: "Computing power is no longer only used for model training, but also supports massive intelligent decisions occurring anytime, anywhere. Scenarios such as autonomous driving, AI4S, financial risk control, industrial quality inspection, smart wearables, and AI PCs are generating full‑spectrum computing power demands from the hyper‑scale to the edge."

Large Models and Agents — 2026 is widely recognized as the breakout year for AI super‑agents. Agents have evolved from auxiliary tools into "digital employees" with autonomous decision‑making and closed‑loop execution capabilities. Current global daily token consumption exceeds 360 trillion. Huawei has internally developed over 600 intelligent applications and more than 19,000 digital employees, and is jointly building industry agents with partners in automotive, healthcare, finance, transportation, and other sectors.

Autonomous Driving and Embodied AI — The cognitive functions of large models are moving from symbolic text processing to environmental interaction and physical reasoning. By integrating multimodal data such as vision and 3D space, they improve decision‑making rationality in scenarios requiring interaction with the real world, such as autonomous driving and humanoid robots. Huawei, together with multiple listed companies, is advancing the implementation of "industry + AI" lighthouse solutions, with smart healthcare and embodied AI becoming key development directions.

Smart Manufacturing and Industrial Quality Inspection — Large model technologies are deeply applied in the manufacturing sector, promoting 500 typical application scenarios, creating 100 high‑quality industrial datasets, and selecting 1,000 benchmark enterprises. Scenarios such as industrial quality inspection and predictive maintenance of equipment impose higher requirements for low‑latency computing power at the edge.

AI for Science — The inference mechanisms of large models are moving from shallow statistical correlations to deep causal inference, providing reliable technical support for scientific discovery scenarios such as drug development and materials innovation. The Zhengzhou node of the National Supercomputing Internet supports trillion‑parameter model training and AI for Science scenarios, making it the largest operational purely domestic AI computing power pool in China.

Financial Risk Control and Quantitative Trading — The financial industry uses large models for real‑time risk control, intelligent investment advisory, and credit approval. A quantitative trading system at one securities firm adopted a hybrid cooling solution, saving RMB 370,000 in electricity costs annually.

Smart Healthcare and Life Sciences — Large models assist in medical image analysis, drug molecular screening, and genomics research. CCID Research Institute pointed out that in 2026, AI technologies will accelerate the transformation and upgrading of traditional industries, deeply integrate with emerging industries, and empower thousands of industries in all directions.

Looking ahead to 2026, CCID Research Institute expects that the penetration depth and integration breadth of AI industry applications will further expand. Technologies are accelerating their extension from front‑end service links to core areas such as R&D and production, and the full‑process intelligent restructuring is becoming the core path for traditional industry upgrades. Embodied AI will break through virtual boundaries to achieve large‑scale implementation in physical scenarios, forming a synergistic development trend with fields such as autonomous driving and XR.

V. Computing Power Leasing Services and Industry Enablement

Faced with increasingly scarce computing power resources and continuously rising hardware procurement costs, flexible and efficient computing power leasing services have become an important path for enterprises to embrace AI at low cost. Yuan Shuai, Deputy Secretary‑General of the Zhongguancun IoT Industry Alliance, pointed out that computing power leasing is expected to become a key solution to alleviate the current AI computing power supply gap. It can quickly match the short‑term computing power needs of AI companies, especially for SMEs lacking sufficient funds and technical capabilities to build their own data centers, significantly reducing upfront investment costs while improving overall computing power resource utilization.

Omniyq (www.omniyq.com) focuses on providing GPU computing power bulk leasing and complete server sales services. Through a massive pool of high‑performance computing resources, it serves upstream and downstream customers in the AI industry, empowering AI application scenarios across thousands of industries. Leveraging its diverse computing power infrastructure including RTX 4090, RTX 5090, H100, H200, B200, B300, and others, along with a flexible, efficient, and cost‑effective leasing model, Omniyq has provided stable and convenient computing power services to a large number of industry customers in fields such as artificial intelligence, scientific computing, graphics rendering, deep learning, autonomous driving, and smart manufacturing. Its vast high‑performance computing power resources are available anytime, anywhere to provide efficient and portable computing power leasing services to upstream and downstream customers in the AI field, empowering AI application scenarios across thousands of industries and contributing to the new productive forces in the digital economy era.

As the core productive force of the digital economy, computing power is driving industrial transformation at an unprecedented pace. With the deepening of the "AI+" initiative, the computing power leasing industry will play an increasingly important role in supporting the intelligent transformation of the real economy. Service providers such as Omniyq will continue to provide a solid and reliable computing power foundation for AI applications across thousands of industries.