East Money Information Securities: H100 leasing prices increase to boost cloud industry sentiment, MAG cloud CAPEX expected to continue investment.

East Money Information Securities released a research report stating that according to data from SemiAnalysis, the GPU leasing market remains tight. As of March 2026, the price of a one-year GPU leasing contract for H100 has skyrocketed to $2.35 per hour per GPU, an increase of nearly 40%. Looking ahead, the expansion of AI inference demand, the promotion of sovereign AI construction, and the implementation of AI on the enterprise side may continue to support high-end computational power and maintain high demand. With MAG Cloud's continued increase in Capex, the industry chain's prosperity is expected to continue, and this round of price increases is expected to boost the revenue and profit margins of major cloud providers and NeoCloud, leading to a potential increase in valuation and EPS for the industry. The main points of East Money Information Securities are as follows: Event: According to SemiAnalysis data, the GPU leasing market remains tight. The price of a one-year GPU leasing contract for H100 has soared from a low point of $1.70 per hour per GPU in October 2025 to $2.35 per hour per GPU in March 2026, an increase of nearly 40%. The price increase of H100 is essentially a result of high demand and delivery constraints coexisting 1) AI model training and inference demand continue to grow rapidly: Top model companies, cloud providers, and enterprise customers are competing for limited high-end GPU resources, directly driving up the demand for H100 leasing. For example, the demand for Anthropic's Claude 4.6 Opus and Claude Code has skyrocketed. Anthropic's annual recurring revenue (ARR) doubled in just one quarter, from $9 billion at the end of last year to $30 billion now. Open-source models like GLM and KimiK2.5 have driven a surge in open-source model application cases. Financing activities of companies like Anthropic, OpenAI, and many Neolabs require a large amount of GPU resources. 2) Delivery involves systemic bottleneck issues: GPU shortages are not just about the number of chips but also systemic delivery bottleneck issues involving server assemblies, network interconnectivity, power supply, data center deployment schedules, and more. Therefore, "having cards doesn't necessarily mean having usable computational power." 3) Structural skew in market resource allocation: Top customers often lock in large-scale computational power, while small and medium customers lean more towards the leasing market, further exacerbating the scarcity of spot and short-term leasing resources, leading to an increase in H100 prices. Looking ahead, the expansion of AI inference demand, the promotion of sovereign AI construction, and the implementation of AI on the enterprise side may continue to support high-end computational power and maintain high demand. Therefore, in the short term, the central price of H100 may still have some support, and the market is more likely to move from "absolute scarcity" to "structural scarcity" gradually. In the medium to long term, the industry's competitive focus may shift from simple GPU supply to cluster delivery capabilities, network efficiency, data center infrastructure, and computational power utilization rate, with platforms and infrastructure vendors with system integration capabilities benefiting more. Although the tight supply and demand of AI computational power have driven up cloud leasing prices, the three major cloud giants MAG are still significantly increasing their capital expenditures. Amazon expects its Capex to reach around $200 billion in 2026, mainly for AWS AI infrastructure; Alphabet's guidance for its 2026 Capex is $175 billion to $185 billion, almost doubling from 2025; Microsoft has already reached about $29.9 billion Capex in FY2026 Q2, an increase of nearly 89% year-on-year, and full-year Capex is expected to maintain high-speed growth. With MAG Cloud's continued increase in Capex, the industry chain's prosperity is expected to continue. Under high-demand conditions, this round of price increases is expected to boost the revenue and profit margins of major cloud providers and NeoCloud, leading the industry to potentially see an increase in valuation and EPS. It is recommended to pay attention to the core beneficiaries of this round of leasing price increases: Oracle, Coreweave, NEBIUS, IREN ltd, Amazon, Google, and Microsoft. Risk warning: It is necessary to pay attention to the unexpected slowdown in AI demand growth, the slowdown in capital expenditures by cloud providers, the risk of leasing price declines due to faster-than-expected release of H100 supply, as well as the impact of export restrictions, supply chain disruptions, and the substitution of new-generation chips on the industry's prosperity.