Recent data from SemiAnalysis indicates continued tightness in the GPU leasing market. As of March 2026, the price for a one-year H100 GPU leasing contract has surged to $2.35 per hour per GPU, a near 40% increase. Looking ahead, the expansion of AI inference demand, the advancement of sovereign AI initiatives, and the implementation of AI solutions by enterprises are expected to sustain strong demand for high-end computing power. With the ongoing increase in capital expenditure by the MAG cloud providers, the positive momentum in the industrial chain is likely to continue. This round of price increases is expected to enhance the revenue and profit margins of hyperscale cloud providers and NeoCloud companies, potentially leading to a simultaneous improvement in both valuation and earnings per share for the sector.
The price surge for the H100 is fundamentally a result of robust demand coexisting with supply chain constraints. Firstly, demand for AI model training and inference continues to grow rapidly. Leading model companies, cloud providers, and enterprise clients are persistently competing for limited high-end GPU resources, directly driving up leasing demand for the H100. For instance, demand for Anthropic's Claude 4.6 Opus and Claude Code has skyrocketed. Anthropic's annual recurring revenue multiplied by more than three times in just one quarter, increasing from $9 billion at the end of last year to $30 billion currently. Open-source models like GLM and KimiK2.5 have spurred a surge in application use cases. Financing activities by companies such as Anthropic, OpenAI, and various Neolabs also require substantial GPU resources.
Secondly, delivery involves systemic bottlenecks. The GPU shortage is not merely an issue of chip quantity but also a problem of system-level delivery bottlenecks, encompassing complete servers, network interconnects, power supply infrastructure, and data center deployment schedules. Therefore, possessing the physical chips does not automatically equate to having usable computing power.
Thirdly, market resource allocation exhibits a structural bias. Major clients often prioritize locking in large-scale computing capacity, while small and medium-sized clients increasingly turn to the leasing market. This dynamic further intensifies the tightness of spot and short-term leasing resources, pushing H100 prices higher.
Looking forward, the expansion of AI inference demand, the progression of sovereign AI projects, and enterprise-level AI adoption are likely to continue supporting strong demand for high-end computing. Consequently, in the short term, the price floor for the H100 may remain supported, with the market potentially transitioning gradually from "absolute shortage" to "structural tightness." In the medium to long term, the focus of industry competition may shift from pure GPU supply to capabilities in cluster delivery, network efficiency, data center infrastructure, and computational utilization. Platforms and infrastructure providers with strong system integration capabilities are positioned to benefit relatively more.
Despite the tight AI computing supply and demand pushing up cloud leasing prices, the three MAG cloud giants continue to significantly increase their capital expenditures. Amazon anticipates its 2026 Capex to reach approximately $200 billion, primarily directed towards AWS AI infrastructure. Alphabet has guided for 2026 Capex in the range of $175 billion to $185 billion, nearly doubling compared to 2025. Microsoft's Capex for FY2026 Q2 alone reached approximately $29.9 billion, a year-on-year increase of nearly 89%, with full-year Capex expected to maintain rapid growth.
As MAG cloud Capex continues to increase, the prosperity of the industrial chain is expected to persist. Underpinned by strong demand, this round of price hikes is anticipated to boost the revenue and profit margins of hyperscale cloud vendors and NeoCloud companies. The industry may experience a dual boost to both valuation and earnings per share. Key beneficiaries of the rising leasing prices to monitor include Oracle (ORCL.US), Coreweave (CRWV.US), NEBIUS (NBIS.US), IREN ltd (IREN.US), Amazon (AMZN.US), Google (GOOGL.US), and Microsoft (MSFT.US).
Investors should be mindful of risks, including potential slower-than-expected growth in AI demand, a deceleration in capital expenditure by cloud providers, and the risk of leasing price declines if H100 supply increases faster than anticipated. Additionally, risks related to export restrictions, supply chain disruptions, and the impact of next-generation chip substitution on industry dynamics warrant attention.
Comments