[SMM Computing Power Daily] H200 price rises again to 140,000, South China RTX 5090 rental 11,500, Zhejiang

SMM, August 6 –

H200 Monthly Rental Quote Climbs to 140,000 Yuan, Market Enthusiasm Unabated

H200 leasing prices extended their rise this week. A Beijing-based IDC service provider reported that the current monthly rental quote for H200 had reached 140,000 yuan per unit. Earlier in the week, the market generally expected H200 quotes to be in the 120,000–130,000 yuan range, but quotes quickly broke through from the weekend to today. SMM previously confirmed that an intelligent computing center in Hubei transacted one H200 unit at 130,000 yuan per unit over the weekend, while today’s new quote of 140,000 yuan indicates that upward momentum has yet to be exhausted. As one of the highest-end training GPUs that NVIDIA can currently supply to China in compliance with regulations, H200 sees very limited spot circulation, and any released supply is quickly absorbed. Market participants noted that current H200 pricing has become “uncomputable,” with wide price fluctuations and widespread markups through intermediary layers; for batches of 32 units or more, price verification of supply authenticity requires particular caution.

South China: 32 RTX 5090 Units Available for Split Lease, Six-Month Short Contract Lowers Entry Barriers

A facility in South China has brought online 32 RTX 5090 8-GPU server leasing resources, with a monthly rental quote of 11,500 yuan per unit, a minimum lease term of six months, and support for single-unit split lease. Adding an ROCe networking solution incurs an additional 1,000 yuan per unit per month. As of press time, these resources have been online for three days and still have availability. A day earlier, another batch of 20 RTX 5090 units without networking was offered for lease in South China at a monthly quote of 13,000 yuan per unit. The current quote of 11,500 yuan is about 12% lower than that batch and offers optional ROCe networking, presenting a clear cost-performance advantage. Under the split lease model, buyers can start with a single unit based on demand, suiting lightweight scenarios such as distributed inference and rendering, and also enabling start-up teams to enter the market with lower upfront costs.

Zhejiang: Single H100 Unit Released, Scattered Resources Offer Negotiable Pricing

A facility in Zhejiang released one NVIDIA H100 8-GPU server, with a monthly rental quote of 76,000 yuan, a minimum one-year lease, and expected delivery by August 16. This resource is a scattered single unit, and the lessor priced it favorably to avoid idle windows. According to SMM’s previous understanding, the mainstream monthly rental range for H100 full servers lies between 78,000 and 85,000 yuan; scattered single units, due to lower liquidity and longer matching cycles, typically see lessors willing to compromise on quotes. This 76,000-yuan quote is close to the lower end of the current market range, offering some appeal to tenants with stable demand who do not require cluster networking.

SMM Analysis:H200 prices rapidly climbed from 130,000 to 140,000 yuan, indicating that the supply-demand gap for high-end computing power is still widening. The price spread between H200 futures and spot cargo is currently significant, and complex intermediary structures further amplify quote fluctuations. Buyers must verify supply authenticity through delivery capability and closed contracts. The growing adoption of split-lease and short-contract models reflects a supply-side evolution from unitary “full-rental long-term contracts” to a more flexible product structure. For the consumer-architecture RTX 5090 GPU, split leasing lowers the usage barrier for small and medium-sized enterprises and start-up teams, but also drives up per-unit operating costs—lessors bear higher vacancy risk and operational fragmentation costs, and the sustainability of the split-lease model warrants attention. On the H100 front, quotes for scattered single units are declining, and the room for negotiation on small, scattered resources is increasing, providing a window of opportunity for small and medium-sized customers whose cluster needs are not urgent.