[SMM Computing Power News] Western Pro 6000D annual rental 36,000, three months' deposit and one month's payment

**Major Recent Advances in Huawei Ascend** **[Large-Scale Deployment]** On 10-16, China Telecom and Huawei launched the world's first Ascend 950DT 10,000-card pool in Inner Mongolia; on the same day, China Mobile launched an Ascend 950DT 1,000-card supernode in Beijing. On 10-15, the first 1,024-card Atlas 950 SuperPoD liquid-cooled supernode was powered on and completed 0Day adaptation for StepFun Step 5 Preview. **[Core Performance]** The 1,024-card liquid-cooled supernode features the following key specifications: computing power of 1 EFLOPS FP8 / 2 EFLOPS FP4, equipped with 256TB of globally unified memory, inter-card interconnect bandwidth at the TB level, and end-to-end communication latency of only 3μs. **[Technology Evolution and Roadmap]** Huawei has achieved cluster upgrades from 384 cards to 8,192 cards; the next-generation 960 supernode will debut the NPO solution, with support for scaling up to 4,096 cards. **[Market Performance]** In H1, Huawei Ascend shipments reached 572,000 units, with a market share of 28%.

Of the 19 indices today, 18 remained flat. The lowest monthly rental for the Yangtze River Delta A100-80G rose from 34,000 yuan to 36,000 yuan (up 5.9%), and the lowest hourly rate rose to 6.25 yuan (up 5.9%). The average monthly rental rose to 38,000 yuan (up 2.7%), while the highest price of 40,000 yuan was not breached. A channel distributor in Jiangsu has fully rented out its inventory of this model at 36,000 yuan per unit per month, clearing out within just one week of listing and reaching the upper end of the 35,000-36,000 yuan range, confirming replenishment at the low end of the index. The western Pro 6000D is quoted at 36,000 yuan per unit per month, with a minimum one-year lease.

When computing power rental market quotes stabilize, the perspective might shift from the rent itself to a more fundamental question: for the same computing power, why do operating costs and rental prices differ when deployed in different cities? Ulanqab offers a sample worth studying. The computing power competitiveness of this grassland city can be summarized in three words: wind, cold, and proximity. SMM view: Site selection for computing power cannot focus on a single indicator alone. Electricity prices determine basic energy costs, average temperatures affect cooling energy consumption, and location and network infrastructure determine whether computing power can effectively reach clients. Only when the three work together can resource endowments be transformed into sustainable computing power rental.