Shen Wan Hongyuan: Advance the upgrading of collaborative computing facilities, focus on normalized scheduling and energy operation in the medium to long term

The Zhitong Finance App learned that Shen Wan Hongyuan released a research report saying that IT collaboration follows the three main lines of facility upgrade, resource operation, scheduling, and interface layout to distinguish between order fulfillment and continuous service space. The facility side focuses on power supply and distribution, liquid cooling, and energy storage requirements; the operating side focuses on power access, customer listing, and capital return; the software side focuses on production system access, paying customers, and cross-project replication. Recently, emphasis has been placed on enterprises with an existing customer and order base. In the medium to long term, attention has been paid to continuous charging capacity brought about by normalized scheduling and energy operations, and distinguishing the company's overall performance from the contribution of electricity computing-related businesses.

Shen Wan Hongyuan's main views are as follows:

The growth of AI electricity consumption and demand for new energy consumption jointly promote electricity computing collaboration, and the policy extends from spatial layout to operation mechanism

The scale of electricity consumption in data centers is expanding, and power supply continuity requirements are high, while Western New Energy is facing problems of fluctuating output and insufficient local consumption capacity. Electricity calculation collaboration is expected to improve the spatio-temporal matching between the two. The domestic policy gradually extended from the “East Digital Western Computing” hub layout to direct connection to green power, two-way collaboration and market-based trading of computing power and electricity. The share of green electricity, PUE, and listing rate separately constrain power sources, facility energy efficiency, and asset utilization efficiency, and promote the industry to simultaneously optimize resource allocation and operation methods.

The electricity price difference creates operating income, and the facility investment required to obtain low-cost electricity determines the project return

Take 10MWIDC as an example. Under the assumption that the PUE is 1.2 and the stable load rate is 90%, the electricity price is reduced from 0.35 yuan/kilowatt hour to 0.25 yuan/kilowatt hour, which can save about 9.46 million yuan in annual electricity costs. After fixing conditions such as 100% direct connection coverage, the IRR for reusing existing access and additional 35kV and 110kV projects was 10.8%, 8.4%, and 5.9%, respectively. The advantages of low electricity prices need to be jointly assessed with investment in dedicated lines, power conversion, and energy storage. Reusing existing facilities has economic value.

Computing power scheduling enhances effective output and load flexibility, and computational interfaces promote cross-system collaboration and commercialization

Separation of inference stages, resource orchestration, and data access optimization reduce computational waiting, and power sensing scheduling further utilizes task delivery margins to adjust execution arrangements according to electricity prices, green power output, and power grid requirements. Tasks such as offline reasoning have a lot of room for adjustment, and online interaction and large-scale synchronous training are limited by time delays and communication. The interface connects computing power and energy systems through unified metering, control authorization, and execution feedback. Commercial revenue covers system construction, predictive operation and maintenance, and transaction assistance, and demand response compensation forms additional revenue.

The goals of stable and low-carbon power supply between China and the US converge, and resource conditions and industrial advantages shape different implementation paths

China has new energy resources, cross-regional power transmission and equipment manufacturing infrastructure. The focus is on transforming these conditions into deliverable computing power and project returns. Some AI clusters in the US are constrained by partial power connections and equipment delivery, while evaluating grid expansion, long-term electricity purchases, and on-site power supply. High-density cabinets drive the upgrading of power supply, distribution and energy storage on both sides. Leading US chip and cloud vendors have strong leadership in architecture and verification requirements. China's power equipment and battery industry chain has a manufacturing foundation, and competition ultimately still needs to be implemented in system reliability, customer certification, and scale delivery.

Risk warning: AI demand and project implementation fall short of expectations; narrowing of electricity price differences and fluctuations in green power supply; supporting investment exceeds expectations and access delays; load adjustment affects business continuity; market rules and revenue delivery are uncertain; commercialization of new technologies falls short of expectations; market competition and overseas business risks