AI data center power supply chain, Goldman Sachs prefers

# Source: Wall Street Insights

## By: Bao Yilong

Goldman Sachs believes that supply shortage-driven pricing premiums have allowed Chinese suppliers to sell their products overseas at prices **10% to 80% higher** than domestic rates. Even after accounting for increased tariffs and logistics costs, profit margins have improved significantly. In terms of product category preferences, Goldman Sachs ranks them as follows: gas turbine blades, power transformers, electrical components, UPS/power racks, liquid cooling systems, and server power supply units (PSUs).

The global wave of AI data center construction is facing a severe **"power bottleneck"**, especially in the U.S. market. Goldman Sachs argues that this presents a historic opportunity for a group of Chinese power supply chain enterprises with advantages in technology, cost efficiency, and delivery speed.

According to news from Zhuifeng Trading Desk, on January 14, Goldman Sachs' research team released a report stating that the surging power demand of global AI data centers has led to a critical shortage of power generation equipment (e.g., gas turbines) and power grid equipment (e.g., transformers), with delivery times stretching to several years.

With European and American suppliers' orders booked out until 2028, the **shorter delivery cycles** of high-quality Chinese suppliers have become a core competitive advantage. Goldman Sachs notes that the supply shortage-driven pricing premium enables Chinese suppliers to price their products 10% to 80% higher overseas than domestically. Even after factoring in rising tariffs and logistics costs, profit margins remain substantially enhanced.

In terms of product category priorities, Goldman Sachs’ ranking is: **gas turbine blades > power transformers > electrical components > UPS/power racks > liquid cooling systems > server power supply units (PSUs)**.

Goldman Sachs forecasts that Chinese enterprises will achieve an average **compound annual growth rate (CAGR) of 23%** in sales from 2025 to 2030. By 2030, the overseas AI data center market will contribute an average of 23% of total revenues, with Chinese companies capturing an average of **4% of the global market share**.

## Global AI Data Center Capacity Surges, but U.S. Power Bottleneck is More Severe

Goldman Sachs points out that global AI data center (AIDC) capacity is expanding rapidly, yet the U.S. faces a far more acute power bottleneck than other countries.

The report projects that between 2025 and 2030, global data center capacity will increase by approximately **73 gigawatts (GW)**, with most of the growth concentrated in the U.S. and China.

Despite the surging demand, U.S. power supply is under severe strain, with bottleneck effects far more pronounced than in China and other regions:

1.  **Severe Local Supply Shortfall**: Domestic U.S. power equipment production capacity can only meet about **40% of local demand**, leaving a massive supply gap.

2.  **Tight Grid Reserve Capacity**: Compared with China, the U.S. has much tighter effective power reserve capacity. Report data shows that from 2025 to 2030, the U.S. is expected to have an average power reserve capacity of around 100 GW, with shortages set to worsen further. Meanwhile, China’s average reserve capacity during the same period will reach as high as 300 GW, providing a much more robust system buffer.

3.  **Extremely High Demand Growth Rate**: Citing internal team forecasts, the report states that by 2030, power consumption of U.S. data centers (including both AI and non-AI facilities) will surge by approximately **175%** compared with 2023, far outpacing the 2.6% annual compound growth rate of the country’s overall power demand.

4.  **Forced Shift to Non-traditional Suppliers**: With lead times for critical equipment (such as gas turbines and transformers) ranging from 3 to 5 years, U.S. power companies and data center operators are increasingly willing to source from non-traditional suppliers to address the years-long supply gap. This creates market opportunities for qualified Chinese suppliers capable of rapid delivery.

Globally, the expansion of AI computing capacity is shifting the focus from **"how many GPUs are needed"** to **"where the power will come from and how quickly it can be deployed"**.

Against this backdrop, the U.S.—plagued by limited domestic supply capacity, tight grid reserve margins and soaring demand—has emerged as the region with the most severe power bottleneck. This makes suppliers offering **"Time-to-Power" solutions** extremely valuable.

## Power Generators and Transformers Face the Tightest Supply, Shortage Cycle to Last Until 2028–2030

Goldman Sachs emphasizes that power generators and transformers are the **tightest supply chain bottlenecks** in global AI data center construction, and this shortage cycle is expected to persist until 2028–2030.

The report notes that multi-year supply shortages have emerged in power generation (especially gas turbines) and grid transmission (especially transformers) sectors. Capacity expansion has fallen far short of initial plans, while demand has surged, resulting in a persistent supply-demand imbalance.

Based on technical barriers and order backlogs, the report concludes that the tight supply situation for gas turbines (particularly blades) will last **at least until 2028–2030**.

Gas turbine blades must withstand extreme temperatures and physical stresses, requiring single-crystal superalloy materials, specialized coatings, complex aerodynamic designs and sophisticated front-end manufacturing processes (such as wax tree fabrication and assembly). This makes them one of the components with the highest technical thresholds and slowest capacity expansion rates.

In addition, power transformers face the most severe bottlenecks due to their high degree of customization and labor-intensive nature.

The average age of power grids in the U.S. and Europe is **35–40 years**, which already makes them ill-equipped to handle the surging power demand of the AI era, exacerbating the pressure for transformer replacement and upgrades.

Each transformer requires custom design based on specific impedance, cooling, tap changer, overload and seismic standards. This **"one-product-one-design" model** makes production highly dependent on skilled labor and difficult to scale rapidly through capital expenditure.

## Technological Transformation: The "Great Migration" to 800V DC Architecture

As the power density of AI server racks exceeds 100kW and moves toward 1MW, traditional copper wire transmission has approached its physical limits.

To carry massive electric currents, a dramatic transformation is underway in data center electrical architectures: shifting from traditional alternating current (AC) to **800V direct current (DC) architecture**. This transition can reduce energy consumption by approximately **5–15%**.

Driven by capacity expansion and technological upgrades, Goldman Sachs forecasts that power supply products for AI data centers covered by China’s industrial technology sector will maintain a **39% CAGR** through 2030.

The research report stresses that this will significantly benefit component suppliers with DC technology reserves, especially leading enterprises in the relay and capacitor sectors.

## Chinese Suppliers Capture Spillover Demand with Delivery Speed and Competitive Products

Faced with 3–5 year wait times for key components, Chinese suppliers are presented with a unique opportunity to capture spillover demand.

Goldman Sachs argues that the core competitive advantages of Chinese suppliers lie not only in cost, but more importantly in **shorter delivery times and competitive product quality**. Specifically:

1.  **Delivery Speed**: This is the decisive factor enabling Chinese suppliers to seize opportunities. Faced with multi-year wait times, end customers are prioritizing "delivery speed" over historical supplier preferences. Chinese suppliers that can provide reliable products and drastically shorten lead times will be favored by the market.

2.  **Product Competitiveness**: The report emphasizes that opportunities are limited to a small number of leading Chinese enterprises that meet international standards in technical strength, quality control and scale production capacity. These companies typically possess expertise in high-voltage direct current (HVDC) technology, high-density power conversion and other areas, with products that can meet the stringent technical standards and rapid scaling needs of AI data centers.

Analysts believe that this market entry, based on speed and product competitiveness, can translate into significant financial gains.

For key product categories, Chinese enterprises can achieve a pricing premium of **10% to 80%** compared with domestic sales. Although they need to bear additional tariffs and logistics costs, the substantial premium margin enables export businesses to achieve significant gross margin expansion, bringing highly predictable profitability improvements for suppliers.

Goldman Sachs identifies Chinese suppliers with global competitiveness that are well-positioned to seize these opportunities, focusing on the following six product categories:

1.  **Gas Turbine Blades**: Leveraging high-end manufacturing technological barriers in single-crystal superalloys, special coatings and complex aerodynamic designs, these products are among the tightest and most technically challenging links in the global supply chain.

2.  **Power Transformers**: Due to their high customization, labor intensity and lengthy certification cycles, these products are the most severe bottleneck on the grid side. Chinese suppliers have a key advantage in meeting overseas demand thanks to their shorter delivery cycles.

3.  **HVDC Relays**: As critical components for upgrading data centers to 800V HVDC architecture, suppliers with advanced HVDC expertise and technical capabilities hold high-visibility global market share in this segment.

4.  **Aluminum Electrolytic Capacitors and Supercapacitors**: These electrical components are fundamental elements in data center power systems (such as UPS, rectifiers, power factor correction, etc.) and respond quickly to demand driven by AI data centers. Supercapacitors, in particular, serve as accelerators for peak load regulation in the medium to long term.

5.  **Uninterruptible Power Supply (UPS)/DC Power Rack Systems**: Through original design manufacturer (ODM) models, these suppliers can provide high-power supply solutions for overseas markets (especially the U.S.) and are actively verifying and developing next-generation 800V DC architecture products.

6.  **Liquid Cooling Solutions**: As the power density of AI servers rises sharply, demand for high-efficiency liquid cooling systems (including cold plates, manifolds, CDU/RPU, etc.) for GPUs/ASICs and other chips is growing rapidly, offering these products structural growth potential and value enhancement.

The above insightful content is sourced from Zhuifeng Trading Desk.

---

## Risk Warning and Disclaimer

The market is risky and investment requires caution. This document does not constitute personal investment advice, nor does it take into account the specific investment objectives, financial situations or needs of individual users. Users should consider whether any opinions, views or conclusions contained herein are consistent with their specific circumstances. Any investment made based on this document shall be at the user's own risk.