News and Information
Insight into forward-looking trends, brand market dynamics
Interpreting Cloud Computing Hardware Trends: Why Are Custom Servers Becoming the Choice of Major Players?
Ten years ago, when major internet companies purchased servers, they almost exclusively chose standardized products from branded vendors.
But today, if you walk into the data center of any leading cloud service provider, you are far more likely to see “white-box” servers carrying the customer’s own logo — systems custom-built by professional OEM/ODM manufacturers according to specific client requirements.
Customized servers are rapidly becoming the mainstream choice in cloud computing infrastructure.
To understand this trend, we first need to look back at the evolution of the server industry.
For many years, the server market was dominated by international brands such as:
Dell
HP
IBM
Purchasing a server was much like buying a PC:
Choose a configuration from a catalog, place the order, receive the equipment, and deploy it.
The advantage was simplicity.
The disadvantage was rigidity.
Everything was fixed:
Chassis structure
Motherboard layout
Fan placement
Expansion architecture
Customers had little to no ability to modify the design.
As internet companies scaled rapidly, they began realizing that branded servers carried significant premium costs.
Many enterprise features were unnecessary for their workloads, yet customers still had to pay for them.
As a result, technically capable companies began purchasing directly from contract manufacturers:
Designing configurations themselves
Managing their own supply chains
Deploying white-box servers at scale
This reduced costs significantly.
However, it also required:
Dedicated hardware engineering teams
Supply chain expertise
Internal validation capabilities
The barrier to entry remained high.
Today, the industry has entered its third phase.
Large cloud companies no longer want to simply “choose configurations.”
Instead, they deeply participate in the server design process itself:
Motherboard layouts
Thermal airflow architecture
Firmware functionality
Management interfaces
Rack integration
Power optimization
Every detail can now be tailored to specific workloads and operational requirements.
Behind this transformation are professional ODM manufacturers providing full-stack engineering support.
Several powerful forces are driving this shift toward customization.
This is the most direct motivation.
Branded servers are typically designed with significant redundancy to satisfy a broad range of customers.
For example:
A branded server may include dual 1600W power supplies when the workload only requires 800W
A storage backplane may support eight drives while the customer only uses four
All of this unused capability becomes unnecessary cost.
Customized servers allow truly demand-driven design:
Only the required CPU performance
Only the necessary storage capacity
Only the needed PCIe expansion
Every dollar is invested with purpose.
One leading cloud provider publicly stated that deep customization reduced per-server hardware costs by more than 20% compared to equivalent branded systems.
For organizations operating millions of servers, this translates into savings worth billions.
Customization is not only about reducing cost — it also enables higher performance.
Standardized products must prioritize broad compatibility and conservative stability.
Customized systems can instead be optimized specifically for the customer’s workloads.
Examples include:
A short-video platform optimizing storage systems for read-heavy workloads, improving read performance by 30%
An e-commerce platform redesigning network architecture to better handle extreme traffic spikes during major sales events
These kinds of workload-specific optimizations are impossible with generic off-the-shelf systems.
Large internet companies often manage:
Hundreds of thousands of servers
Massive distributed infrastructure
Operational efficiency becomes critically important.
Customization allows operational requirements to be embedded directly into hardware design.
For example:
Standardized fault indicator colors and blinking patterns for rapid maintenance identification
Uniform asset label placement for automated inspection robots
These details may seem minor individually, but at hyperscale they produce major operational gains.
Under global carbon reduction initiatives, data center energy efficiency has become increasingly important.
Customized servers offer natural advantages in this area.
By precisely matching workloads:
Unnecessary compute resources are eliminated
Fan power consumption is reduced through optimized airflow
Higher-efficiency power modules improve overall energy utilization
One cloud provider reported that through deep hardware customization, their data center PUE improved from 1.5 to 1.3, reducing annual electricity expenses by more than RMB 200 million.
Customization itself is not one-size-fits-all.
Depending on customer involvement, it can generally be divided into four levels.
This is the simplest form.
Customers select combinations from existing OEM product lines:
CPU models
Memory capacity
Storage configuration
The manufacturer handles assembly and validation.
Beyond configuration, customers can customize:
Chassis colors
Front panel structures
Logo printing
This model is common for customers with branding requirements.
Customers specify physical design requirements based on their data center environment, including:
Chassis depth
Rail specifications
Cable routing
Rack integration
This level requires stronger engineering capability from the manufacturer.
This is the deepest level of customization.
Customers participate throughout the full development process:
Motherboard design
Firmware development
Thermal architecture
Management software
Products are essentially designed from scratch according to specific business requirements.
While the barriers are higher, the long-term benefits are also significantly greater.
Today, leading cloud providers commonly adopt Level Three and Level Four customization models, collaborating deeply with ODM manufacturers to co-develop next-generation server platforms.
The rise of customization is fundamentally reshaping the role of ODM manufacturers.
In the past, ODM providers primarily acted as contract manufacturers:
Customers supplied the designs
Manufacturers handled production
Today, ODM providers are evolving into full solution partners.
Customers define requirements.
ODM manufacturers now provide:
System design
Product development
Manufacturing
Testing
Validation
Deployment support
This transformation raises the bar significantly.
Modern ODM manufacturers must possess:
Strong R&D capability
Advanced supply chain management
Comprehensive quality control systems
Fast-response engineering support
Manufacturers unable to meet these demands are gradually being eliminated from the market.
Meanwhile, companies with true system-level design capability are entering a new growth phase.
Looking ahead, what will the next stage of customized servers look like?
We believe the future lies in deeper software-hardware integration.
In the past, customization focused primarily on hardware.
In the future, software will increasingly define hardware.
Hardware will be designed around software requirements — and may even dynamically adapt to changing workloads.
Examples already exist:
Google’s TPU was specifically designed for AI workloads
Amazon’s Nitro system offloads networking, storage, and security functions to dedicated hardware, freeing CPU resources
These are classic examples of deep software-hardware co-design.
For server OEM/ODM manufacturers, this means:
Understanding hardware alone is no longer enough
System software expertise is equally important
Future leaders must know not only how to design motherboards, but also how to optimize entire computing systems.
This represents both a major challenge and a tremendous opportunity.
Related Recommendations
Learn more news and information
