As the power density of AI and HPC systems continues to rise, liquid cooling is moving from a high-end option to a default architecture. GPUs and CPUs have already pioneered this transition, with storage (SSDs) emerging as the next critical component.
It’s not about whether SSDs are three seconds faster than other types of storage, but about a more systemic problem. When compute fully enters the era of liquid cooling, what happens to performance, energy consumption, and ROI if we continue to use air-cooled SSDs?
Read on to learn how Solidigm solved this system-wide constraint with an engineered liquid-cooled SSD solution.
Data storage is facing a serious imbalance. As the PCIe protocol advances from 4.0 to 7.0, power consumption of SSDs per drive has jumped from 25W to 60W. In AI architectures in which the computing nodes are fully liquid-cooled, storage heat dissipation is still stuck in the old era of "blowing air." Between this leap in SSD power density and outdated heat dissipation methods, an obvious generational mismatch has formed, restricting the reliability boundary of the entire system.
| No Throttle | Threshold 1 | Threshold 2 | |
| SSD Temperature | <77°C | ≥77°C | ≥79°C |
| SSD Performance | 100% | ~58% | <1% |
Table 1. SSD temperature and performance thresholds
What does this mean? Once it reaches the upper thermal limit, the data channel is almost completely cut off.
In AI systems, this behavior is not just a local problem, but one that can be quickly amplified.
Figure 1. Impacts of SSD temperature increase
The result is not simply that "SSDs are slower," but rather that the most expensive GPUs in the AI server start idling.
Figure 2. Limits of air cooling
When SSD power consumption moves toward 40W to 60W:
This not only increases cooling costs but also limits system density.
When the GPU and CPU switch to liquid cooling:
The result: SSDs are among the most vulnerable components to heat-related impacts in the most expensive AI servers.
Figure 3. Solidigm fully-liquid-cooled SSD solution
The core challenge of liquid-cooled SSDs is not about attaching a cold plate to the drive. Instead, it lies in:
Ultimately, this requires a system-level rethinking of liquid-cooled storage architecture.
In conventional E1.S SSDs, the controller, DRAM, and NAND components are distributed across both sides of the printed circuit board (PCB). Attaching a cold plate to only one side of the drive leaves the opposite side as a thermal blind spot, creating an inherent limitation in heat removal.
Addressing this challenge requires architectural changes to pull heat from both sides of the PCB using a single cold plate. This not only improves the SSD thermal performance but also increases system level heat recovery from liquid.
In real-world operations, liquid cooling introduces a fundamental tension. Applying sufficient pressure between the cold plate and the SSD is necessary for effective thermal conduction, but overly tight coupling eliminates hot-plug capability.
By introducing a precision-engineered mechanism, it is possible to maintain firm physical contact between the SSD and the cold plate for optimal heat transfer while still allowing instantaneous release during maintenance. This enables hot-plug and serviceability without system shutdown or coolant disconnection, achieving an operational experience comparable to air-cooled environments.
Sustaining a high level of serviceability is critical. Without it, liquid-cooled SSDs risk becoming impractical for data center deployment instead of a viable production solution.
Standardizing liquid-cooled SSDs is a critical step toward enabling high-density, energy-efficient, fully liquid-cooled data center architectures. Leading industry players are actively driving the development and adoption of formal specifications. Solidigm, in collaboration with SNIA, has released the SFF TA-1006 (Rev 2.0) specification, which defines the liquid-cooled implementation for the E1.S form factor and marks the transition from proprietary, custom-built solutions to open, reusable industry standards.
The specification tightly defines parameters such as cold plate contact area, surface flatness, and roughness tolerances to ensure predictable and repeatable thermal performance. In addition, features such as chamfered edge design address mechanical insertion and removal challenges under high contact pressure conditions.
The core value of this effort lies in transforming liquid-cooled SSDs from case-by-case custom engineering into standardized, mass-deployable components establishing a solid foundation for industrial-scale adoption in data centers.
Looking ahead, ecosystem-level collaboration will be essential to accelerating the standardization and large-scale adoption of liquid-cooling technologies in next-generation data centers. Founded in 2014 the ODCC, has become a leading open platform in China’s data center industry. ODCC has actively promoted technical standardization and innovation across servers, networking, liquid cooling, edge infrastructure, and data center facilities. It has established a wide range of mature open technical specifications.
By working closely with ODCC, enterprises can better align their liquid-cooling products and solutions with advanced industry standards, gain access to timely technical insights, and benefit from ODCC’s ecosystem influence and platform credibility. Such collaboration can help accelerate market acceptance, improve solution trustworthiness, and support the broader deployment of efficient, reliable, and sustainable liquid-cooling infrastructure.
By publishing this article via ODCC channel,1 ODCC is helping to share insights from Solidigm into industry-leading liquid-cooled SSD design with the data center ecosystem increasingly focused on scalable, efficient liquid-cooling adoption.
Figure 4. SSD cold plate design improvements
Through liquid cooling, SSDs can operate in the non-throttling range for extended periods, ensuring a continuous and stable data supply. This stability directly eliminates GPU idle time (wait states) caused by storage-side cooling bottlenecks which allows compute resources to remain fully utilized. From a system perspective, improved ROI is not a separate objective, but a natural outcome of the physical alignment between the storage architecture and computing requirements.
Figure 5. Advantage of cold plate cooling
At the server level—and even across the entire data center—liquid-cooled SSDs deliver not just localized optimization, but overall improvement in system efficiency. Compared with traditional methods, liquid cooling requires much less energy usage than air cooling under the same heat dissipation requirements, and when combined with a higher-density, fanless design, it greatly reduces rack space needs and the complexity of ongoing operations and maintenance. This architectural streamlining translates directly into meaningful improvements in data center energy efficiency, creating gains from the component level all the way to the overall infrastructure.
In today’s AI architectures, the GPU is responsible for delivering compute, while storage is responsible for maintaining data supply. Your cooling solution determines whether these two components can cooperate continuously with stability in real-world deployments. Solidigm liquid-cooled SSD solutions are designed not merely to cool a component, but rather to bridge the generational gap between computing and storage. Our approach allows the entire AI system to achieve a stable physical balance between extreme performance, overall energy efficiency, and architectural consistency. When this deep system-level coupling is achieved, improved ROI is no longer a separate business metric, but an inevitable result of this technical balance.
Wayne Gao is a Principal Engineer and Solution Storage Architect at Solidigm. He has worked on Solidigm’s Cloud Storage Acceleration Layer (CSAL) from pathfinding to commercial release. Wayne has over 20 years of storage developer experience, has four U.S. patent filings/grants, and is a published EuroSys paper author.
Bo Li serves as a senior storage solutions architect at Solidigm. With over two decades of experience in system design and development across multiple organizations, he specializes in optimizing the performance of networked and storage solutions. In recent years, Bo has concentrated his efforts on advancing the industry-wide adoption of non-volatile storage technologies.
Hardeep Singh is a Thermal Mechanical Engineering Manager and Principal Engineer focused on next-generation SSD thermal mechanical solutions for data center application. With 18+ years of experience across semiconductors, turbomachinery, power plants, and electronics cooling, he leads teams developing air-, direct-liquid-, and immersion-cooled SSD technologies. Hardeep holds an M.S. in Mechanical Engineering from Arizona State University.
Subscribe for product updates, technical resources, and insights on accelerating AI workloads with Solidigm Storage
