AI infrastructure is often framed as a race for more chips and more electricity. That overlooks a third constraint: every unit of compute eventually becomes heat, and removing that heat can consume a substantial share of a data center’s energy.
Hitachi Vantara has deployed an operational phase-change material, or PCM, installation at its data center in the Netherlands using technology developed by Duraflow. The measured results suggest that better thermal management can do more than lower energy use. In a power-constrained location, it can turn electricity that would have powered cooling equipment into additional capacity for computing.
Why power is becoming the bottleneck
The important question for AI infrastructure is not simply how much electricity an operator can obtain. It is how much useful computing capacity can be produced from the electricity already available.
Hitachi Vantara faced that problem as demand increased for high-density computing and AI infrastructure. According to PR Newswire, the company explored ways to make its existing electrical capacity work harder instead of assuming that additional computing always required additional power.
That distinction matters because thermal management can demand as much as 40% of a data center’s total energy needs. Cooling is therefore not merely a facility-support function. In a constrained power environment, it directly competes with computing for both electricity and capacity.
The practical implication is straightforward: reducing the energy required to manage heat may allow more of an existing power connection to support AI workloads. The benefit comes from making constrained infrastructure more productive, not merely from lowering an energy bill.
Phase-change materials in practice
PCM technology stores thermal energy through a change of state. Hitachi Vantara’s installation demonstrates the approach at operational scale, rather than presenting it only as a controlled pilot.
The system provides 240 kW of cooling capacity through two 120 kW PCM units. Operational monitoring shows that the units collectively consume 5–8 kW of electricity, depending on operating conditions.
| Cooling measure | Hitachi Vantara and Duraflow PCM installation | Conventional DX comparison |
|---|---|---|
| Cooling capacity | 240 kW | 240 kW |
| Electrical consumption | 5–8 kW | Estimated 60–80 kW of compressor power |
| Cooling water | No water required for cooling | Not specified |
| Evidence basis | Operational monitoring data | Estimated comparison |
These figures are application-specific, not a universal efficiency guarantee. Even so, they illustrate why thermal management can become a capacity strategy: the same 240 kW of cooling requirement can be met with a markedly different amount of electricity.
The installation is also described as one of the world’s largest operational PCM installations. Its importance lies in providing a real-world proof point for data-center operators evaluating whether the technology can work at meaningful scale.
Why it matters for AI growth
The wider infrastructure chain is becoming more visible. A Crypto Briefing report on DTE Energy describes a utility that has executed 2.4 GW of power agreements with hyperscale data-center operators. It also reports a pipeline of approximately 8.4 GW when earlier-stage opportunities are included.
Demand for power and demand for efficient power use are two sides of the same issue. Utilities can secure larger data-center commitments, but operators still need to turn those connections into productive computing infrastructure. If too much available electricity is consumed by cooling, the original power agreement does not translate cleanly into additional usable compute.
This is where the Hitachi Vantara deployment matters beyond its own facility. It provides evidence that thermal management can be treated as part of the AI capacity plan rather than as a secondary facilities decision.
The likely winners are operators and technology providers that can increase the productive output of existing infrastructure. The main risk is treating application-specific performance figures as universal. Operators would need to compare local climate, workload behavior, operating conditions and available power before assuming the same results.
How to evaluate the opportunity
A data-center operator should begin with the binding constraint rather than a predetermined cooling upgrade. If electricity is constrained but the facility already has an approved connection, the relevant question is how much of that connection is being consumed by thermal management.
The next step is to compare the electrical demand of the existing cooling arrangement with a PCM option serving the same required capacity. Hitachi Vantara’s installation provides a useful benchmark, but operators should not transfer its figures to another site without local analysis.
The business case should also include deployment timing, actual energy use and the resulting computing capacity. A thermal-management upgrade is most compelling when it allows high-density workloads to operate sooner or more consistently within an existing power connection.
The broader lesson extends beyond PCM. As AI demand raises the value of power, energy efficiency becomes an infrastructure growth strategy. Companies that help customers produce more from scarce electricity are addressing the constraint that can delay deployment.
What to watch next
The next test is sustained operating performance. Hitachi Vantara’s figures should be evaluated over time, across changing conditions, rather than treated as a one-time result.
Watch whether other operators adopt PCM for comparable workloads, whether vendors can reproduce the operational savings in different environments, and whether cooling water requirements remain absent across applications. Also watch whether thermal efficiency becomes a standard part of AI capacity planning.
For utilities such as DTE Energy, the relevant signal is whether proposed data-center projects progress from discussion into signed agreements. For cooling and infrastructure providers, the signal is whether operational evidence leads to wider deployment. For data-center operators, the ultimate measure is simple: more useful computing capacity from power that was already available.
Key takeaways
- Thermal management can consume as much as 40% of a data center’s total energy needs.
- Hitachi Vantara’s PCM installation delivers 240 kW of cooling while consuming 5–8 kW of electricity.
- Because PCM can reduce thermal-management electricity by up to 90% depending on the application, it can release existing power capacity for computing.
- Energy efficiency becomes an AI growth strategy when grid access, rather than demand for computing, is the limiting factor.
Frequently asked questions
How can data centers increase AI capacity without always securing more electricity?
They can improve thermal management so more available power is used for computing rather than cooling. In constrained locations, that can make an existing power connection more productive.
What is phase-change material technology?
PCM stores thermal energy through a change of state. Hitachi Vantara deployed an operational installation using Duraflow’s technology at its data center in the Netherlands.
How much energy can the approach save?
PR Newswire says PCM technology can reduce electricity used for thermal management by up to 90%, depending on the application. Hitachi Vantara’s two 120 kW units collectively consume 5–8 kW while providing 240 kW of cooling.
Does PCM cooling require water?
The reported approach requires no water for cooling. Operators would still need to assess site-specific conditions before applying the result to a different facility.
Why does this matter to AI infrastructure investors and operators?
Computing expansion depends on the entire power-and-cooling system. Improving thermal efficiency can increase usable capacity within an existing power connection and make AI infrastructure expansion less dependent on grid availability. END: