Free tools Windows power users keep installed
One-click scans. No signup required.
COOLERCHIPS is an ARPA-E research and development program designed to make data-center cooling more energy-efficient, reliable, and affordable as computing loads rise. Its headline target is to bring total cooling energy below 5% of a typical data center’s IT load for a high-density compute system, at any time and any U.S. location. That figure is a program goal—not a reported portfolio-wide result. A Department of Energy notice dated August 26, 2026 describes a follow-on phase, COOLERCHIPS 1.5, in which selected teams are expected to expand, test, and validate cooling systems for AI heat loads of up to 1 megawatt per rack.
Why data-center cooling is a harder problem at AI scale
Servers convert nearly all of their electrical input into heat. Cooling equipment must capture that heat at the chip or server, move it through one or more liquid or air loops, and reject it to the surrounding environment. As processors become denser, the cooling system consumes more power and can become a constraint on reliability, facility capacity, and operating cost.
The U.S. Department of Energy’s 2023 funding announcement reported that data centers accounted for approximately 2% of total U.S. electricity consumption and that cooling could represent up to 40% of data-center energy use. Those figures are the scope and date of that announcement; they should not be read as a new 2026 measurement.
COOLERCHIPS addresses the thermal system around computing equipment. Its funding opportunity excludes chip design and cooling inside the chip itself, so it is not a general semiconductor-architecture program.
What COOLERCHIPS is trying to achieve
A system-level energy target
ARPA-E states a goal of reducing total cooling energy expenditure to less than 5% of a typical data center’s IT load for a high-density compute system, regardless of time or U.S. location. This is a performance target for research and development, not evidence that every funded design has reached it.
Lower thermal resistance
The program also describes a design aim of reducing thermal resistance so coolant can operate closer to chip temperature. Its stated chip-to-coolant temperature-difference target is below 10°C. That is an engineering objective, not a universal operating specification or a demonstrated result for a particular project.
Efficiency without sacrificing uptime
COOLERCHIPS frames efficiency, reliability, availability, and total cost of ownership as connected requirements. A system that saves electricity but increases failure risk or maintenance burden would not satisfy the program’s broader purpose.
The four parts of the cooling chain in the program
Secondary-loop components
These components move heat from server electronics toward facility water or another primary cooling loop. Improvements here can reduce pumping, heat-exchanger, and temperature penalties between the IT equipment and the building system.
Modular and edge data-center systems
This track covers integrated cooling from facility water to ambient conditions in smaller or modular data centers. Such systems must work within tighter space, power, and deployment constraints than a large centralized facility.
Software and decision tools
COOLERCHIPS supports software that evaluates energy efficiency, reliability, and cost together. Modeling these factors in one design process can expose trade-offs that a cooling-only calculation would miss.
Rank #3
Testing facilities and protocols
New cooling concepts need repeatable measurements. The program therefore includes facilities and test methods for comparing thermal performance, energy use, reliability, and operating behavior under controlled conditions.
What the first project portfolio illustrates
On May 9, 2023, DOE announced $40 million for 15 COOLERCHIPS projects. The portfolio demonstrates that the program is not committed to one cooling method or one equipment scale.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problems| Example announced effort | Approach or role | What the announcement establishes |
|---|---|---|
| Intel Federal | Two-phase immersion cooling | An announced project direction; not proof of commercial deployment or a measured program-wide result. |
| JETCOOL | Microconvective cooling | An announced heat-transfer research approach; final comparative performance was not established in the cited material. |
| NVIDIA | Modular data-center cooling system | An announced system-level concept for modular facilities, not a retail product recommendation. |
| National Renewable Energy Laboratory | Testing protocols and a digital-twin effort | Work intended to improve evaluation and modeling of cooling systems. |
| University of Maryland | Integrated decision-support software | A tool-oriented effort to weigh efficiency, reliability, and cost in design decisions. |
These descriptions identify proposed or funded work. They do not establish that one approach won a head-to-head test, met the below-5% target, or is available for purchase.
What COOLERCHIPS 1.5 adds for AI data centers
DOE’s August 26, 2026 notice describes COOLERCHIPS 1.5 as a continuation for selected first-phase teams. It provides additional funding, extended periods of performance, and new milestones. The teams are expected to expand, test, and validate primary and secondary cooling loops for AI heat loads of up to 1 megawatt per rack.
The notice says ARPA-E will select a common test location for seven project teams. It also assigns the University of Maryland to provide software and testing support during final system evaluations. These are planned activities: the notice does not say that testing is complete or that a one-megawatt-per-rack system has already been validated in ordinary data-center operation.
The notice characterizes the continuation as development of advanced, water-free cooling systems for high-power AI data centers. “Water-free” is an objective for the systems being developed, not a claim that all COOLERCHIPS facilities or all data centers have eliminated water use.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
How the approaches can be compared
The available official descriptions do not provide enough common test results to rank the projects. A useful comparison instead asks where heat is captured, what boundary the system covers, and how evidence is being produced.
| Comparison axis | Questions to ask |
|---|---|
| Heat-capture location | Is heat captured at a chip or server component, transferred through a secondary loop, or managed across a modular facility? |
| Heat-transfer mechanism | Does the design use immersion, microconvective transfer, another liquid method, or a combination of technologies? |
| System boundary | Does the claim cover a component, a server or rack, a cooling loop, or the complete data-center facility? |
| Energy metric | Is the number for cooling equipment alone or for total facility energy, and what IT load and conditions were used? |
| Reliability and availability | How does the design behave during faults, maintenance, changing loads, and extreme ambient conditions? |
| Evidence stage | Is the result a proposed design, laboratory work, controlled system testing, or validated operation in a live data center? |
This framework prevents a component-level temperature result from being mistaken for a facility-level energy result.
Why reliability belongs beside efficiency
Cooling is part of the data center’s resilience plan. A pump, heat exchanger, control system, or coolant loop can become a single point of failure if redundancy and maintenance access are not designed in. Higher coolant temperatures may reduce chiller work, but only if servers remain within their safe operating envelope. Software that models cost and efficiency without availability can therefore select a design that is cheap to run but difficult to keep online.
DOE Secretary of Energy Jennifer M. Granholm framed the policy context this way: “Climate change, including severe weather events, threatens the functionality of data centers that are critical to connecting computing and network infrastructure that power our everyday lives,” she said in the May 9, 2023 announcement. The statement explains why resilient cooling matters; it is not a technical result from a COOLERCHIPS project.
What the program could change in practice
- More heat captured close to the source: reducing the distance and temperature penalty between processors and coolant.
- Lower parasitic power: improving pumps, heat exchangers, controls, and heat-rejection equipment so less electricity is spent moving heat.
- Cooling designed for dense racks: supporting AI systems whose heat output exceeds what conventional air cooling can economically handle.
- Better design decisions: using digital twins and integrated software to evaluate energy, reliability, and cost before construction.
- Comparable testing: creating protocols and shared facilities so different technologies can be measured under similar conditions.
Whether those changes deliver real-world savings will depend on validated results, deployment conditions, maintenance requirements, and the total facility boundary used for measurement.
Quick Recap
What is established—and what is not
- Established: ARPA-E’s below-5% cooling-energy and below-10°C chip-to-coolant figures are stated program targets.
- Established: DOE announced a first phase of 15 projects and $40 million in 2023.
- Established: COOLERCHIPS 1.5 is described in 2026 as extending selected work toward testing at up to 1 megawatt per rack.
- Not established by these sources: that all projects use the same cooling method.
- Not established: that any one approach is the winner, that the targets have been met portfolio-wide, or that a commercial product is ready for general purchase.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
