part 3: the Heat has to go somewhere
Reading Data Center Frontier's coverage of liquid cooling, I found the most useful comparison was between complete thermal systems, not competing claims about a single cold plate. Different operators were choosing different architectures according to rack density, facility design and their appetite for operational change. That is the discussion I want to have here.
So the key to understanding cooling is: follow the heat. Start at the processor and keep going until the heat leaves the facility or reaches a useful destination. Permission to compute depends on that whole journey. A good result at the first connection is not the final answer.
Begin at the chip
The first choice is how to collect the heat. NVIDIA's enterprise guidance includes air cooled systems as well as denser configurations requiring different power and cooling capabilities. I would match the design to the workload and building before declaring one architecture the winner.
For direct liquid cooling, just like in your favorite car, cold plates bring the cooling circuit to the components producing the heat, while manifolds and coolant distribution units connect those circuits into a larger installation, as Vertiv's portfolio illustrates. That is already several pieces of equipment to coordinate.
Single phase cooling keeps the fluid liquid through the cooling process; CoolIT Systems's water glycol cold plate work is one example described by Data Center Frontier. The operating familiarity of the complete installation would matter to me alongside its thermal specification.
Two phase cooling uses a change of state to carry heat away, as the fluid boils and is condensed for reuse in systems such as ZutaCore's HyperCool. Different mechanism, same work. The heat still needs a path beyond the chip.
The distribution system earns its place
The CDU is a connection between parts of the thermal system. ZutaCore describes configurations that circulate dielectric fluid through chip cold plates, condense the returning vapor and transfer heat either through an air based arrangement or to facility water. I would specify the exact configuration before discussing site requirements.
At the facility level, Vertiv supplies chilled water systems, chillers, rear door heat exchangers and refrigerant based systems alongside its liquid distribution products. A cooling procurement decision can therefore extend well beyond the server vendor.
That is where I would ask about responsibility. Who checks fluid compatibility, pressure limits and connections? Who owns the handover between the server circuit and the building? I would want a single commissioning plan even where several suppliers contribute equipment.
Israel's cooling contribution
ZutaCore, which has R&D in Israel, provides a direct comparison with global suppliers in the two phase chip cooling category. Its HyperCool design uses a dielectric fluid and controlled boiling at the chip, with different CDU arrangements for rack and larger installations. That is a defined engineering role.
There is a related comparison with Accelsius, whose NeuCool systems also use two phase cooling and dielectric refrigerant. I would compare integration, service procedures and qualification evidence rather than reduce the decision to a single efficiency claim.
The water boundary deserves its own question. ZutaCore offers configurations without facility water as well as configurations transferring heat to facility water, so “waterless” at the chip does not describe every possible installation. Ask where the fluid circuit ends.
Water technology companies are lining up to fill the cooling gaps with monitoring, management and control systems. Fortunately, Israel has a long history of water management technologies and processes that can easily be accessed. In addition to IoT technology companies like Datastream IoT (via its Miltel subsidiary), Ayyeka Technologies and IDEX that have built monitoring solutions for distributed assets, companies like Watergen and H2OLL are harvesting water from the atmosphere so that remote location data centers have access to nature’s primary coolant.
For me, that is ordinary diligence. I want to know what is inside the rack, what is elsewhere in the building and what must be installed outdoors. The site drawing should make the terminology unambiguous.
Coolant health is another technology
The fluid itself has to be managed. In August 2026, Ecolab introduced direct chip cooling monitoring covering concentration, temperature, pH and flow, supported by laboratory analysis and field service. That creates a different supplier relationship from buying the CDU.
There is a second distinction within that offering. Ecolab separately describes cooling tower, chiller and evaporative cooling water programs. I would keep the coolant circuit and the facility water treatment plan separate in the review.
Leak monitoring adds another operational layer. WINT monitors chilled water lines, mains and other facility water systems, using anomaly detection with alerts and automatic or remote shutoff. Its role is to help protect the facility around the cooling installation.
I would want any shutoff strategy reviewed with the cooling operator. What can be isolated safely? What must remain available while a fault is investigated? Detection is valuable, but the response needs an owner.
Our view
I would judge cooling technology on three levels: component performance, complete system integration and serviceability. The first can be demonstrated on a test stand. I would ask for separate evidence on the other two.
The commercial questions follow from that. How long does qualification take? Who holds replacement parts? What happens during maintenance? I would give more weight to a supplier that answers those questions clearly than one that wins every comparison on its own presentation.
Final Thoughts
Follow the heat, then follow the maintenance procedure. That is how I compare global systems with Israeli chip cooling technology and the monitoring around it. Permission to compute depends on a thermal path the operator can keep working, not just a processor temperature achieved in a demonstration.
