Data Center Nation Warsaw 2026: 600 kW Racks and Liquid Cooling in Poland

Today I am at Data Center Nation Warsaw with my colleague Kateryna Kravtsova, electrical engineer and project manager at TEBIN. The agenda at Expo XXI is full and the conversations have already started. The data centre market is moving fast: new capacity, artificial intelligence (AI) infrastructure and growing power demand are creating new engineering challenges across Europe. The question I came with is what the next wave of development asks of the engineers who design it. These are my notes from four sessions, with the speakers’ own numbers, read by a mechanical engineer who leads a design and engineering company’s work in Poland.

Where is the power, and how fast is the Polish market growing?

Faster than its reputation. Schneider Electric’s session put last year’s Polish market at around 140 data centres drawing roughly 200 MW, which matches CBRE’s Europe Data Centres report for the second quarter of 2026: about 200 MW today, around 500 MW expected by 2030. The second session added that roughly 80 percent of that capacity sits in the Warsaw area and that the market is now actively searching for power elsewhere. Poznań was the example, where Beyond.pl’s campus is growing past 100 MW for AI loads; Beyond.pl and the Polish Data Center Association (PLDCA) were both referenced.

The operators spoke in concrete numbers. Equinix’s speaker expected about 60 MW in the Warsaw area by the end of the year and wants to expand further, with 100 MW outside Warsaw and possibly 300 MW. A 300 MW AI campus running on gas turbines was raised in the same session and judged beyond what Poland is ready for today. Atman’s speaker said WAW-3 has just opened and that the company is looking for land for a WAW-4 of more than 100 MW at higher density.

Nobody called power availability itself the blocker in Poland. Price is: at around 16 euro cents per kilowatt-hour, Polish energy is expensive compared with the Nordic markets that attract the most power-hungry workloads, although those markets are small. Atman named Poland’s design and engineering capacity as its strength and the power price as its weakness, and said the same weakness applies to Germany, France and the Netherlands, which is why Spain and Portugal were named as the new markets to watch.

Two things give Poland a longer runway. Speakers described the country as a major power construction site today, with gigawatts of new generation expected from the end of 2027 and nuclear planned for the longer term. And Schneider Electric’s session quoted the grid operator’s plan of around 5,000 km of new transmission lines and 10 GW of new connections.

AI redraws the design map: power, permit, scale, speed

The second session put the new sequence into four words: power, permit, scale and speed. Where a site has power and a permit, the questions are how large it can go and how fast it can be built. Nine months from start to a working data centre was described as the current must, and six months as the new aim. The main technical focus on the IT side is the next generation of graphics processing units (GPUs), with NVIDIA setting the pace.

On the policy side, Equinix argued that public sector and financial data must stay in Poland, which keeps a floor under domestic demand for both hyperscale and colocation capacity, and that Poland has the knowledge, the resources and enough investors for AI data centres, so if it does not enter the AI race now it will be late. The national digitalisation strategy has been published, and speakers said a national AI strategy is awaiting government approval, expected at the end of 2026 or early 2027. One speaker in the Schneider Electric session counted Poland at about 2 percent of the world’s data centres by number.

For engineers this moves the project boundary outward. A data centre design today starts at the grid connection, not at the hall door. TEBIN has already worked on the grid side, with the civil and electrical basic design of a grid substation of over 200 MW at 150/20/10 kV serving a data centre cluster in the Netherlands, and the same logic applies to the Polish pipeline.

From 10 kW to 600 kW racks: what changes downstream of the chip

Everything. For years a typical rack drew well under 10 kW, and the Uptime Institute’s 2025 survey still finds the most common deployed rack density below 10 kW. A current AI training rack such as NVIDIA’s GB200 NVL72 draws roughly 120 to 130 kW and is liquid-cooled by design. Schneider Electric’s session went further: racks of up to 600 kW, the figure NVIDIA has announced for its next rack generation, are now on the roadmap, where 150 kW was the ceiling a short time ago.

Air cooling does not scale across that range: moving enough air through a 130 kW rack, let alone a 600 kW one, needs velocities and temperature differences that neither the IT equipment nor the people in the hall can tolerate. Andrew Green, a regional data centre practice lead at JLL, put it plainly: rack densities have gone beyond what the physics of air cooling allows. The industry response is direct liquid cooling, and the Uptime Institute’s 2025 cooling survey found roughly one in five operators already using it, with the majority of the rest considering it. The strongest driver is not energy efficiency. It is heat that has nowhere else to go.

Liquid cooling as a hydraulic design problem

Liquid cooling turns the white space into a hydraulic system and brings the mechanical engineer into the rack. In direct-to-chip cooling, cold plates sit on the processors and accelerators, a technology cooling loop carries the heat to a coolant distribution unit (CDU), and the CDU exchanges that heat with the facility water loop. Each of those three loops has its own pressures, temperatures, water quality requirements, leak detection and redundancy logic, and they have to be designed as one system.

The facility water temperature is the decision that shapes the whole plant. ASHRAE’s thermal guidelines for liquid-cooled equipment define facility water classes by their upper supply temperature, from W17 up to W45 and W+ above 45 °C. The lower classes imply chillers. W40 can run almost year-round on dry coolers in Central Europe, while W32 still needs trim cooling on summer peaks. W45 and above are the band where waste heat becomes genuinely useful. Choosing the class early, with the IT vendor’s cold plate limits, decides whether the roof carries chillers or dry coolers, how large the pipework is, and the building’s energy balance.

Schneider Electric’s message on this point was the one I would underline: for the foreseeable future, the design is always a combination of air and liquid cooling components. Memory, drives, power supplies and network switches still reject a share of their heat to the room, so every liquid-cooled hall is a hybrid hall with two systems that must be sized and controlled together. The second consequence is weight. JLL puts the floor loading for liquid-cooled, high-density equipment at 20 kPa or more, compared with the 12 to 15 kPa that conventional halls were designed for. That is a structural question as much as a mechanical one, and in a retrofit it can decide the project.

Why does cooling pull power, structure and fire safety along with it?

Because every kilowatt of cooling capacity has an electrical, structural and life-safety consequence. Pumps and CDUs become part of the critical path and need the same redundancy as the fans they replace. The electrical distribution behind a 130 kW rack has to deliver that power through busbars and switchgear that take floor area and clear routes. Coolant pipework in a data hall changes the fire strategy and the leak scenarios. All of it has to be maintainable under customer load.

This is where TEBIN has delivered much of its data centre work, as design partner to lead engineering firms in Germany and the Netherlands: electrical, building management and fire safety design for the conversion of a printing works into a data centre of under 10 MW, a permanent load bank strategy for a facility in full operation, and fuel and urea systems for five data centre projects. None of these projects was about cooling alone, and none of them could have been designed one discipline at a time.

What do the standards and regulators ask for?

A documented baseline, measurable efficiency and measurable heat reuse. The EN 50600 session covered the European standard series for data centre facilities and infrastructures, now in its regular five-year review. The message was that every new facility should be designed against it, from power distribution and environmental control to physical security. Two caveats were noted: the standard does not cover cybersecurity, and physical security is no longer an abstract chapter in Poland. The war across the eastern border makes the security of sites in the east of the country a real design input.

On the regulatory side, the European Union’s delegated regulation 2024/1364 requires data centres above a given size to report annually on energy use, power usage effectiveness (PUE), water usage effectiveness (WUE) and the energy reuse factor (ERF), and Poland is inside that scheme. Germany’s Energy Efficiency Act goes further: new data centres commissioned from 1 July 2026 must achieve a PUE of 1.2 or lower and reuse a rising share of their waste heat, from 10 percent to 20 percent for facilities commissioned from 2028. An amendment adopted by the federal cabinet in June 2026 keeps the 1.2 cap while softening the heat reuse obligations and still has to pass parliament, so the exact requirement should be checked at the start of every project.

A Polish project team should watch those rules even where they do not apply yet: they show where the European baseline is heading, and liquid cooling at W40 or W45 is the technical bridge, because its return water is warm enough to feed a heat network, which the exhaust air of a conventional hall is not.

The next wave: integration, earlier, and faster

The projects coming to Poland and Central Europe will be denser, hotter, heavier, more tightly regulated and built on shorter programmes than the colocation halls of the last decade. Flexibility and the grid were named as the main points everywhere, and interconnection matters as much as capacity: data needs to flow, so fibre routes between countries are part of the site decision, which is why Portugal’s direct connection to Africa came up in a Warsaw conference room.

The engineering answer is a different sequence: grid connection, cooling class, rack density, floor loading, heat export, physical security and fire strategy decided together at concept stage, held in one coordinated Building Information Modeling (BIM) model, and carried through to documentation the contractor, the commissioning team and the operator can build and run from. If nine months is the must and six months is the aim, there is no time left for finding these answers on site. That is the work TEBIN does as a design and engineering company, from our Polish office in Szczecin and our teams across Europe. If you are at Data Center Nation Warsaw today, come and say hello. Kateryna and I are here to talk about projects, engineering, and exactly these questions.