Vertiv just told the data center industry where it thinks the real constraint lives. And it is not at the chip. It is not at the rack. It is at the point where all that concentrated heat has to leave the building.
On March 23, Vertiv announced an agreement to acquire ThermoKey S.p.A., an Italian heat exchanger manufacturer founded in 1991 and headquartered in Rivarotta di Rivignano Teor, in the heart of northeastern Italy's HVAC-R corridor. ThermoKey makes dry coolers, air-cooled condensers, and microchannel heat exchangers. About 230 employees. Roughly EUR 60 million in 2023 revenue, up 20% year-over-year. All products designed and built in-house across 32,000 square meters of production floor space. Deal terms were not disclosed. Expected close: Q2 2026.
This is not a headline acquisition. No flashy price tag, no dramatic press tour. But if you understand what is happening to the thermal architecture of AI-class data centers, this deal makes a specific and aggressive bet about where the supply chain breaks next.
Vertiv CEO Giordano Albertazzi framed it plainly: "Heat rejection is becoming increasingly critical for data centers and AI factories as the industry seeks new ways to unlock capacity, improve energy efficiency, and scale with confidence."
Read that again. He did not say "cooling." He said "heat rejection." The distinction matters.
Liquid cooling, in all its forms, does one thing well: it absorbs heat at the source. Cold plates pull thermal energy off GPU dies. CDUs transfer that energy to a secondary loop. But that loop has to go somewhere. The heat has to physically leave the facility. Dry coolers, condensers, cooling towers, and heat exchangers on the roof or outside the building wall are where rejection happens. They are the final link in the chain. And that final link is becoming the bottleneck.
Vertiv is now building toward ownership of every stage of what it calls the "thermal chain," from chip-level liquid cooling through facility-level heat rejection. ThermoKey fills the gap at the far end of that chain. This is vertical integration with a purpose: sell a complete, optimized thermal system rather than a collection of components that customers have to stitch together on their own.
The math is straightforward and unforgiving.
NVIDIA's GB300 racks hit 163 kW per rack. Vera Rubin NVL144 systems are targeting 300+ kW per rack in 2026. Rubin Ultra NVL576 racks are projected to exceed 600 kW by 2027. The B300 GPU alone carries a 1,400-watt thermal design power. Seven years ago, top-end GPUs generated roughly 300 watts.
Liquid cooling handles those densities at the rack. That problem, while far from trivial, is being addressed by dozens of vendors with direct-to-chip and immersion systems. But here is the part the industry has spent less time talking about: liquid cooling does not eliminate heat. It concentrates it. It makes the heat denser, hotter, and more difficult to reject at the facility boundary. A single high-density AI cluster can dump thermal loads onto outdoor heat exchangers that were originally sized for a facility running 5 to 10 kW racks.
The outdoor rejection equipment at most existing sites was never designed for this. Neither was the piping infrastructure connecting indoor CDUs to outdoor dry coolers. Neither were the physical footprints allocated on rooftops or in mechanical yards. When a data center operator upgrades a few hundred racks from air-cooled CPU workloads to liquid-cooled GPU training clusters, the indoor cooling system might scale. The outdoor rejection system almost certainly does not.
Steven Dickens, CEO and principal analyst at HyperFRAME Research, put it bluntly: "With inference workloads set to overtake every other compute workload combined over the next 10 years, then how we move the heat generated by accelerated compute through the DC will be vital, not a nice-to-have."
ThermoKey is not a startup. The company has been building heat exchangers for over three decades, serving OEM and system integrator customers across EMEA in refrigeration, HVAC, industrial process cooling, and increasingly data centers. Their TKMicro line, launched in 2010, pioneered aluminum microchannel technology in their product category. Microchannel heat exchangers use smaller, flattened tubes with higher surface-area-to-volume ratios, which means better heat transfer per unit of physical space. For data center operators fighting for every square meter of outdoor mechanical area, that density advantage is real.
ThermoKey also builds systems compatible with low-GWP and natural refrigerants, which matters as F-gas regulations tighten across Europe. If you are building or retrofitting facilities in the EU, your heat rejection equipment increasingly needs to work without the high-GWP refrigerants that older systems relied on.
Prior to this deal, ThermoKey was backed by IGI Private Equity, with co-investors LGT Capital Partners and BNP Paribas BNL Equity Investments. CEO Giuseppe Visentini held a 5% stake and led the company through a growth plan targeting EUR 100 million in revenue. The Vertiv acquisition likely represents an exit for IGI and an acceleration of that growth trajectory under a much larger parent.
This acquisition does not exist in isolation. Vertiv closed its $1 billion acquisition of CoolTera (formerly PurgeRite) in 2025, adding fluid management services for coolant loops. Now ThermoKey adds the hardware at the rejection end. Together, the two deals give Vertiv coverage from the cold plate to the dry cooler, plus the services layer that keeps the loop clean and running.
The financial context reinforces the urgency. Vertiv posted $10.23 billion in 2025 revenue, up 28% year-over-year. Q4 2025 orders were up 252%. Backlog reached approximately $15 billion. The company is guiding for 27-29% organic growth in 2026. That backlog is not abstract demand. It is customers placing orders for thermal infrastructure they need built and installed on aggressive timelines. Having in-house heat rejection manufacturing, especially in EMEA where ThermoKey's production base sits, shortens the supply chain for exactly the components that are hardest to source right now.
The global data center cooling market was valued at $10.8 billion in 2025 and is projected to hit $25 billion by 2031. The liquid cooling segment alone is expected to reach $10 billion by 2026. Dell'Oro Group has Vertiv and Schneider Electric virtually tied for global cooling market share. But Vertiv is making a specific bet that owning the rejection layer, not just the cooling layer, creates a defensible position that competitors will struggle to replicate through partnerships alone.
If you manufacture cooling components, this deal should sharpen your focus. Vertiv is pulling heat rejection in-house. That means one of the largest thermal infrastructure buyers in the world just became a direct competitor for dry cooler and heat exchanger capacity. OEMs that previously sold rejection equipment to Vertiv or to the same customer base will find the competitive dynamics shifting.
For data center operators, the signal is different but equally clear. Heat rejection is no longer background infrastructure you spec once and forget. It is a capacity constraint that determines whether your next liquid cooling deployment actually works at full load. The operators who figure out rejection planning early will be the ones who can actually deliver on the AI infrastructure commitments they are making today.
Vertiv is betting that the thermal chain is a single system, not a collection of parts. ThermoKey is how they close the loop.