NextFin News - Danfoss's data-center business almost doubled in 2025, reaching about 7% of the Danish engineer's EUR 9.4 billion in total sales, as the artificial-intelligence buildout turns cooling from a back-office utility into one of the fastest-growing revenue streams in industrial equipment. The question now is whether the AI capex wave is a cyclical surge that will fade, or a structural shift that rewires the company's long-term growth profile.
The answer matters because Danfoss is not a listed company chasing quarterly expectations. It is owned by the Clausen family and its foundation, which control 99.89% of the votes, and that ownership structure lets it place a multi-year bet on data-center cooling without the pressure of public-market earnings guidance. In 2025, that bet started to pay off: North America's Climate Solutions segment grew 26% organically, driven by data-center demand, while the company raised R&D spending to 5.3% of sales (EUR 503 million) and generated a record EUR 734 million in cash flow to fund further expansion.
But the AI spending splurge is a double-edged sword. Hyperscaler capital expenditure is climbing at a pace that few industrial suppliers anticipated, and the cooling market is expanding with it. Analysts project the top five hyperscalers will spend roughly $600 billion in 2026, up about 36% from 2025. The global data-center cooling market is projected to grow from USD 18.8 billion in 2025 to USD 21.0 billion in 2026 and USD 54.4 billion by 2034, a compound annual growth rate of 12.6%. The liquid-cooling slice is growing even faster, from USD 4.8 billion in 2025 to USD 6 billion in 2026 and USD 27.1 billion by 2035, an 18.2% compound rate. Danfoss's challenge is to convert that market growth into durable share before better-capitalized rivals lock it out.
The Mechanism: Why AI Makes Cooling a Growth Engine, Not a Commodity
Cooling used to be a cost line on a data-center operator's spreadsheet, sized conservatively and replaced infrequently. AI changes the physics of the problem. Traditional air-cooling systems max out at roughly 25 kilowatts per rack, while AI training clusters routinely demand 30 to 40 kilowatts per rack and next-generation liquid-cooled designs handle up to 100 kilowatts. That density jump does two things at once: it forces operators to retrofit or rebuild, and it shifts the value pool toward the thermal-management layer because a cooling failure at those densities means millions in destroyed compute.
Danfoss's entry point is the chiller and compressor, not the full rack-level system. Its Turbocor TGS380 compressor, deployed through partners such as ThermalWorks, is oil-free and uses magnetic bearings, which cuts mechanical friction and maintenance. The system supports a temperature delta of up to 60 degrees Fahrenheit between supply and return fluid, three times higher than traditional chillers, allowing the same circuit to serve hybrid, all-liquid, and all-air data halls. That flexibility matters because most operators are not ripping out air cooling overnight; they are blending the two during a multi-year transition.
The economics show up in power usage effectiveness, the industry's scorecard for efficiency. Edged, a data-center operator that deployed ThermalWorks chillers with Danfoss Turbocor compressors across multiple facilities by the end of 2024, reported an average PUE of 1.15 and energy savings of 20% to 30% versus traditional chillers. The system can also operate in ambient temperatures as high as 135 degrees Fahrenheit, which widens the geography where free-cooling strategies work.
"The oil-free and magnetic-bearing technology offered by Danfoss Turbocor compressors provide exceptional energy efficiency and reliability, making them ideal for mission-critical environments like data centers," said Rogerio Federici, vice president of Danfoss Turbocor.
The transmission channel from AI spending to Danfoss's revenue is therefore not linear capex pass-through. It runs through rack density, which forces a technology substitution from air to liquid and hybrid systems, which raises the value per facility and the stickiness of the thermal supplier. A chiller selected during a hyperscale campus build is rarely swapped out; the switching cost is a full system redesign. That is why Danfoss's data-center sales almost doubling to 7% of group revenue is more significant than the percentage itself: it signals the company is winning design-ins during the installation wave that will define the next decade of capacity.
Cyclical Surge or Structural Shift? The Ownership Test
The central analytical question is whether this is a cyclical upswing in capital spending or a structural regime change. The evidence points to both operating at different time horizons, and separating them is the only honest reading.
On the cyclical side, hyperscaler capital expenditure is a notoriously lumpy cycle. Analysts expect top-five hyperscaler spending of roughly $600 billion in 2026, up about a third from 2025, but the pace of revisions has been steep and the direction can reverse within a single fiscal year if AI monetization disappoints. Danfoss itself flagged the uncertainty implicitly: its 2026 sales outlook is a wide EUR 9.1 billion to EUR 10.6 billion, a range of EUR 1.5 billion that reflects genuine uncertainty about how much of the AI order flow converts into recognized revenue this year. The company's total sales fell slightly in 2025 on a reported basis because a weakening U.S. dollar subtracted 3% from the top line, and Europe ended the year with negative 2% growth. Agriculture, a core end-market for the Power Solutions segment, remains in a downturn. None of that is AI-related, and all of it can mask or distort the cooling signal.
On the structural side, three forces will not self-correct. First, rack density is a one-way door: chipmakers are not going back to lower-power designs, so the thermal problem compounds with each generation. Second, energy and water constraints are becoming siting constraints. A data center's access to power now determines where it can be built, and cooling efficiency directly governs how much compute fits inside a fixed power envelope. Third, PUE is moving from a sustainability metric to a competitive one, because operators with better efficiency can run more workloads inside the same contracted capacity. Danfoss's 20-30% energy-savings claim, if it holds across deployments, is a structural value proposition, not a cyclical discount.
The ownership structure sharpens the structural call. Because the Clausen family and its foundation hold 99.89% of the votes, Danfoss does not face the quarterly-earnings discipline that would push a listed peer to prioritize near-term margin over market-share capture. President and CEO Kim Fausing framed the strategy plainly in the company's 2025 results: "With a strong balance sheet and record-high cash flow, we are well positioned to continue investing in our business, driving organic growth and M&A." That sentence is a statement of optionality. The company spent EUR 342 million on capital expenditure in 2025, expanded factories across the Americas, Europe, and Asia, and grew R&D to 5.3% of sales. A private owner can afford to let a cooling bet compound for five years before it shows up as a segment-level margin driver.
"The name of the game is to be as energy-efficient as possible for our customers," said Julian Bastiani, Edged's vice president of operations in North America, after a 20-week rollout of the cooling technology across multiple facilities.
So the cyclical-versus-structural verdict is: the order flow is cyclical in its timing, but the technology substitution underneath it is structural. Danfoss's bet is that it can use the cyclical wave to lock in structural design-ins. If it succeeds, the 7% revenue share today understates the long-term earnings contribution because cooling content per facility rises with density. If it fails, the company is left with a larger cost base and a capex cycle that reverted to the mean.
The Second-Order Effect: Power, Water, and the PUE Arms Race
The first-order effect of AI spending is obvious: more data centers, more cooling orders. The second-order effect is less discussed and more consequential. As cooling becomes the binding constraint on compute capacity, the value in the data-center stack migrates toward the companies that can move heat and power most efficiently. That migration redistributes pricing power away from generic components and toward integrated thermal and power suppliers.
Vertiv, the closest listed benchmark, illustrates the magnitude. It reported 2025 net sales of $10.2 billion, up 28% from a year earlier, with backlog more than doubling to $15.0 billion at year-end. Schneider Electric, Daikin, Johnson Controls, STULZ, Rittal, Asetek, and CoolIT Systems are all competing for the same design-ins. The competitive field is crowded, but the market is large enough that share gains do not require share losses: the total cooling market is expanding by roughly USD 2 billion a year in the near term, and the liquid-cooling segment by even more.
The third-order implication is an expectation gap. The market has priced AI infrastructure as a growth trade, which means any deceleration in hyperscaler orders is punished disproportionately. Danfoss's wide 2026 revenue guidance is itself a signal that management refuses to commit to a linear extrapolation. Investors in listed peers have learned that a book-to-bill ratio above one can reverse within two quarters if chip supply or power availability bottlenecks construction. For a private company, the penalty for being wrong is lower, but the opportunity cost of over-investing ahead of demand is real: the EUR 342 million in CapEx and the R&D ramp are commitments made before the revenue is certain.
There is also a water dimension that the market underweights. Liquid cooling reduces energy use but shifts the constraint toward water availability in some designs, and waterless or closed-loop approaches are becoming a differentiator in drought-prone regions where hyperscalers are building. Danfoss's waterless chiller approach via ThermalWorks is a direct response to that constraint. It is a small detail in a results release, but it is the kind of product decision that determines which suppliers win in the American Southwest and Southern Europe, where power is available but water permits are not.
The Counter-Thesis: Why the Cooling Bet Could Underperform
The strongest case against Danfoss's cooling thesis is not that AI spending stops. It is that the value accrues elsewhere in the stack, and that air-cooling improvements close the gap faster than liquid advocates expect.
Air-cooling technology keeps improving. Hot and cold aisle containment, higher-efficiency fans, and smarter controls have repeatedly extended the economic life of air-cooled designs, and some operators have pushed traditional systems beyond the 25-kilowatt-per-rack threshold that liquid proponents treat as a hard ceiling. If air cooling continues to creep upward in capability, the addressable market for liquid and hybrid systems grows more slowly than the density curve implies.
Second, Danfoss is not the primary contractor. It supplies compressors and components into systems built by chiller OEMs and integrated by larger players such as Vertiv and Schneider Electric. In a consolidating market, the integrators capture the margin and the customer relationship, while component suppliers face price pressure. Vertiv's $10.2 billion revenue base and its deep partnerships with chipmakers give it an advantage in winning full-hall contracts where Danfoss appears only as a sub-component. The risk is that Danfoss wins the technology argument but loses the value capture.
Third, the capex cycle could turn sooner than the structural story assumes. If hyperscaler spending growth decelerates from its current pace, or if AI revenue per watt disappoints, operators will extend equipment refresh cycles rather than build new halls. Danfoss's 2025 data-center sales almost doubled off a small base; sustaining that growth rate requires the absolute dollar additions to keep compounding, which is mathematically harder each year.
The falsifying signal is specific: if hyperscaler capital expenditure growth falls below 10% year over year for two consecutive quarters while Danfoss's data-center revenue growth drops below 20%, the structural-substitution thesis is wrong and this is a cyclical wave peaking. A secondary signal would be liquid-cooling adoption failing to exceed 40% of new high-density builds within three years, which would indicate air-cooling improvements are containing the technology shift.
What Comes Next: Beneficiaries, Exposure, and the Watch List
Translating the mechanism into impact, the near-term beneficiaries are the thermal and power suppliers with design-ins already locked at hyperscale campuses. Danfoss is one of them, but so are Vertiv, Schneider Electric, and the specialized liquid-cooling firms. The exposed parties are generic component suppliers without a thermal narrative and the traditional HVAC players that cannot migrate up the density curve. For investors, the asymmetry is that a private company can absorb a cyclical downturn better than a listed peer, but it also discloses less, so the market learns about problems later.
Split by time horizon: in the short term, sentiment tracks hyperscaler capex headlines and any sign of order deceleration. In the medium term, fundamentals depend on conversion of design-ins into recognized revenue, which is where the wide EUR 9.1 billion to EUR 10.6 billion 2026 guidance range will tighten or widen. In the long term, the structural question is whether liquid and hybrid cooling becomes the default for high-density builds, which would make Danfoss's compressor and chiller portfolio a durable annuity rather than a cyclical order book.
Three scenarios frame the path. The base case is steady hyperscaler spending with Danfoss's data-center sales holding in the mid-single-digit percentage of group revenue and growing faster than the group. The upside case is a continued capex splurge that pushes data-center revenue toward 10% of sales and forces competitors to license or partner rather than displace. The downside case is a 2027 capex pause that leaves Danfoss with excess capacity from its EUR 342 million CapEx program and forces margin defense.
The watch list is concrete. First, Danfoss's next results update and whether data-center sales remain above 7% of the total. Second, hyperscaler capex guidance from the largest operators, specifically whether full-year 2027 commitments hold above the current consensus. Third, liquid-cooling adoption rates in new high-density builds, which determine whether the technology shift is real or rhetorical. Fourth, PUE benchmarks from new facilities, because efficiency claims are only valuable if they survive deployment.
Danfoss's data-center bet is not a bet on AI demand in the abstract. It is a bet that heat density is the bottleneck AI cannot engineer around, and that the company which solves it first captures value regardless of which chipmaker or cloud provider wins. The AI spending splurge is the fuel; the cooling constraint is the engine. If the bottleneck holds, Danfoss's 7% revenue share is the beginning, not the ceiling.
Explore more exclusive insights at nextfin.ai.
