The pace of adoption encodes a directional shift in density expectations that energy planning cycles have not universally absorbed
Decision Focus
A July 2026 market analysis from Straits Research projects the global data center immersion cooling market growing from approximately $290 million in 2025 to $2.45 billion by 2034, at a compound annual growth rate of 26.76%. North America holds roughly 40% of current market share, and cloud service providers account for nearly 46% of demand. The operational signal buried in these figures is not about a new cooling vendor—it is about the power density assumptions baked into infrastructure that your procurement, interconnection, and capacity teams are planning against right now.
90-Second Brief
In recent days, the Straits Research report sizes the global immersion cooling market at $290 million in 2025 and projects a near-ninefold expansion to $2.45 billion by 2034. Hyperscale data centers represent the largest segment by facility type, at 44.58% share. Cloud providers, your direct counterparts in energy procurement, are the dominant end users. The pace of adoption encodes a directional shift in density expectations that energy planning cycles have not universally absorbed.
What Is Really Happening?
The proximate driver is GPU-class compute. AI and HPC workloads have pushed rack-level power densities into ranges that conventional air cooling—and even direct liquid cooling—cannot address economically at scale. Immersion, whether single-phase or two-phase, allows operators to pack substantially more compute into a fixed physical footprint, which is precisely what hyperscalers are incentivized to do as land, interconnection slots, and grid capacity become constrained.
Single-phase systems already command 63% of the market, suggesting a preference for lower operational complexity before two-phase adoption matures. Recent vendor activity reinforces the commercialization momentum: in March 2026, UNICOM Engineering and GRC launched integrated immersion-cooled AI infrastructure combining immersion-ready compute platforms with single-phase systems, targeting HPC, GPUaaS, and edge deployments through a single-contract procurement model. Separately, fluid supplier Perstorp initiated OEM trials of new dielectric fluids for AI and hyperscale applications in February 2026, signaling that the supply chain upstream of the cooling hardware is also being built out.
Together, these developments point toward a market moving from pilot deployments to repeatable at-scale procurement—the moment when infrastructure specifications become standards rather than exceptions.
Why It Matters for Global Heads of Data Center Energy
The energy consequence of immersion adoption is not PUE improvement in isolation—it is a reconfiguration of the power delivery problem. When rack density scales from the conventional 5–15 kW range toward the 50–100 kW range that immersion enables, every upstream infrastructure element requires re-examination: substation feed capacity, transformer sizing, busway and switchgear ratings, UPS topology, and backup generation dispatch logic.
If your capital planning cycle is still modeled on air-cooled density assumptions, you are likely underspecifying infrastructure for the loads your hyperscale or cloud-provider tenants will deliver within the current decade. Hyperscale facilities already represent 44.58% of immersion cooling demand—a segment you either serve or lose to operators who anticipated the shift earlier.
There is also a grid interconnection implication. Higher density per square foot means a smaller facility can draw the same or greater aggregate power. That changes the economics of interconnection requests—fewer physical sites for equivalent MW load—but it does not reduce queue timelines or transformer lead times. Energy teams need to model the density trajectory explicitly in interconnection applications, not just in facility design, to avoid under-requesting capacity that dense AI deployments will require within the contract horizon of a 10–15 year PPA.
Forward View
Three fronts carry the most strategic weight if the immersion adoption curve continues at or near the projected rate.
First, behind-the-meter power infrastructure will require active redesign. Immersion-ready deployments demand power delivery systems engineered for sustained high-density draw, not peak-event headroom. The transformer and switchgear procurement implications arrive 2–3 years before the density load does, which means the planning window is now.
Second, cooling system energy consumption itself becomes a more complex optimization variable. Immersion cooling reduces cooling energy as a share of total facility power—potentially improving PUE materially—but the absolute power draw per rack still rises sharply. Scope 2 accounting and 24/7 carbon-free energy matching programs will need to track this shift: lower PUE does not automatically mean lower carbon intensity when the compute load per MW doubles.
Third, fluid supply chain dependencies are emerging as a new vendor relationship to manage. Perstorp’s dielectric fluid trial activity in early 2026 indicates that fluid chemistry and supply reliability will become operational concerns similar to transformer supply chains. Energy and infrastructure teams without visibility into fluid sourcing may face deployment delays that are invisible to procurement today.
What Is Still Uncertain
The Straits Research projection carries the limitations common to market sizing analyses: the $290 million 2025 baseline and the $2.45 billion 2034 forecast are derived from vendor surveys, company disclosures, and analyst modeling—not from audited deployment data. The 26.76% CAGR extrapolates a specific adoption trajectory that could slow if hyperscaler capital allocation shifts, if GPU architectures alter thermal profiles, or if competing approaches such as rear-door heat exchangers or direct liquid cooling chip packages capture a larger share of the density problem.
The geographic split also warrants scrutiny. The report identifies North America at approximately 40.6% share—consistent with hyperscale concentration in US markets—but does not break down where within North America the density-driven demand is concentrated. Northern Virginia, Phoenix, and Silicon Valley each carry different grid constraint profiles, and PUE improvement in one market does not translate to interconnection relief in another.
One Question for Your Team
Does your current interconnection queue strategy and capital infrastructure specification account for the density loads that immersion-cooled AI deployments will require from your largest tenants or internal build programs within the next five years—or are your power delivery assumptions still anchored to air-cooled baseline densities?
Sources
- Straitsresearch — Data Center Immersion Cooling Market Size, Share, 2034 (Link)
