The Liquid Cooling Market Going Into 2026

Liquid cooling stopped being a science project the moment accelerator racks crossed 40 kW. Air can be pushed to roughly that point with containment and rear-door assistance; past it, the physics stop cooperating and the conversation moves from airflow management to fluid engineering. That threshold is why the sector's numbers look the way they do.

Sizing estimates vary by analyst, which is worth knowing before you put one in a pitch deck. Grand View Research puts the global data centre liquid cooling market at $6.65 billion in 2025, rising to about $8.17 billion in 2026. Mordor Intelligence models $6.77 billion in 2026 growing to $18.79 billion by 2031 at a 22.65% compound rate. Dell'Oro Group is the conservative voice, forecasting the market to approach $7 billion by 2029, while Precedence Research projects $25.80 billion by 2035. A credible plan cites at least two of these and explains why it chose the base case it did. Lenders notice when a founder picks the biggest number on the page without justifying it.

The more useful figure for a new entrant is penetration rather than dollars. Roughly 10% of data centres use liquid cooling today, and adoption is expected to roughly triple to about 30% by 2028, according to analysis published by DatacenterDynamics. That is the installed base that has to be retrofitted, commissioned, monitored and serviced by somebody. Ninety percent of the world's data halls are still air-cooled and most of them will not be demolished.

The revenue mix is about to invert

The single most strategically important number in this niche is buried in that same analysis: today's market revenue splits roughly 80% solutions and 20% services, and it is projected to flip to 20% solutions and 80% services within four to five years. Most operators stop at market size. The number that actually determines whether your company is worth anything in year five is what share of your revenue is contracted, recurring service work rather than one-time hardware resale.

Component-level growth confirms it. The coolant distribution unit market was valued at about $1 billion in 2024 and is projected to reach $3.6 billion by 2031, a 20.5% compound rate (Persistence Market Research). Every one of those units needs commissioning, filtration management, fluid chemistry monitoring and an annual service visit. That is the annuity.

Who already holds the ground

This is not an empty field. Mordor's share analysis puts Vertiv above 11.3% in 2025, with the top five vendors — Schneider Electric, Vertiv, Rittal, STULZ and Boyd Corporation — holding around 35% between them. Below that sit CoolIT Systems, Motivair, nVent Schroff, Delta Electronics, Airedale and Envicool on the direct-to-chip side, and LiquidStack, Submer, Green Revolution Cooling, Midas Green Technologies and DCX in immersion. Chilldyne and JetCool occupy specialist niches in negative-pressure and microconvective cold-plate designs.

A 35% top-five share means 65% of the market is fragmented. That fragmentation is the opening, and it sits almost entirely in the integration, retrofit and service layer rather than in manufacturing. Very few new companies should try to out-engineer Vertiv on CDU hardware. A great many can win the work of getting somebody else's CDU installed correctly in a 2003-vintage colocation hall with a chilled-water loop that was never designed for it.

The UK and European picture

UK data centres consume around 2.5% of national electricity, and that draw is forecast to rise roughly four-fold by 2030, according to the House of Commons Library. The pipeline of schemes without planning consent that aim to be operational by 2030 stands at about 6.2 GW. For an integrator, that is a demand curve you can put a number against: every gigawatt of new-build capacity at AI density is a multi-hundred-million-pound cooling opportunity, and the UK does not currently have the specialist commissioning workforce to absorb it.

In continental Europe the driver is regulatory rather than purely thermal, and it is covered in detail in the compliance section below. Germany's national rules are the sharpest: new facilities commissioned from July 2026 must reach a PUE of 1.2 within two years, which is effectively unreachable in a dense hall without liquid.

Six Questions Buyers Ask First

These are the questions that come up in every first meeting with a facilities director. Your business plan's market and operations sections should answer all six, because your sales conversations will be nothing but these six for the first two years.

1. What does it cost per rack?

Direct-to-chip retrofit work runs approximately $50,000 to $80,000 per rack and supports densities above 60 kW. In-row cooling units, the half-measure, cost $20,000 to $35,000 and consume a rack position. Cold plates and manifolds add $5,000 to $10,000 per server, piping installation runs $50 to $100 per linear foot including insulation and leak detection, and CDUs land at $75,000 to $150,000 per 500 kW of capacity. Whole-facility numbers come in near $3M to $4M per megawatt for new build and $2M to $3M per megawatt for retrofit.

2. What's the payback?

Published ranges cluster at 18 to 36 months. Dense AI clusters running current generation accelerators frequently pay back inside twelve months; general enterprise compute typically breaks even at two to two and a half years. Immersion economics sharpen above 50 kW per rack with electricity at $0.10/kWh or higher, and payback can compress toward 1.6 years at 80+ kW per rack. Anything shorter than a year quoted without a density figure attached should be treated as marketing.

3. What PUE does it actually deliver?

Air-cooled halls typically run between 1.25 and 1.50. In-rack and direct-to-chip liquid cooling routinely delivers below 1.10, with deployments reported at 1.03 to 1.10 once computer room air handlers come out of the loop. The number to underwrite in a proposal is not the PUE headline but the 20% to 40% reduction in total facility energy, because that is what the customer's finance team can convert into a cash figure. Over ten years, a 10 MW facility is looking at $30M to $50M in combined electricity and maintenance savings depending on regional tariffs.

4. At what density does this become mandatory?

The practical break is around 30 kW per rack, where air containment starts to require heroic measures, and it becomes absolute around 50 kW. Above 100 kW per rack, immersion begins to outperform direct-to-chip on total cost of ownership. Below 20 kW, liquid is usually a solution looking for a problem, and telling a prospect so is the fastest way to be trusted with their next project.

5. Can it be retrofitted or does it need new build?

Retrofit is the larger market and the better business. Retrofit capex runs roughly $2M to $3M per megawatt against $3M to $4M for new build, and a well-scoped retrofit delivers a large fraction of new-build thermal performance at a fraction of the cost. The constraint is rarely the cooling hardware. It is floor loading, the availability of a facility water loop, riser space, and whether the hall can be worked on live. Those four questions belong in your site survey template on day one.

6. What about water use?

This is the question that has grown fastest. Under the EU framework, water usage effectiveness is now a reported metric, and targets under discussion cap WUE at 0.4 L/kWh for new facilities in water-stressed areas. Closed-loop direct-to-chip with dry coolers uses essentially no evaporative water, which is a commercial advantage worth pricing rather than giving away.

What It Costs to Stand the Business Up

Be precise in your plan about which business you are in, because the capital requirement varies by nearly an order of magnitude across the three viable entry models. A design and integration firm that specifies, installs and commissions other vendors' equipment is the lightest. A service and fluid-management firm that contracts for filtration, chemistry monitoring and annual maintenance is lighter still on capex but slower to revenue. A product company building its own cold plates, manifolds or CDUs is a different animal entirely and belongs in a venture conversation, not an SBA one.

For an integration and service business, budget $180,000 to $900,000 in the US or £140,000 to £700,000 in the UK to reach first revenue with a credible team. The spread is driven almost entirely by two decisions: how many engineers you hire before your first contract, and whether you hold demo and inventory stock.

Where the money goes

• Thermal and mechanical engineers (2 FTE, year one): $190,000–$300,000 (£110K–£180K). This is the dominant line. Data centre thermal engineers are scarce and priced accordingly.
• Commissioning toolkit: $45,000–$120,000 (£35K–£95K) for flow benches, differential pressure rigs, particle counters, leak detection and calibrated torque equipment. Without it you cannot sign off a loop.
• Demo or loaner CDU (100–500 kW class): $12,000–$35,000 (£10K–£28K). Enormously effective in enterprise sales where the buyer has never seen one running.
• Reference rack build: $5,000–$10,000 per server (£4K–£8K) for cold plates, manifolds and blind-mate quick disconnects.
• Fluid inventory and lab testing: $25,000–$90,000 (£20K–£70K). Water-glycol at PG25 is cheap; single-phase dielectric fluids are not.
• Insurance: $18,000–$45,000/yr (£14K–£35K). Professional indemnity, contractor's all-risk and product liability. A leak over live compute is the claim that ends underinsured firms.
• Certification, bonding and licensing: $3,000–$25,000 (£2.5K–£9K), depending on how many states or how many engineers you certify.
• Working capital: $80,000–$350,000 (£60K–£280K). Capital projects settle in 90 to 150 days. This line kills more of these businesses than competition does.

The fluid decision that changes your capital plan

One choice reshapes the whole cost model. Cold-plate coolants for direct-to-chip cost roughly $1.50 to $3.00 per litre. Single-phase immersion using synthetic oil starts at about $10 to $13 per litre. Two-phase dielectric fluids historically ran $25 to $300 per litre. If your plan includes an immersion line, fluid is a working-capital item that will dominate your balance sheet, and you must show a supplier agreement, not an intention.

Funding routes

In the US, an integration or mechanical services business classifies under NAICS 238220 (Plumbing, Heating and Air-Conditioning Contractors), where the SBA size standard is $19 million in average annual receipts over the preceding five fiscal years (SBA size standards table). That comfortably qualifies a new entrant for SBA 7(a) financing up to $5 million, and a borrower may carry up to $5M in SBA loans in each of two different NAICS codes at once, which matters if you later separate the service entity from the fluid distribution entity. Lenders in this classification are used to contractor cash-flow patterns, so present a backlog schedule and a receivables ageing assumption rather than a hockey stick.

An important structuring point: SBA 7(a) will fund working capital and equipment for a services business far more readily than it will fund inventory-heavy hardware manufacture. If your plan has both, separate them. We have seen otherwise fundable plans stall because the lender could not tell whether it was underwriting a contractor or a manufacturer.

In the UK, the Start Up Loans scheme offers up to £25,000 per director at 6% fixed with free mentoring, which is meaningful for a two-founder consultancy but nowhere near sufficient for an equipped integrator. Realistic UK routes are asset finance against the commissioning fleet, invoice finance against main-contractor receivables, and equity from infrastructure-adjacent angels. Innovate UK grant competitions periodically cover heat reuse and data centre efficiency, which suits a company with genuine engineering IP rather than a pure services model.

Elsewhere, Canada's BDC lends against equipment for mechanical contractors, and in the UAE the Khalifa Fund supports technical services ventures. Wherever you raise, the section a sophisticated lender reads first is not the market size. It is the contracted backlog and the assumed days sales outstanding. Our bespoke business plan service builds both as linked schedules so the working capital requirement falls out of the model rather than being asserted.

Three Architectures, Three Business Models

The architecture you build around is not a technology preference; it determines your capital requirement, your sales cycle, your warranty exposure and your regulatory position. Pick one as the primary line and name the second as the planned extension. Plans that claim to do all three from day one do not get funded.

Why two-phase immersion is missing from that table

Deliberately. Two-phase immersion was the highest-performance option in the category until its supply chain evaporated. In December 2022, 3M announced a global exit from all PFAS manufacturing by the end of 2025, and the last date to place a Novec order was 31 March 2025. Novec 7100, Novec 649 and Fluorinert FC-72 — the fluids the entire two-phase approach was built on — are no longer produced (DatacenterDynamics).

Regulation is closing behind it. The EPA's PFAS reporting rule under TSCA Section 8 takes effect in 2026 and expands disclosure duties on fluorinated substances, and the EU's universal PFAS restriction dossier under REACH targets roughly 10,000 compounds; if adopted as drafted it would restrict effectively all fluorinated fluids in new applications in the EU around 2027 to 2028 (Schneider Electric).

This is not a reason to avoid the sector. It is a reason to be explicit in your plan. If you sell two-phase, name the replacement fluid, name the supplier, state the contracted volume and state your regulatory position. If you do not sell it, say so and explain that the decision is a deliberate supply-chain risk position. Either answer reads as competence. Silence reads as having not looked.

Adjacent markets are worth mapping in the same section of your plan. If your model leans toward whole-facility thermal design rather than component integration, the data center cooling solutions business plan template covers a broader scope, and founders targeting greenfield AI campuses should look at the hyperscale data center business plan template.

Where the Margin Sits

Liquid cooling businesses have four revenue lines, and they carry radically different margins. Plans that model a single blended gross margin percentage are the ones that miss their numbers, because the mix shifts violently between a year with one large capital project and a year built on service renewals.

The four lines

• Hardware supply and resale — CDUs, cold plates, manifolds, dry coolers. Gross margin 12–22%. High revenue, low quality of earnings. Dry coolers alone run $800–$1,600 per kW of capacity.
• Engineering, installation and commissioning labour — Gross margin 38–55%. This is where a competent integrator makes its money in year one and two.
• Recurring service and maintenance contracts — Priced at 8–14% of installed hardware value annually. Gross margin 45–65%. This is the line that gets a valuation multiple.
• Fluid, filtration and chemistry management — Billed per litre plus scheduled sampling. Gross margin 45–60% and near-total customer retention, because nobody changes fluid vendors on a live loop casually.

Blended net margin after overhead and warranty reserve typically lands between 9% and 18%. Anyone modelling 30% net on a business that passes through hardware is modelling a manufacturer, not an integrator.

Worked example: a 2 MW retrofit

Take a 2 MW direct-to-chip retrofit in an enterprise colocation hall, contracted at $2.6 million. That splits roughly as follows:

• Hardware pass-through: $1.55M at 16% gross = $248,000
• Engineering and commissioning labour: $780,000 at 46% gross = $359,000
• Project management and controls integration: $270,000 at 40% gross = $108,000
• Total gross profit: $715,000 on $2.6M revenue = 27.5% blended gross

Now attach the service contract. At 11% of installed hardware value, that is $170,000 per year at 58% gross, producing $99,000 of recurring gross profit. With fixed overhead at $520,000 per year — two engineers, a project manager, a part-time controller, insurance, vehicles and tooling — the project year nets roughly $294,000, and the service annuity alone then covers about 19% of fixed cost before the next contract is signed.

Run that forward. Four completed 2 MW projects produce roughly $680,000 of annual recurring service gross profit, which exceeds fixed overhead. That is the moment the business stops being a contractor and starts being an asset. Model it explicitly, because it is the argument that persuades an equity investor to price you off recurring revenue rather than off project margin.

The pricing mistake to avoid

Integrators reflexively price service as a percentage discount on a maintenance schedule they copied from a chiller contract. Liquid loops are not chillers. The value you are pricing is water chemistry governance, particulate control, corrosion monitoring and leak-risk management over live compute worth tens of millions of dollars. Price against the asset you are protecting, not against your labour hours. Firms that make this shift routinely add six to eight points to service gross margin without losing renewals.

Sales cycle and cash conversion

Expect six to fourteen months from first meeting to signed contract on a retrofit above 1 MW, and three to six months on a single-row or proof-of-concept deployment. Payment terms follow construction convention, meaning 90 to 150 days with retention held against commissioning sign-off. Build the cash model on that basis. A plan showing 30-day terms in a capital-project business tells an experienced lender that the founder has not run one.

Certification, Licensing & Reporting Duty

Liquid cooling sits at an awkward intersection: it is IT infrastructure sold to facilities teams, installed by mechanical trades, and increasingly governed by energy and chemical regulation. Three separate regimes apply and your plan should address each.

United States

• EPA Section 608 technician certification — Required under the Clean Air Act for anyone who maintains, services, repairs or disposes of equipment that could release refrigerants. Exams cost $25–$150 per technician and the credential does not expire (US EPA).
• The 2026 threshold change — From 1 January 2026, the EPA lowered the applicability threshold from 50 lb to 15 lb of charge for systems using refrigerants with a global warming potential above 53. Plant that previously sat outside leak-repair and recordkeeping duties now falls inside it. Any plan touching DX-based rear-door or chiller plant should state this explicitly.
• State mechanical or HVAC contractor licence — 36 states operate a statewide licence; 14 leave it to county and municipal authorities. Licensing states typically require two exams, a business-and-law paper of 50–80 questions and a trade paper of 80–120 questions, and the experience path from zero to a full contractor licence commonly takes two to five years (licensing overview). If your founding team lacks a qualifying individual, your first hire is a licence holder, and your plan should say so.
• TSCA Section 8 PFAS reporting — Effective 2026, expanding disclosure obligations across fluorinated substances. Directly relevant to anyone importing, distributing or handling dielectric coolant.
• Contractor bonding and OSHA compliance — Standard for site work, plus arc-flash and lockout/tagout procedures when working live halls.

United Kingdom

• F-Gas qualification — Engineers working on equipment containing fluorinated gases need a recognised qualification, commonly the City & Guilds 2079 award. The five-day lifetime course runs at roughly £1,280 per engineer (GOV.UK).
• Company F-Gas registration — Individual qualification is not enough; the business itself needs certification, typically via a scheme such as REFCOM.
• Data centres as Critical National Infrastructure — Designated in September 2024, which raises security vetting expectations for anyone with physical access to a hall.
• NSIP planning route — The Infrastructure Planning (Business or Commercial Projects) (Amendment) Regulations 2026, made on 8 January 2026, allow major data centre schemes to opt into the Nationally Significant Infrastructure Projects regime, moving consent from local planning authorities to the Secretary of State (House of Commons Library). For a supplier this shortens and concentrates the pipeline: fewer, larger schemes decided centrally.
• CDM 2015 duties — You will be a contractor or principal contractor on most installs, with the documentation obligations that follow.

European Union

The EU regime is the one that creates demand rather than merely permitting supply. Under Commission Delegated Regulation (EU) 2024/1364, implementing the Energy Efficiency Directive (2023/1791), any data centre with installed IT power demand of 500 kW or more must report annually to the European Database on Data Centres, submitting 24 data points spanning energy, sustainability metrics, ICT capacity and data traffic. Four indicators are calculated from them: Power Usage Effectiveness, Water Usage Effectiveness, Energy Reuse Factor and Renewable Energy Factor (European Commission). The annual deadline for full-year 2025 data fell on 15 May 2026.

The Directive itself sets transparency rather than binding thresholds, leaving performance targets to member states. Germany went furthest: facilities commissioned from July 2026 must reach PUE 1.2 within two years, with waste-heat reuse obligations escalating from 10% Energy Reuse Factor in July 2026 to 15% in July 2027 and 20% in July 2028 (EU data centre regulation summary). A PUE of 1.2 with a 20% heat-reuse requirement is close to a legislative mandate for liquid cooling with heat capture, and it is the strongest single demand signal in the sector. If you have any European ambition, that paragraph belongs in your market section.

Six Ways These Plans Fall Over

We have reviewed enough infrastructure-services plans to see the same six failures recur. Each one is avoidable in a paragraph.

1. Pricing the hardware, ignoring the annuity

Founders model revenue as units of equipment sold because that is the number they can find. In a market where the revenue mix is forecast to invert toward 80% services, a plan built on hardware markup is describing the shrinking half. Model the service attachment rate as a named assumption and defend it.

2. A two-phase immersion line with no fluid contract

Since the Novec exit completed in 2025, any plan proposing two-phase immersion without a named replacement fluid, a named supplier and a stated position on the EU REACH restriction is unfinanceable. Technical reviewers on the lender's side now ask this question first.

3. Forgetting the enabling works

In many retrofits the enabling work — pumps, dry coolers, riser piping, structural load assessment, leak containment — costs as much as the cooling hardware itself. Quoting the hardware and discovering the rest in week three is how integrators turn a 27% gross project into a break-even one. Price a paid site survey as its own deliverable.

4. Underwriting hyperscale margins on an enterprise pipeline

Hyperscale buyers negotiate like hyperscale buyers and increasingly design in-house. The winnable pipeline for a new firm is enterprise and mid-tier colocation retrofit, where the buyer has a thermal problem and no internal thermal team. That segment pays for expertise. Build the forecast on it and treat hyperscale as upside.

5. No fluid governance in the service contract

Galvanic corrosion between dissimilar metals, biofilm growth in a poorly inhibited loop and particulate fouling in cold-plate microchannels are the three failure modes that generate warranty claims. If your contract does not specify water chemistry limits, filtration intervals and sampling responsibility, you have accepted those risks for free. Named material compatibility and chemistry schedules belong in the operations section of the plan.

6. Modelling contractor cash flow like a SaaS business

Ninety to 150 day settlement with retention held to commissioning sign-off is the norm. A plan showing 30-day terms understates the working capital requirement by a factor that will surface in month four, usually at exactly the moment a second project starts. Build a monthly cash model, not an annual one.

Sample Business Plan Preview

Here's an extract from a liquid cooling business plan written by our team, so you can see the level of specificity lenders and investors expect:

What's in the Template

Every Avvale business plan template includes these sections, pre-structured for your industry:

• Executive Summary — Your business at a glance, written to hook investors in 60 seconds
• Company Overview — Legal structure, ownership, location, and founding story
• Industry Analysis — Market size, growth trends, and the regulatory picture
• Customer Analysis — Target segments, procurement behaviour, and buying triggers
• Competitor Analysis — Vendor mapping and your differentiation strategy
• Marketing Plan — Channels, messaging, and customer acquisition strategy
• Operations Plan — Delivery workflow, staffing structure, and key milestones
• Management Team — Founder bios, advisory board, and key hires planned

For a liquid cooling business specifically, we recommend expanding the operations section to cover your site survey methodology, commissioning protocol, water chemistry and filtration governance schedule, and the material compatibility matrix for the fluids you intend to handle. Those four items are what separate a services company from a group of engineers with a van, and technical reviewers on the lender's side look for them.

The optional Financial Forecast add-on (included in our $300/£250 and $1,000/£800 packages) provides a 5-year Excel model with income statement, cash flow, balance sheet, break-even analysis, and startup capital requirements. For capital-project businesses we build the cash flow monthly with an explicit backlog schedule and retention assumption, because annual models hide exactly the problem that sinks contractors. You can see the depth of the underlying research in our market research and content service, or start from the free business plan template hub.

Frequently Asked Questions