Liquid cooling startups: sell the ugly AI plumbing before everyone panics
Liquid cooling startups can turn AI heat, power density and heat reuse into paid audits, buyer reports and retrofit tools. Start with one wedge.
Data center heat should not be wasted while Europe worries about energy prices.
That sounds obvious.
Still, too many AI founders talk about models, GPUs and funding while pretending the heat disappears into the sky like startup confetti.
It does not.
The heat is real. The power bill is real. The cooling loop is real. The local politics are real. The buyer budget is real.
TL;DR: Liquid cooling startups help data centers remove heat from dense AI chips through direct-to-chip cooling, immersion cooling, rear-door heat exchangers, coolant distribution units, monitoring, leak detection, retrofit planning and heat reuse. For bootstrapped founders, the first sale is rarely a full cooling system. The sharper entry is a paid audit, heat-flow report, site readiness memo, buyer calculator, leak-risk review, district heat feasibility note or cooling procurement helper that turns AI heat into a business case.
I am Violetta Bonenkamp, founder of Mean CEO, CADChain, and F/MS Startup Game. I care about this topic because deep tech has a habit of hiding the important work behind unglamorous parts.
Cooling is one of those parts.
If you already understand data center energy demand created by AI inference growth, this is the heat chapter. AI inference does not live in a clean little cloud. It runs inside facilities that need electricity, racks, cooling, pipes, pumps, heat exchangers, water choices, chips, technicians and buyers who can afford the whole thing.
What Liquid Cooling Means In Data Centers
Liquid cooling means using liquid to move heat away from servers, chips, racks or data halls instead of depending only on air.
In data centers, the main methods include:
- Direct-to-chip cooling: liquid moves through cold plates attached to hot processors such as GPUs and CPUs.
- Immersion cooling: servers or components sit in a special fluid that absorbs heat.
- Rear-door heat exchangers: a liquid-cooled door captures hot exhaust air from racks.
- Coolant distribution units: equipment that manages coolant flow between IT equipment and the facility water loop.
- Hybrid cooling: air and liquid work together because many sites will not change everything at once.
The ASHRAE liquid cooling white paper explains why liquid cooling moved from high-performance computing into mainstream data centers, while the Open Compute Project advanced cooling group works on public guidance for liquid-cooled IT and facility water systems.
That matters for founders because standards and shared designs create a market around parts, audits, documentation, retrofits, training, monitoring and procurement support.
You do not need to own the whole data center to sell into the cooling problem.
You need to own one painful decision.
Why AI Made Cooling A Buyer Problem
AI has changed rack density.
Older data halls could lean heavily on air cooling because the heat per rack was easier to handle. AI servers can pack far more power into a smaller space. More power becomes more heat, and that heat has to move somewhere.
The IEA Energy and AI report says data centres used about 415 TWh of electricity in 2024 and projects demand could reach 945 TWh by 2030. The same report notes that cooling can range from a small share in tuned hyperscale sites to more than 30% in less lean enterprise sites.
For a founder, that means cooling is no longer a facilities footnote.
Cooling affects:
- Whether a site can host dense AI racks.
- Whether the operator can add more compute.
- Whether cloud and colocation prices rise.
- Whether heat can be reused.
- Whether water use becomes a local concern.
- Whether the buyer can prove energy claims.
- Whether the site gets permits, partners or local trust.
This is why Europe’s AI infrastructure gap is not only about GPU access. The gap includes power, cooling, data center space, site timing, supplier control and the operational evidence buyers need before they trust an AI workload.
Here is the blunt founder read:
If AI racks become hotter and denser, someone has to sell the tools that help buyers decide what to cool, where to cool it, how to monitor it, how to finance it and what to do with the heat.
That someone does not have to be a giant.
Heat Reuse Is The Part Founders Should Watch
Data centers turn most of the electricity they consume into heat.
Cooling usually treats that heat as a problem to reject.
Europe should treat more of it as a product input.
The technical reality is messy. Heat temperature, distance to heat demand, district heating access, contracts, heat pumps, seasonality, local rules, site layout and buyer appetite all matter. A data center far from heat demand may have no easy reuse path. A site near a campus, housing block, greenhouse, swimming pool, hospital or district heat network may look very different.
That is a founder opening.
The Tallaght District Heating Scheme in South Dublin shows the practical version: waste heat from a nearby data center helps warm local public, residential and commercial buildings. Stockholm Exergi’s heat recovery model shows another path, with excess heat from data centers and other businesses fed into district heating. Microsoft has also described a Danish project where surplus datacenter heat will warm homes near Høje-Taastrup.
The lesson for startups is not "build a district heating network."
The lesson is narrower:
Sell the proof layer around heat reuse.
Founders can help with:
- Site heat mapping.
- Buyer heat-demand research.
- Contract templates for heat offtake.
- Heat pump sizing notes.
- Permit and reporting packets.
- Temperature and flow monitoring.
- Heat reuse payback estimates.
- District heat partner search.
- Investor and lender evidence packs.
- Public-sector explanation memos.
The product can start as service work.
That is not embarrassing.
It is how a bootstrapped founder learns what buyers actually pay for.
The Regulation Angle Is Getting Less Optional
The EU is making data center energy and water data more visible.
Delegated Regulation (EU) 2024/1364 sets out a common Union scheme for data center reporting. The regulation covers indicators connected with energy consumption, power use, temperature set points, waste heat use, water use and renewable energy.
Do not read that only as a legal burden.
Read it as market signal.
When buyers must report more data, they need:
- Cleaner measurement.
- Better records.
- Easier internal handoffs.
- Credible supplier evidence.
- Benchmark-ready reports.
- Less chaos around energy and water data.
That creates startup openings around data capture, reporting prep, heat reuse documentation, cooling asset inventories and evidence packs.
This is very close to the logic behind infrastructure startups when energy and compute get expensive. Infrastructure buyers often pay first for clarity, not for a huge new system. They want someone to reduce uncertainty before money moves.
Liquid Cooling Startup Wedge Table
Use this table before you build hardware, raise money or pitch "green data center" magic.
Colocation operator or enterprise data center
One report showing power, cooling and rack constraints
City, campus, energy utility or data center owner
One site model with heat source, heat demand and distance
Operator planning direct-to-chip cooling
Pilot sensors on one loop with alert records
AI cloud team buying capacity
Vendor comparison with risk, timing and cost notes
Facility team with liquid loops
Monthly sample report and maintenance advice
Engineering firm or operator
Before-and-after site plan for one data hall
District heat network and data center owner
One signed letter of intent or paid search project
Lender, insurer, buyer or public partner
Diligence-ready folder with site data and assumptions
The pattern is simple.
The first paid product should reduce a decision risk.
That is the part a small team can sell before pumps, pipes and hardware stock eat the company.
Where Bootstrappers Can Enter
Bootstrapped founders should not start by trying to outbuild Vertiv, Schneider Electric, Siemens, Microsoft, NVIDIA, Google or a major colocation provider.
That is ego with an invoice attached.
Start where the buyer has confusion, urgency and a narrow budget.
Good entry points include:
- A diagnostic for companies planning dense AI racks.
- A heat reuse calculator for local energy partners.
- A supplier risk memo for direct-to-chip cooling parts.
- A leak and maintenance checklist for early liquid loops.
- A district heat partner search for new data center sites.
- A reporting prep service for EU data center indicators.
- A water and coolant monitoring dashboard for one facility team.
- A training workshop for non-technical data center buyers.
- A due diligence memo for lenders financing AI data center upgrades.
- A procurement pack for startups buying AI cloud capacity from providers.
This is also where the F/MS lean validation framework fits. Test the buyer’s willingness to pay before building a platform. The F/MS Startup Game landing page test guide is useful here too, because a founder can test one offer around liquid cooling audits, heat reuse notes or facility reporting before writing code.
Deep tech founders sometimes dislike that advice because it feels too small.
Good.
Small paid proof is how you stay alive long enough to earn the right to build bigger things.
The CADChain Lens: Engineering Data Will Matter
Liquid cooling is not just a pipe story.
It is also an engineering data story.
Cooling projects involve drawings, facility layouts, supplier files, inspection records, component specs, service logs, water chemistry reports, warranties and handover documents. If a founder wants to sell into this market, she needs respect for engineering data and access control.
That is why CADChain is relevant to my lens here. CADChain works around CAD data, IP, technical evidence, blockchain, machine learning and engineering workflows. Data center cooling projects will depend on similarly boring but powerful records.
The startups I would watch are not the ones with the loudest climate claim.
I would watch the ones that can answer:
- Who owns the site drawings?
- Which supplier touched which design?
- Which component changed after the audit?
- Which water chemistry record proves maintenance happened?
- Which heat reuse assumption changed?
- Which buyer signed off on risk?
- Which files can be shown to lenders, insurers or public partners?
If you cannot trace the evidence, you cannot sell trust.
What Founders Should Build First
Start with a service-shaped product.
I know. Everyone wants software margins.
But in hard infrastructure markets, service work is often the cheapest research you will ever do.
Build one of these first:
- A liquid cooling readiness report for one operator, one facility or one AI team buying capacity.
- A heat reuse feasibility memo for one site near a campus, district heat network, housing plan or public building.
- A vendor comparison packet for buyers choosing between direct-to-chip, immersion, rear-door heat exchangers or hybrid setups.
- A cooling risk evidence folder for insurers, lenders, public partners or board approval.
- A sensor and reporting pilot on one cooling loop, one water chemistry process or one heat offtake path.
Price the first version so the buyer feels the pain but can still say yes without a six-month committee dance.
Then collect:
- What data the buyer had.
- What data was missing.
- Who delayed the sale.
- Which report section mattered.
- Which supplier created confusion.
- Which number changed the buyer’s decision.
- Which internal team had budget.
- Which risk made the buyer nervous.
That is your product spec.
Not your pitch deck.
Mistakes To Avoid
- Selling climate virtue before you show buyer savings.
- Treating heat reuse as easy because the physics sounds simple.
- Building hardware before you know the buyer’s procurement path.
- Ignoring water chemistry and maintenance.
- Pretending every data center site can reuse heat.
- Assuming local politics will welcome a new data center.
- Selling a dashboard when the buyer needs a decision memo.
- Overusing "green" language when the buyer wants payback, permits and proof.
- Forgetting that liquid cooling changes service skills and spare-parts planning.
- Copying hyperscaler architecture when your buyer owns an older enterprise site.
The expensive mistake is assuming the hard part is the technology.
Often, the hard part is the buyer path.
What To Do This Week
Use this as a seven-day founder test.
Day 1: Pick one buyer type. Data center operator, AI cloud buyer, city heat planner, district heat utility, engineering firm, lender, insurer or hardware supplier.
Day 2: Interview three people near that buyer. Ask what makes liquid cooling, heat reuse or energy reporting annoying right now.
Day 3: Build a paid diagnostic offer. Keep it narrow: one site, one report, one decision.
Day 4: Create a landing page for the offer. Do not explain the whole market. Explain the buyer’s headache and the report they get.
Day 5: Send it to 20 relevant people. Ask for a paid call, not compliments.
Day 6: Turn objections into a checklist. Price, data access, site access, insurance, supplier risk, heat buyer, permits, reporting.
Day 7: Decide whether to sell the same diagnostic again, change the buyer, or kill the idea.
This is the Mean CEO version of climate infrastructure work:
Stop admiring the category.
Find the budget.
The Bottom Line
Liquid cooling startups are interesting because AI has made heat a business constraint.
Europe does not need another founder saying "green data centers" with soft music in the background. It needs founders who can help buyers make cheaper, cleaner and faster decisions about heat, cooling, water, power density, heat reuse and data center evidence.
The ugliest layer may be the best wedge.
Pipes. Pumps. Records. Contracts. Sensors. Reports. Heat offtake. Facility diagrams. Maintenance logs.
That is where serious founders should look.
If AI infrastructure keeps growing, the companies that explain and manage the heat will sit closer to the money than many founders expect. The next adjacent questions are already visible: grid flexibility software for power systems under stress, nuclear AI infrastructure for base-load debates, and GPU FinOps for founders who need to know what compute really costs.
FAQ
What are liquid cooling startups?
Liquid cooling startups are companies that help data centers, AI teams, facility owners, engineering firms or energy partners manage heat from dense computing systems. They may build hardware, software, services, monitoring tools, audits, leak detection, retrofit plans, heat reuse reports or cooling procurement support. For bootstrapped founders, the strongest starting point is usually a narrow paid service that helps one buyer make one cooling or heat reuse decision.
Why does AI increase demand for liquid cooling?
AI increases liquid cooling demand because AI servers can place much more power and heat into each rack. Air cooling can struggle when chip density rises. Liquid can capture heat closer to the chip or rack, which helps operators support hotter AI systems. The buyer issue is practical: can the facility add AI capacity without blowing up power, cooling, water, permit or service assumptions?
What is direct-to-chip liquid cooling?
Direct-to-chip liquid cooling uses cold plates attached to hot components such as GPUs and CPUs. Coolant flows through the cold plate and carries heat away. It is attractive for AI racks because it captures heat near the source. The startup openings around direct-to-chip cooling include readiness audits, leak detection, maintenance training, component tracking, water chemistry checks and procurement support.
What is immersion cooling?
Immersion cooling places servers or components in a special fluid that absorbs heat. It can handle dense computing loads, but it changes service routines, equipment choices, facility planning and buyer comfort. A founder should not pitch immersion as magic. The better first offer is a feasibility report that compares immersion with direct-to-chip, rear-door heat exchangers and hybrid cooling for one site.
How does heat reuse work in data centers?
Heat reuse takes excess heat from a data center and sends it to a place that can use it, such as a district heating network, campus, housing block, greenhouse or public building. The project usually needs heat exchangers, heat pumps, pipes, contracts, monitoring and local partners. It works best when heat demand is nearby and steady enough to justify the cost and coordination.
Can bootstrapped founders enter liquid cooling?
Yes, but they should start with the proof layer, not the heaviest capital layer. A small founder can sell audits, feasibility notes, data cleanup, monitoring pilots, buyer reports, supplier comparisons, heat reuse partner search or EU reporting prep. Those offers create buyer knowledge and revenue before the founder decides whether to build software, sensors or hardware.
What should founders sell first?
Founders should sell the smallest paid decision aid that a buyer understands. Good first offers include an AI rack readiness report, a heat reuse feasibility memo, a liquid cooling vendor comparison, a leak-risk checklist, a water chemistry service or a cooling evidence pack for lenders and insurers. The goal is to learn which buyer has urgency, data and budget.
How does liquid cooling connect with data center energy demand?
Liquid cooling connects with data center energy demand because AI workloads increase heat density and can make cooling a bigger part of facility planning. Better cooling can support denser racks, change water and energy choices, open heat reuse paths and affect the cost of AI capacity. Founders building AI products should care because data center constraints can show up later as cloud prices, capacity shortages or buyer questions.
What are the risks in heat reuse projects?
Heat reuse projects can fail when the heat source is too far from demand, the temperature is too low, the seasonal pattern is wrong, contracts are weak, local partners disagree, or the data center owner cannot justify the work. Founders should treat heat reuse as a commercial project with physics inside it, not as a nice environmental story.
What should female founders know about liquid cooling startups?
Female founders should know that liquid cooling is an opening into serious infrastructure without needing permission from the usual startup club. The market needs technical confidence, buyer discipline, documentation, sales, and the courage to ask boring questions that reveal money. If you can help a buyer turn AI heat into a decision, a report, a contract or a paid pilot, you belong in the room.