The cheapest megawatt is the one you already have.

Search AI infrastructure and you will find billion-dollar campuses, record power commitments and GPU clusters measured in the tens of thousands. Those projects are real. They are also not the decision most organisations are actually making.

The more common decision looks like this: a workload needs to grow, the racks it would grow into are already at their thermal ceiling, and somebody proposes leasing more space. That is a defensible answer. It is frequently not the cheapest one, and the gap between the two is large enough to be worth an afternoon.

Power at the meter is not the same as usable density

Two constraints get conflated constantly.

The first is power — what the utility will deliver to the building. When that is maxed out, no amount of clever engineering helps, and you are into interconnection queues and bridge generation.

The second is heat rejection — whether the cooling system can remove the heat that power would produce once it is drawn. This is where a great many facilities actually stop. There is capacity at the meter, there is space on the floor, and the racks cannot go denser because the air handling cannot cope.

The distinction matters because the two have completely different remedies and completely different price tags. New capacity is a multi-year, multi-million commitment. A cooling upgrade is measured in weeks and does not require a new lease.

Before signing for more space, it is worth establishing which of the two is actually stopping you. In our experience it is not always the one people assume.

Where OptiCool fits

OptiCool Technologies is a partner in our ecosystem, and this is the problem they address. Their Delta4T™ system uses two-phase refrigerant cooling rather than conventional air.

The mechanism is worth understanding, because it explains the performance. Conventional cooling floods a hot aisle with cold air to mask heat. A two-phase system instead uses the latent heat of a vaporising refrigerant as the heat sink — the coolant absorbs heat by changing state rather than merely warming up. Because a phase change carries far more energy per unit of volume than a temperature change in water, the system moves the same heat with dramatically less fluid and much lower pumping energy.

MITRE's chief engineer put the practical consequence well: the approach extracts considerably more heat energy than water can, while circulating small volumes of refrigerant instead of many gallons per cabinet.

The record

• Over 7,000 active rear-door heat exchangers deployed across more than 125 sites — this is not a pilot technology.
• A 120 kW rear-door heat exchanger absorbing an 80 kW rack load while drawing 550 W, with surplus capacity left over to cool the surrounding room. From an aerospace deployment, and the ratio is the point: the heat removed against the energy spent removing it.
• Deployments at MITRE, Sandia National Laboratories, Comcast, T-Mobile and AT&T among others, plus named partnerships with Sabey Data Centers, CoreSite and Belden.
• Water-free at the rack. Refrigerant-based rather than water-based heat rejection.

Why this matters commercially

1. It changes which transaction you are doing, and the economics of it. You are still buying something — but a rack-level cooling upgrade instead of a multi-year lease on new space. If a retrofit takes existing racks from 20 kW to 80 kW, the capacity you were about to sign for may already be sitting in a hall you pay rent on. Same compute, a materially smaller commitment, and no new term to negotiate.
2. Energy cost is the recurring half of the return. Cooling is one of the largest operating line items in any dense facility, and it is a cost you carry every month for the life of the deployment. Delta4T™ delivers energy reductions of up to 90% against conventional approaches, and the aerospace case study is the mechanism in miniature: 80 kW of rack heat absorbed for 550 W of draw. On any dense footprint, the operating saving is usually the larger number over the term, not the capex avoided.
3. Water-free sidesteps the fastest-moving permitting risk in the sector. Water consumption is becoming the constraint that stalls projects and attracts community opposition — it is on our own diligence list for exactly that reason. A refrigerant-based system removes that exposure at the rack.
4. Rack-level and modular means capex tracks demand. You are not underwriting a facility-wide cooling plant for density you may not need for three years.
5. Non-invasive retrofit protects live workloads. The aerospace deployment scaled cooling without service disruption, which is the difference between a project that gets approved and one that does not.
6. It gives you a credible alternative — and that is leverage. This is the part most buyers miss. The single most expensive habit in infrastructure procurement is arriving at a negotiation with no option but to sign. If densifying your existing footprint is genuinely viable, you are negotiating from a different position entirely, whether or not you end up using it.

If you are the operator, the maths runs the other way

Everything above is written from the buyer's side. For an operator or colocation provider the same technology is a revenue argument, and a sharper one.

• Revenue per square foot. A hall that can only sell 10 kW cabinets is leaving money on a floor you have already built, permitted and financed. Raising deliverable density raises what that same footprint earns — without new construction, new power or new entitlement risk.
• Access to the workloads that actually pay. High-density AI tenants are the demand with budget and urgency, and they will not shortlist a facility that cannot cool them. Density is increasingly the qualifying criterion, not a premium feature.
• Lower operating cost on the space you keep. Cooling energy you do not spend is margin, whether the hall is full or filling.
• Deploy against live workloads. Retrofitting rack by rack means you are not taking revenue offline to increase what the hall can carry.

If that is your position, the commercialization mandate is the relevant page — we bring qualified demand to operators, and deliverable density is one of the first things that demand asks about.

Where this is the wrong answer

A partner spotlight that only lists upside is a brochure, so here is the other half. A cooling retrofit will not solve your problem when:

• The utility feed is genuinely maxed. Heat rejection is irrelevant if the power is not there. Establish this first.
• Floor loading or structural provision cannot carry the rack spec. High-density racks are heavy, and retrofitting structure is expensive or impossible.
• The target density requires cooling at the die. Rack-level heat rejection has a ceiling; beyond it you are into direct-to-chip, and the specification changes.
• You are a colocation tenant without consent. Your landlord has to permit the modification, and the commercial terms of doing so are their own negotiation — one we are happy to run.

What we would do with this

If you are weighing a capacity decision, the sequence is short. Establish whether power or heat rejection is the binding constraint. If it is heat rejection, price a retrofit against the lease you were about to sign — including the terms, not just the rate. If it is power, the retrofit is a distraction and the power mandate is the right route.

We can put you in front of OptiCool directly. We are not the seller in either route, and we have no reason to prefer the larger deal — what we are paid to protect is the relationship, which means saying so when the cooling is the cheaper transaction than the lease.

Related: high-density colocation sourcing, the capacity diligence list, and power procurement. More on the technology at opticooltechnologies.com.