The public has told us plainly what it will no longer accept: data centers that take its water, strain its grid, and raise its bills.

Within months, data centers have become a political hot button, threatening to spiral out of control. Gallup, polling in March 2026, found 71 percent of Americans oppose an AI data center in their area, against 53 percent who would oppose a nuclear power plant. Heatmap watched opposition climb from 42 to 71 percent in nine months. Asked why, opponents cluster on water, power, and the utility bills they expect to follow. Now, the politicians are following suit.

The public perception is so acute that no amount of marketing will bring them back. The only answer for our industry is to stop insisting our designs use the resources that the public refuses to give us – their water, their grid headroom, and their land.

Luckily, we already have the solution: sealing the server.

The building was the wrong unit of design

For most of our history we treated the facility as the unit of design and the environment as the cooling system, pulling in air, evaporating water, relying on the world outside the walls to carry our heat away. Almost every objection the public now raises flows from that one decision.

Change the unit of design from the building to the server and the equation changes with it. Seal the server, take the heat at the die, carry it in a sealed circuit and reject it through a sealed exchanger, and the machine stops depending on the room it sits in.

Proving it where its hardest

In December 2024 we field trialled two sealed prototype servers outdoors at a telecom site in Arlington, Texas, unattended for eight months with no service visit and no component change, through ambient from minus 7°C to 40°C and solar loading above 1,000 watts per square meter. Inside the sealed enclosure the air reached 59°C. There was no throttling, no instability and no hardware fault. The only downtime was the site losing power.

Same enclosure, same day, same load: the liquid cooled network interfaces held between 55 and 65°C while the air cooled one sat at 80 to 85°C. On teardown, the outside had collected dust and dirt; inside there was no water ingress, no dust, no corrosion, and coolant chemistry that had not moved in eight months.

What that changes for the community

The effect compounds across four fronts:

• Water usage goes to zero. Because the loop returns water above 60°C, that heat can be rejected through a dry cooler instead of an evaporative tower, the exact step the Water Foundation singles out as the main water driver.
• It attacks power where the waste actually is. Fans do not vanish; a liquid-first design cuts their draw rather than eliminating it. What disappears is the building's air handling, the CRAHs and plenums whose only job is moving air around a room. In Arlington, cooling a unit drawing 350 to 380 watts took around 40 watts typically, peaking at 80 under stress.
• It provides heat as a usable energy asset for the community. Water returning above 60°C is usable: district heating, process heat, freeze prevention, an energy offset that belongs to the community hosting the building rather than the operator.
• It shrinks the building. Modelled with Collen Construction and four engineering partners, a 2.5MW retrofit turns 310 air cooled racks at 8kW into 100 racks at 25kW, freeing about a third of the floor plate, and the gap widens as AI racks pass 100kW.

The commercial consequence

One thing follows: when the sealed server carries its own thermal environment, cooling stops being a facilities decision taken late and per site. It becomes a component specification, a qualified sub-assembly with known performance, bought and warranted like any other part of the machine. That is the shift we think defines the AI era in infrastructure. Not a new cooling product inside the old model, but a new layer in the stack that OEMs and operators design around.

The fair challenge is not whether it works, but what it costs and who fixes it at three in the morning. A sealed sub-assembly with no field serviceable fluid path answers the second: nothing to open, nothing to top up, nothing to spill. A closed loop still has to put its heat somewhere; running it hot, without evaporation, is what shrinks that cost rather than relocating it.

The offer

The natural solution to this public discussion is an architecture that seals the server, closes the loop and stops asking the environment for favors.

Get it right and there is something specific to bring to the next community meeting: no evaporative cooling water from the local supply, a lower cooling load on the grid, a building a third smaller for the same compute, and waste heat hot enough to be worth something to the people living beside it. That is an offer rather than an apology.