The AI Boom's Hidden Bottleneck Isn't Only Power. It's Water.
Updated: Aug 17
Published as part of iQ-LOOP's Insights series.
A LinkedIn post by Prof. Steve Keen circulated widely this week, and it is worth engaging with directly because it identifies a constraint that much of the AI infrastructure conversation still skips past.
Every discussion about the limits of the AI buildout eventually arrives at electricity:
• Not enough gigawatts.
• Not enough turbines.
• Not enough grid capacity.
• Not enough time.
That conceptualisation right in some ways, but it overlooks another constraint that is already influencing programs on the ground.
Water.
Data centres require continuous cooling, and in many locations that cooling still depends heavily on water drawn from the same regional supplies used by households, agriculture and existing industry.
As data-centre development accelerates, this creates an increasingly difficult question.
What happens when digital infrastructure and local communities begin competing for the same finite water resource?
This is no longer a hypothetical trade-off.
Across water-stressed regions, proposed data-centre developments are facing greater scrutiny from local authorities and residents. The issue is not opposition to AI or digital infrastructure. It is the practical reality that large new industrial users cannot assume unlimited access to water simply because the investment capital is available.
The power side of the story is just as constrained.
Data-center developers are entering grid-interconnection queues that already extend years into the future. New generating capacity, transmission infrastructure and major industrial equipment all operate on long development and manufacturing cycles.
The result is a fundamental mismatch.
AI investment is moving on a hyperscale technology timeline.
Energy, water, and infrastructure still move on an industrial timeline.
Capital can be committed quickly.
• A grid connection cannot.
• A new turbine cannot.
• A new transmission line cannot.
• A new regional water system cannot.
That mismatch may become one of the defining constraints on the next phase of AI growth.
The problem is not only supply
The industry is not simply looking for more electricity.
It is looking for reliable electricity.
It is looking for power that can be delivered on schedule.
It is looking for greater certainty over future energy costs.
And increasingly, governments and operators are asking where that energy comes from, who controls it, and whether the supply chain is strategically secure.
That means the real infrastructure requirement is broader than generation alone.
It includes:
• Reliable power.
• Adequate cooling.
• Lower water consumption.
• Greater pricing certainty.
• And a secure, sovereign source of energy.
Most proposed solutions address one or two of these requirements.
The more interesting question is whether they can be addressed together.
The question may not simply be how to generate more
Much of the current response is focused on building more infrastructure.
More power stations.
More transmission.
More batteries.
More cooling capacity.
All of those will be required.
But there is another question worth asking.
What if some of the energy required by data centres did not have to arrive through the electricity grid at the exact moment it was needed?
What if energy could be produced elsewhere, stored safely, transported using existing logistics and positioned close to demand in advance?
That is where aluminium becomes interesting.
Aluminium is more than a construction material.
Most people think of aluminium as a material used in aircraft, vehicles, buildings, packaging and electrical systems.
But aluminium also represents a substantial amount of embodied electrical energy.
Producing aluminium from alumina requires large quantities of electricity. Once produced, the metal is stable, solid and easily handled.
• It can be stored for long periods.
• It can be transported by road, rail or ship.
• It does not require cryogenic storage.
• It does not need to be compressed.
It can be counted, insured, warehoused, and managed as inventory.
This creates a different way of thinking about energy.
Instead of requiring electricity to travel continuously through wires, some energy can potentially be converted into a physical material and moved through conventional industrial logistics.
• The electricity is used where it is available.
• The resulting energy inventory is then moved to where it is needed.
• This is the principle behind iQ-LOOP.
Energy that can move as inventory
iQ-LOOP is developing a circular aluminium energy platform.
Surplus, off-peak or low-cost electricity is used to restore aluminium inventory.
That aluminium can then be formed into transportable FuelBlades and positioned close to the point of demand.
When required, the stored energy is released locally.
The aluminium is not intended to be used once and discarded. It is recovered and returned for regeneration, creating a circular material and energy system.
• The important point is not simply the machinery.
• It is the possibility of treating energy as inventory.
A stable solid that can be produced where electricity is abundant, moved through existing logistics, and stored close to the customer.
That matters because a material-based energy inventory is not permanently tied to one power station, one transmission route, or one grid connection.
It can move.
Energy would no longer be purchased only as a flow.
It could also be owned and managed as inventory.
Why this matters for data centres
Data centres require more than electricity.
They also require continuous cooling.
In conventional systems, that cooling can impose substantial additional electrical demand and, depending on the cooling architecture, significant water consumption.
iQ-LOOP is being designed to provide useful thermal energy alongside power.
That thermal energy can drive absorption cooling, reducing dependence on electrically driven compressor chillers and avoiding the need to place every element of the cooling demand onto an already constrained grid connection.
The system is also being designed around dry heat rejection and closed-loop water management.
Preliminary engineering indicates that iQ-LOOP has the potential to reduce onsite water usage by up to 80% compared with conventional evaporative cooling systems, depending on climate, operating conditions and final system configuration.
It would not eliminate water use entirely.
Water remains part of the energy-release, cooling and heat-recovery processes.
The objective is therefore not to claim a water-free system.
It is to reduce water consumption while delivering power and cooling simultaneously.
That distinction matters.
Grid-independent, not anti-grid
For a data-centre operator, the practical consequence could be significant.
New power and cooling capacity would not necessarily have to wait entirely on a multi-year interconnection queue, a new gas turbine order, or a major regional water infrastructure project.
The energy inventory could be regenerated where electricity is available and then transported to the site.
At the point of use, the system could provide a grid-independent or grid-supported source of power and cooling, depending on the operator's requirements and the final configuration.
That does not mean disconnecting every site from the grid.
It means giving operators another option.
A site could remain grid-connected while using local energy inventory to reduce peak demand, strengthen resilience, or avoid adding load to a constrained connection.
In other situations, the same principle could support more independent operation where grid access is limited, delayed, or unreliable.
The important point is flexibility.
The operator is no longer entirely dependent on every unit of energy arriving through the local grid at the precise moment it is needed.
From volatile pricing to greater certainty
Availability is only one side of the problem.
Price matters too.
A data centre is a long-life infrastructure asset. Its economics depend not only on whether power is available today, but on what that power may cost over many years of operation.
Wholesale electricity prices can fluctuate.
Network charges can change.
Capacity constraints can increase local costs.
Fuel markets can move sharply in response to events far beyond the operator's control.
An aluminium energy inventory can separate part of the customer's energy requirement from short-term market volatility.
The inventory could be regenerated using electricity secured under long-term contracts, at known locations and during defined operating periods.
That does not make every future cost perfectly fixed.
Regeneration, transport, maintenance and financing costs will still matter.
But it may provide something operators value almost as much as low cost:
Greater pricing certainty.
Rather than purchasing every unit of energy at the prevailing price at the point of use, an operator could contract, store and manage a physical energy inventory over a longer time horizon.
That could make future energy costs more predictable and give operators greater control over their facilities' economics.
Sovereign energy supply
There is also a strategic dimension.
Many countries remain dependent on imported fuels, international pipelines, overseas refining capacity or infrastructure controlled beyond their own borders.
Data centres are increasingly regarded as critical national infrastructure.
The question of where their energy comes from is therefore not only commercial.
It is also about resilience, security and national control.
A circular aluminium energy system could regenerate energy inventory domestically using national electricity resources.
That electricity could come from wind, solar, hydro, nuclear or any other suitable generation source.
The resulting inventory could then be stored within the country, transported through domestic logistics and positioned close to nationally important infrastructure.
This creates the possibility of a more sovereign energy supply.
Not because every part of the supply chain would automatically be domestic.
And not because aluminium removes all international dependencies.
But because the energy itself could be regenerated, stored and controlled within national infrastructure rather than remaining permanently dependent on imported fuels or real-time cross-border supply.
For governments, defence organisations and operators of critical infrastructure, that distinction may become increasingly important.
The constraint is infrastructure, not ambition
The AI sector does not lack capital.
It does not lack demand.
It does not lack ambition.
Its problem is that physical infrastructure cannot always be delivered at the speed of digital investment.
A data-centre operator may be ready to build, but still face delays tied to grid access, generation capacity, cooling infrastructure, water availability, and permitting.
A transportable aluminium energy inventory offers a different possibility.
Energy can be regenerated where electricity is available.
The inventory can be transported and positioned behind the meter, closer to the point of use.
Power and cooling can then be delivered locally, with the potential for:
• Grid-independent or grid-supported operation.
• Up to 80% lower on-site water usage compared with conventional evaporative cooling.
• Greater long-term pricing certainty.
• And a more secure, sovereign energy supply.
This does not remove every infrastructure constraint.
It does not replace the grid.
It does not replace batteries, renewables, nuclear generation or gas turbines.
But it may provide another route for supplying power and cooling without placing all requirements on the same constrained local infrastructure.
The bill, not just the machine
The real question is not simply whether a new machine can be built.
It is whether the wider infrastructure burden can be reduced.
Can new power demand be added without waiting years for a larger grid connection?
Can cooling be delivered without placing the same pressure on local water resources?
Can energy be positioned close to demand before the system becomes constrained?
Can operators gain greater certainty over future energy costs?
Can countries retain greater control over the energy supporting their critical digital infrastructure?
These are the questions iQ-LOOP is now working to answer.
We are advancing a 1 MW system design for the Canadian market, including configurations for power, heat and cooling.
We are in active discussions around a first pilot, and the response so far gives us confidence that the underlying problem is both real and urgent.
We are not proposing that aluminium replaces the electricity grid.
We are proposing that it gives operators another option.
A grid-independent or grid-supported source of power and cooling that can be positioned where it is needed, when it is needed.
A system with the potential to reduce onsite water usage by up to 80%.
An energy inventory that could provide greater long-term pricing certainty.
And a supply model that could strengthen national energy security by allowing energy to be regenerated, stored and controlled domestically.
For data-centre operators facing the combined constraints of power, cooling, water, price and time, that may be the more important conversation.
Because the hidden bottleneck in the AI boom is not simply how much electricity can be generated.
It is whether secure, predictable and sufficiently sovereign energy and cooling can be delivered where they are needed, without exhausting the infrastructure and water systems already in place.
Comments