AI is running into the physical world.
Three constraints are becoming visible at the same time: grid queue congestion, local water stress, and critical mineral concentration. The market is still pricing AI as software demand. Manathric tracks the physical constraints that decide where capacity can actually exist.
Grid
104
UK withdrawals / month
Grid connection attrition is already visible before it becomes a portfolio impairment. (Source: NESO TEC Register, 2012–2026.)
Water
4.8/5
Colorado stress screen
Data center expansion competes with local resource constraints, not abstract cloud capacity. (Source: WRI Aqueduct, Colorado water stress screen.)
Minerals
0.97
Gallium HHI
Critical mineral concentration creates an input risk that is absent from most AI capex models. (Source: USGS critical mineral data, gallium HHI calculation.)
1. Compute demand is becoming a grid queue problem.
The limiting factor is no longer only data center financing, chips, or hyperscaler demand. Connection capacity is becoming a gating variable. When withdrawal velocity rises, it is a signal that projects are discovering constraints after capital has already been allocated. (Source: NESO TEC Register, 2012–2026; withdrawal trend data.)
2. Water stress changes the map.
AI campuses are physical assets. They need power, cooling, land, and permitting. Regions with strong fiber or cheap land can still fail the deliverability screen when local water stress and grid scarcity move together. (Source: WRI Aqueduct; NESO connection data.)
3. Mineral concentration is an overlooked input risk.
Critical minerals do not need to stop supply to affect valuations. Concentration alone can change procurement timing, capex certainty, and the option value of alternative locations. (Source: USGS critical mineral data, gallium HHI 0.97.) The right research object is therefore not AI demand. It is AI demand constrained by physical bottlenecks.
Investment Implications
Investors
Treat AI infrastructure as a physical supply-chain trade, not only a compute demand story.
Developers
Prioritize regions where grid deliverability and water permits can be defended together.
Research
Track withdrawals, constraints, and sponsor execution as a single system rather than separate datasets.