AI Does Not Scale Like Software
Why AI scaling is constrained by physical reality.
The scaling misconception
Boards and executives often assume that scaling AI resembles scaling software: elastic and inexpensive.
It does not.
Artificial intelligence grows through physical systems (energy, chips, cooling, land, infrastructure, and supply chains) and each of these introduces constraints that software never had to face.
The issue here is the widening gap between how leaders think AI scales and what scaling AI actually requires.
When digital assumptions meet physical limits
In classical software, scale is largely a design question.
In AI, scale becomes a negotiation with physics.
Systems that appear infinitely expandable at the interface level quickly encounter physical limits beneath:
· power grids,
· semiconductor capacity,
· cooling density,
· data center infrastructure,
· network throughput,
· and geopolitical access to compute.
The challenge for organisations is to understand the industrial conditions they are stepping into.
How misunderstanding accumulates
Misalignment rarely appears as failure.
It builds progressively:
· Software assumptions are applied to physical systems.
· Infrastructure dependencies remain underestimated or invisible.
· Scaling decisions are made without full visibility on constraints and exposure.
· Execution accelerates faster than oversight can adapt.
Each step appears reasonable on its own.
Across steps, organisations lose sight of what scaling actually commits them to.
What remains constant
Technologies evolve rapidly (architectures, models, and platforms continuously shift) but one requirement does not change: leaders must understand the system they are scaling.
Because the constraints are physical, economic, and increasingly geopolitical, they cannot be abstracted away.
They must be recognised and governed.
The challenge for leadership
For boards and executive teams, the challenge goes beyond mastering the technical details of AI infrastructure.
It becomes to take decisions under conditions where:
· the cost of scale is opaque,
· dependencies are global,
· and risks accumulate outside the software layer.
The question “Can we scale?” becomes “What does scaling commit us to?”
The scaling tensions
Scaling AI creates recurring tensions that leadership cannot fully eliminate:
· expanding capacity without losing control,
· increasing speed without weakening oversight,
· depending on infrastructure that sits outside the organisation,
· accelerating deployment while visibility decreases,
· pursuing scale while exposure accumulates.
Together, they reshape what responsible scaling actually requires.
In such environments, technological ambition alone is not enough.
Scaling decisions require clear anchor points where constraints are explicitly recognised and strategic dependencies deliberately assessed.
Decision anchor
As AI systems scale, organisations can become dependent on infrastructure, energy, and compute layers they neither control nor fully understand.
This creates the risk of committing to scaling assumptions that remain structurally fragile.
If leaders cannot clearly explain the dependencies, constraints, and trade-offs behind an AI scaling decision, the decision is not understood and should not be taken.
Executive Reflection
Artificial intelligence changes how easily accountability can become obscured by assumptions of infinite scale.
AI does not scale like software. It scales like industry.
And in industry, every decision carries weight, friction, dependency, and consequence.
Understanding this is not a technical requirement but leadership responsibility.
The full AI as an Industrial System concept, including the supply chain, infrastructure dependencies, and control layers shaping AI scaling, is developed in the white paper:
[AI Does Not Scale Like Software] It can be read online or downloaded.
Igor Allinckx
AI & Governance
May 2026
Back to Insights
Part of an ongoing exploration of governance, AI, and human judgment.