It took the internet nearly ten years to reach 100 million users. Smartphones took just over five. But when ChatGPT came along, it hit that milestone in a matter of weeks. This isn’t just a quirky fact about speed, it’s a sign of something deeper.
Not just about how fast we adopt new tools, but about how quickly technology is starting to reshape the systems around it.
Because when adoption moves that fast, the challenge isn’t creating demand, it’s meeting it. And that’s why the real story in AI today isn’t intelligence anymore: it’s infrastructure.
AI no longer behaves like traditional software, running quietly in the background. Instead, it’s beginning to look, and act, like an industrial system.
Every prompt you send travels through a physical chain: processors, power grids, cooling systems, even water networks. In other words, intelligence has become physical.
And like every industrial revolution before it, the limits now aren’t just technological, they’re structural. Not about what models can think, but what our systems can support.
This shift is revealing four hard constraints that will define the next phase of AI.
- First, there’s compute: the most advanced chips are in such short supply that some delivery times now reach into 2027.
- Then there’s power: a single model rollout can add 70 megawatts to a data center’s load, roughly the energy needs of 50,000 homes.
- There’s also water: cooling these machines can consume millions of litres per year, especially in regions already under pressure.
- And finally, geopolitics: as AI expands, data is becoming a strategic asset, one that’s regulated, restricted, and increasingly nationalised.
This calls for a fundamental shift in how investors think about AI.
The opportunity isn’t only about who builds the smartest models anymore, it’s also about who builds the systems that make those models run.
We’re entering a classic capital cycle.
The first wave is already underway: a massive buildout of chips, energy infrastructure, and cooling capacity.
But eventually, the focus will shift. Efficiency will take over: doing more with less, and maximising what’s already been built.
History tells us that the biggest returns rarely go to the pioneers alone.
They often go to those who quietly build the foundation underneath the breakthroughs.
AI may be advancing at remarkable speed, yet it remains constrained by tangible, physical limits.
As the industrial age of AI takes shape, the fundamental question is no longer who will develop the most advanced models.
But who will build, own, and sustain the infrastructure that enables them.
