A growing number of people think they’ll wake up tomorrow and find that AI has replaced them at work. That’s not gonna happen for most people because the bottlenecks for AI growth are too real. Energy, compute, and governance all constrain how fast AI can move through the economy. The bottom line being that commoditized intelligence won’t arrive as a single shock and awe moment but will emerge methodically as a long march.
That matters because the speed of change will shape real decisions long before the endpoint is obvious. Where to live, what to work on, what to invest in, which industries are expanding, which ones are being compressed, and what kind of world to prepare your children for. If AI is going to reorganize society, then the important question is not whether change is coming but how fast.
That’s why I want to track the bottlenecks instead of the discourse. I don’t place much trust in the judgment of the crowd on X or LinkedIn, although I do find signal in it. I want to get as close to primary data as possible because if you can measure the constraints on AI adoption, you can build a better read on the actual velocity of change.
Energy is one clear lens. Capital is pouring into power generation, grid expansion, and related infrastructure. To be clear: the signal is not just that money is flowing but also whether projects are getting approved, financed, built, and delivered on time. Compute follows the same logic. Data center construction, semiconductor supply, capex, utilization, and local political resistance all tell you something about how quickly AI capacity can scale in the real world.
Labor is another signal. Where are companies hiring because AI is expanding their frontier? Where are they cutting because automation is beginning to compress headcount? Watch the pressure points. Watch which roles are gaining leverage and which ones are being absorbed into software. Then layer in commodities. If AI infrastructure is truly scaling, demand should show up in raw materials, equipment orders, and capacity expansion across the supply chain. Each vertical gives you one piece of the picture. Stack them together and you get a much clearer read on the pace of change.
From there, the framework gets more interesting. Migration data from sources like U-Haul and Zillow can show where economic gravity is shifting. Are people moving toward regions benefiting from energy and compute buildout, or away from places making that buildout difficult? Governance matters here too. Local regulation, permitting friction, and political hostility can slow deployment just as surely as a physical shortage can. Are people moving toward places with welcoming AI regulation or toward antagonistic AI regulation?
You can go even deeper. Track new AI-specific chip architectures, fab expansion, grid interconnection queues, and model capability benchmarks. Watch pricing. Frontier labs are still heavily subsidizing model usage, and it’s hard to see that as permanent under the current infrastructure regime. Too much demand is chasing too little compute, supported by an energy system that is not scaling fast enough. If pricing changes, that tells you something important. It tells you where intelligence is becoming abundant, where it is still scarce, and which use cases are valuable enough to justify the cost.
That’s the real point.
We know AI will change the world. What we don’t know is the rate at which that change will propagate through society. The bottlenecks give us a way to measure it. If you want to understand how fast the future is arriving, stop watching the noise and start tracking the constraints.