“Crypto is libertarian. AI is communist.”

In his 2018 talk at Stanford University, Peter Thiel meant it as a shorthand for power, not politics.

Crypto, he said, disperses trust. It removes the need for a central authority and lets coordination emerge from the edges.

AI, he noted, is the opposite: enormous compute, concentrated data, and decision-making clustered inside a few institutions.

That framing feels right if you freeze the picture at the moment of creation. Training models is capital-heavy and centralized. Control appears to move upward.

BUT, it turns out, it’s not that simple…

In 2018, AI was still largely a lab-and-cloud phenomenon. What wasn’t yet fully obvious—even with Palantir—was small, specialized models would scatter like cockroaches into factories, grids, hospitals, and robots.

And it has a lot to do with the increasing rush into atoms.

Atoms > Bits

For the last generation, growth in America lived mostly in code.

That dynamic has put AI and crypto in the same arena as every other software story…

Competing for attention, valuation, and belief.

As a result, these emerging technologies were treated by many as weird abstractions: powerful ideas, future-facing tools, still searching for the problems to justify their scale.

But that framing is already breaking down…

And it’s because the rotation back into the physical economy is underway.

Consider that McKinsey estimates a collective $106 trillion in investment will be necessary through 2040 to meet the need for new and updated infrastructure.

Meaning, this cycle requires at least 50% more physical capital, at a significantly higher annual intensity, than the 1990s–2000s globalization buildout.

And, no less…

As this buildout progresses, the role of AI and crypto will stop being theoretical and begin to clarify.

Here’s why.

Let’s start with AI.

Coordination > Computation

AI delivers its real payoff when it has something expensive, slow, and physical to work on.

An easy example…

In environments where an hour of downtime costs hundreds of thousands of dollars, preventing a single failure justifies the system. That’s why predictive maintenance showed up as one of the first—because inefficiency is visible, measurable, and expensive.

Heavy industry uses AI to predict wear and schedule maintenance on machines that cost millions and can’t be easily replaced.

This logic is spreading faster than most realize.

Power grids use AI to forecast demand, balance load, integrate intermittent renewables, and spot failures before they cascade. Logistics networks optimize routing, fleet utilization, and warehouse flow to reduce idle time and working-capital drag.

Retailers use it to manage inventory and demand. Hospitals use it to schedule beds, staff, and equipment without building new facilities.

At first, AI squeezes existing assets—fewer outages, higher utilization, less slack.

So far, so good.

But as systems scale and interconnect, the problem shifts. AI moves from optimizing individual assets to coordinating systems.

Grids respond to factories and data centers in real time. Supply chains adapt dynamically to congestion, energy prices, and inventory stress. Production schedules adjust to shocks in days instead of weeks.

Over time, the economy becomes anticipatory. It stops reacting to failures and starts modeling how constraints propagate across power, transport, labor, and capital.

Maintenance, staffing, inventory, and routing converge into a distributed prediction layer.

To see where this may lead, look to the trend.

Small > Big

The vast majority of physical industries don’t use frontier AI models as operational workhorses.

Day-to-day operations are increasingly dominated by smaller, specialized models (SLMs).

Physical systems prioritize low latency, reliability, explainability, and cost control over raw intelligence. Small, local models run closer to the edge, integrate cleanly with existing control systems, preserve data privacy, and behave predictably under stress.

Frontier models still matter, but mainly upstream, in code generation and system design.

The net effect of this? AI lowers the friction that once made scale a requirement for efficiency.

AI shifts forecasting, scheduling, maintenance, and routing from people and process into software, making them less organization-dependent.

The buildout is heavy, but the state that follows is leaner, more competitive, and requires less organizational mass per unit of output.

So, in short: the future rarely reads the memo.

And opportunities abound as a result (more on that soon in our Paradigm Mastermind Group).

But wait…

What of “physical AI”. Isn’t it an inherently centralizing force?

Well…

The result, once again, is counterintuitive.

More tomorrow.