[ad_1]
Ray Dalio has been saying for a while that the dollar is in trouble. Not right now, not tomorrow, but rather at a structural level. His theory on the rise and fall of empires points to an intriguing pattern: roughly every 100 years, the world’s reserve currency gets replaced.
Not because someone decides to replace it. But because change is unavoidable – and the underlying force that gave power to that currency shifts into something else.
We’re about 100 years into dollar dominance. We’re getting closer.
What Actually Backs a Reserve Currency
Reserve currencies don’t just happen out of thin air. They’re backed by whatever the dominant economic force of the era is.
Before the dollar it was the British pound, backed by the largest navy in the world and control over global trade routes. Before that, the Dutch guilder, backed by the most sophisticated merchant fleet of the time. Each transition happened because a new empire became dominant in the thing that mattered most for commerce.
For the dollar, that thing was oil.
The Petrodollar Was Never a Conspiracy
After Bretton Woods collapsed in 1971, the dollar survived, and even consolidated, because oil was priced in dollars. You want oil, you need dollars. Every country needs oil, so every country needs dollars. Simple, unavoidable, effective.
The dollar wasn’t backed only by abstract American values or military trust. It was backed by the one commodity the entire world had to buy, every single day.
What If Oil Stops Mattering?
Let’s try an exercise of imagination, and no, I’m not talking about electric cars. I’m talking about something way deeper.
AI is already beginning to do what oil did for manufacturing — becoming the input for almost everything. It’s already at the foundation of drug discovery, legal work, financial modeling, logistics, content, code. The list grows every month.
And there’s a wilder version of this argument. AI is even accelerating energy research. Fusion, which has been “20 years away” my entire lifetime, is suddenly getting real traction. Solar and battery optimization is increasingly AI-driven. If AI helps us get cheap, abundant energy, the physical scarcity of oil — the very thing that made it a geopolitical weapon — starts to dissolve.
You could make energy at home. Not today, maybe not in five years. But it may happen soon.
When that becomes a reality, the petrodollar loses its foundation.
Inference Is the New Oil
Unlike oil, which you had to drill for in specific places controlled by specific people, inference can be run anywhere you can build compute.
It has all the properties that made oil work as a backing. It’s scarce — quality compute isn’t free, and good models need tons of energy to train. It’s universally needed — every sector of the economy is becoming dependent on it. And it’s measurable. We already have a unit: the token.
Which is where PPT — price per token — becomes interesting. Not as a currency someone declares tomorrow, but as an index. The way price per barrel was the pulse of the oil economy, price per token might become the pulse of the inference economy.
The Models Keep Getting Better
Every six months, the frontier moves. What was cutting-edge a year ago is now available for almost nothing. The gap between the best proprietary model and a capable open source alternative keeps narrowing, and the compression has real consequences.
The US currently leads on proprietary models. OpenAI, Anthropic, Google DeepMind — the frontier is American, backed by an overheated investment market pouring money into compute, talent, and infrastructure.
China is doing something different. Instead of competing dollar for dollar on proprietary development, they’re doing what they’ve always done — taking the open layer and making it theirs. DeepSeek wasn’t a surprise. It was the result of a deliberate strategy: work within the open source ecosystem, optimize hard, and ship something affordable and at least as capable.
The result is that you don’t need a billion-dollar data center to run useful inference anymore. You need a decent GPU, the right model, and electricity. We’re moving toward a world where someone can have serious compute in the back of their garage and use it to generate daily income — running local models, offering inference services, solving real problems for real people.
This gives everyone a place at the table. A small place, yes, but still a place.
But having a place at the table doesn’t mean you get to eat. The concentration of power we’re describing isn’t new — it echoes patterns from history. I’ve written before about how showing up is not enough anymore. The world is increasingly run by a handful of corporations, much like the Mongolian Empire consolidated power across continents. Those who were conquered had a choice: swear allegiance and deliver real value, or be erased. In an inference economy, the math is similar. To survive, you need to generate at least 5x your current value — enough to justify your seat. To thrive, you need 100x. The table is open, but the entry fee keeps rising.
Inference Doesn’t Need a Country
I’ve been thinking a lot about this during the last few years. We’re at a point where the nation-state framing starts to break down.
The old model — one country controls the dominant resource, prices it in their currency, projects power through that control — made sense when the resource was physical. You can blockade oil, invade a country, take their president, problem solved. You can’t blockade a model weight file. You can’t invade it.
If inference becomes the primary economic force, power won’t necessarily concentrate in Washington or Beijing. It will concentrate around whoever controls the compute layer, the data pipelines, and the distribution networks. That might be a country. Or it might be a corporation. Or it might be something we don’t have a word for yet.
Neal Stephenson imagined something like this in Snow Crash, back in 1992. In that world, nation-states have fragmented into franchulates — corporate-run micro-nations, floating enclaves, sovereign territories defined not by geography but by who you pay allegiance to and what network you’re on. That famous novel reads less and less like fiction with every year.
Language barriers disappear when AI makes communication frictionless. Cultural friction softens when every interaction is mediated and translated in real time. The things that historically kept people inside national containers start to matter less. What matters is access to compute, and who sets the rules of the network you’re on.
Whoever controls the inference layer controls the economy that runs on top of it. That might look like a country. It might look like a platform. Dalio was right that the dollar is running out of road — he just observed the cycle, showing it on the map. What he didn’t map is that the next dominant force might not belong to any nation at all. The petrodollar logic was built for a world that is quietly becoming something else.
These things move slowly, and then all at once.
[ad_2]
dragos@dragosroua.com (Dragos Roua)
Source link
