The electricity grid is a ghost. It disappears when you need it most.
Trump just gave a speech about AI. He didn't talk about models. Didn't talk about chips. He talked about power plants. The man who built a reality TV empire on deals is now fixated on the reality of physics: you can't train a frontier model without boiling a river.
I've been tracking this for years. Since 2020, I manually mapped energy consumption of Bitcoin mining operations. Back then, the narrative was 'mining is bad for the environment.' Now, the same critics are silent as AI data centers guzzle 100 MW per cluster. The hypocrisy is loud. But the data is louder.
Let me stress-test the narrative.
Context: The Global Liquidity Map
Macro watchers know this: capital flows where energy is cheap. In 2021, crypto miners chased stranded natural gas. In 2023, AI companies chased nuclear power plants. The underlying asset is the same: baseload electricity. The difference is scale.
Trump's speech is a political signal. He's telling state governors: 'Ignore the environmentalists. Approve the permits. Or lose the AI race to China.' This is not a technical debate. It's a resource allocation war.
But the crowd is missing the real story. Everyone focuses on compute. They obsess over H100 allocations. They ignore the fact that US grid capacity is already maxed out. The North American Electric Reliability Corporation (NERC) warned in 2023 that peak demand growth is outpacing capacity additions. AI data centers are the straw breaking the camel's back.
Core: Crypto as a Macro Asset
Here's the original analysis. Trump's push for AI infrastructure creates a direct resource competition with crypto. Bitcoin miners and AI data centers both need high-availability, low-cost power. They both prefer locations with fast permitting (Texas, Ohio, New York). They both face public backlash.
I've audited the numbers. A typical Bitcoin mining facility uses 50-100 MW. A typical AI training cluster uses 100-200 MW. The difference is that AI demand is growing exponentially. Train a GPT-4 class model? 20,000 A100s running for months. That's 100 MW continuous. Add inference? Multiply by 10.
Now, the liquidity is a ghost. The US government is not printing money for energy. They're printing rhetoric. Trump's promise of 'fast approvals' doesn't build a single kilowatt. The actual capital expenditure for new nuclear or gas plants is $5-10 billion per GW. The construction timeline is 5-10 years. The AI compute demand is doubling every 18 months.
This is the asymmetry. The market is pricing in a future where power is abundant. The reality is a future where power is scarce. Which means something has to give.
I've spoken to a hedge fund manager in Beijing. He told me: 'The US is making the same mistake they made in 2021 with crypto. They treat energy as an infinite resource. It's not.'
Smart contracts don't enforce themselves. But physics does.
Contrarian: The Decoupling Thesis is Wrong
Everyone says crypto is decoupled from AI. They say crypto is a hedge against inflation, AI is a growth trade. They're wrong.
Both assets are sensitive to the same macro variable: the cost of energy. If Trump's policies succeed in building more power plants, energy costs drop, and both crypto and AI benefit. If they fail, energy costs spike, and both suffer.
But there's a twist. AI's demand is so large that it could crowd out crypto miners. Look at what happened in New York. The state banned proof-of-work mining. The reason? Energy consumption. The same logic is now being applied to AI data centers. In Virginia, local communities are suing to stop Amazon's 2.5 GW data center. The argument is the same: 'You're taking our power, and raising our bills.'
This is the blind spot. The market assumes AI's growth is unconstrained. It's not. The social license to build is the real bottleneck. Trump's speech is a attempt to override that license with federal authority. But even he can't stop the NIMBYs.
Takeaway: Cycle Positioning
So where does this leave us?
The macro cycle is shifting. The narrative was 'AI is the next internet.' The reality is 'AI is the next energy crisis.'
For crypto investors, the signal is clear: track energy infrastructure. The Ethereum merge was a big step. But Bitcoin still needs cheap power. If AI eats the cheap power, Bitcoin mining becomes unprofitable at scale. The hash rate will consolidate.
For macro watchers, the question is: will Trump's policies actually accelerate energy buildout? History says no. The US has been trying to build nuclear plants for decades. The only success stories are in China and South Korea.
My thesis: This is a liquidity mirage. The promise of abundant energy is just a promise. The real constraint is physical. And physical constraints always win.
Don't fight the tape. But don't buy the hype. The ghost is still in the machine.