Similarities in the Bitcoin and AI Economies
A subscriber recently observed that trust in AI-generated content collapses the moment a reader spots a hallucination. Once the classic “this isn’t X, it’s Y” error appears, the relationship changes. Readers stop accepting what they see and start verifying every claim. This shifts the work of reading from comprehension to validation.
The parallel to Bitcoin is immediate. Bitcoin’s foundational principle is “don’t trust, verify”. The system assumes participants cannot trust what they see (an address balance, a transaction) without verifying the cryptographic chain that produced it. AI content now operates the same way, but for different reasons. Bitcoin makes verification possible through mathematics. AI makes verification necessary through unreliability.
This observation reveals deeper similarities between these two economies that extend beyond the verification problem.
Global, Free Market Technologies
Both Bitcoin and AI developed outside traditional government research programmes. Neither required state funding or direction. Both emerged from distributed communities operating across borders, implementing open protocols that anyone could adopt and modify.
The technologies themselves are languageless and jurisdictionless. A Bitcoin transaction validates identically whether initiated in Tokyo or London. An LLM processes text without reference to the legal framework governing its deployment. This creates global markets for both computational products (hashing power, inference capacity) and the outputs those products generate (verified transactions, generated text).
Neither technology respects geographic boundaries or regulatory frameworks by design. Both create economic activity that moves faster than policy can contain it.
Shared Infrastructure Patterns
The surface similarities are visible in the infrastructure both require:
Energy intensive: Bitcoin mining and AI training both consume substantial electrical power. Data centres supporting either operation show similar power profiles and cooling requirements.
Distributed: Neither technology concentrates in single locations. Bitcoin nodes run globally. AI inference happens across cloud providers and edge devices. The protocols enable anyone to participate.
Digital-first: Both exist entirely in computational space. Physical control is impractical. Attempts to restrict either technology through hardware export controls or infrastructure seizure face the same challenges. The protocols persist as long as any implementation exists anywhere.
Specialised hardware: Bitcoin moved from CPUs to GPUs to ASICs. AI training moved from consumer GPUs to data centre accelerators (H100s, TPUs). Both followed the same trajectory from commodity hardware to purpose-built silicon.
Protocol-based development: Bitcoin Core, alternative implementations, and forks demonstrate how open protocols enable competing implementations. AI follows the same pattern. TensorFlow, PyTorch, JAX, and dozens of other frameworks implement similar capabilities. Libraries rise and fall based on community adoption rather than centralised control. Bitcoin Core currently faces criticism over perceived corporate influence. TensorFlow lost ground to PyTorch despite Google’s backing. The pattern holds.
Energy-Derived Units of Value
The comparison becomes precise when examining the fundamental unit of economic value in each system.
Bitcoin’s unit is the hash. Miners expend electrical energy performing SHA-256 operations. The network difficulty adjusts so that computational work (measured in hashes) correlates directly with energy expenditure. Hash rate is measured in hashes per joule. This creates a direct relationship between electrical energy and the computational proof that secures the blockchain.
AI’s unit is the token. Training and inference expend electrical energy processing transformer operations. Model providers measure efficiency in tokens per joule. Larger models require more energy per token. Optimised architectures improve the energy efficiency of token generation. The relationship between electrical energy and computational output is similarly direct.
Both systems convert electrical energy into discrete computational units that carry economic value. Bitcoin hashes secure transaction history. AI tokens generate information. The mechanisms differ but the economic structure is identical: energy becomes computation becomes value.
This parallel makes the economies directly comparable. A data centre running Bitcoin mining hardware and a data centre running AI inference hardware perform the same fundamental operation: converting grid electricity into computational outputs sold on a market. The hash rate and token throughput measure productive capacity in equivalent terms.
Energy Dependence and Geographic Distribution
Both economies require energy-rich countries to exist and develop. Bitcoin mining concentrates in regions with cheap electricity (historically China, then Kazakhstan, now United States). AI development concentrates in countries with grid capacity to support large-scale training (United States, China, France).
This creates geographic constraints despite the digital-first nature of both technologies. Countries without surplus electrical capacity cannot participate in either economy at scale. The protocols may be global and languageless, but the infrastructure that implements them depends on local energy resources.
The energy requirement also creates volatility. Bitcoin mining operations migrate when energy costs change or regulations shift. AI training costs fluctuate with electricity prices and data centre capacity. Both markets respond to the same input: the cost of converting grid power into computational work.
Conclusion
Bitcoin and AI share more than surface similarities. Both convert electrical energy into discrete computational units through specialised hardware. Both operate on open protocols in global, free markets outside traditional government funding models. Both require energy-rich infrastructure to exist at scale.
Neither Bitcoin nor AI has found its “killer app”. Perhaps that’s because there isn’t one to find. These technologies transform societies so fundamentally that adoption happens wholesale, not through specific applications. The hash becomes a unit of security. The token becomes a unit of computation. A society can now measure its information security and processing capability as functions of energy generating capacity. That’s the killer app: information security and processing become engineering problems. We can measure them, and we know how to make the numbers go up.