The price of a single thought now has a meter.
Each token costs a fraction of a cent, minted in distant data centres where electricity turns into language. Heat rises from racks of GPUs, electricity becomes syntax, and meaning gets assembled, sold, and forgotten. Somewhere nice, a tree dies. This is how we have chosen to price intelligence.
There is a temptation to look backward for a map, and Bitcoin offers one. It is not a clean map, or even a safe one, but it is a good place to start. Bitcoin is built on a hard limit – 21 million coins, no more, ever – a designed scarcity that governs everything that follows. LLM tokens have no such ceiling. Their supply expands with capital, with silicon, with ambition. That difference matters. It is structural. It cannot be waved away.
And still, the comparison holds just long enough to teach us something, because beneath both systems lies the same quiet truth: computation is not abstract – it is physical. It costs.
Bitcoin miners learned this first. They learned it in electricity bills and burned-out hardware, in the steady arithmetic of survival. When the market price of Bitcoin falls toward the cost of producing it, miners begin to fail. They shut down, and the network sheds them like dead skin. What remains is a floor, a price below which the system resists falling for long.
LLM tokens carry their own version of this floor. Each generated word requires compute: GPU cycles, memory bandwidth, cooling systems that fight entropy hour by hour. Providers can subsidise costs for a time, can blur margins with strategy and scale, but they cannot escape the underlying physics. If they price tokens below the cost of running the machines, they go broke.
This is the spine of the analogy. Energy binds both systems to reality.
But cost alone does not explain price. It never does.
Bitcoin’s most dramatic movements came not from changes in mining cost but from waves of belief. In 2020 and 2021, demand surged and previously sceptical institutions entered, narratives hardened, and speculation fed on itself. The price rose far above the cost of production, and the gap between what it cost to make a coin and what someone would pay for it became a space filled with expectation, fear, and momentum.
Something similar has begun to happen with LLM tokens. Demand has arrived quickly and unevenly. Enterprises are embedding models built on token-thirty agents that they don’t fully understand into their workflows. Consumers, meanwhile, ask questions they once carried alone. Here’s where it gets interesting. Companies like OpenAI and Anthropic price their most capable systems at levels that reflect not just cost, but value–perceived, contested, and expanding. The price of a token becomes less about the electricity required to produce it and more about what that token can replace: labour, time, and uncertainty.
As demand stretches the system, it creates a premium above the floor. And then comes efficiency, the quiet antagonist in both stories.
Bitcoin mining evolved through generations of hardware, from CPUs to GPUs to ASICs, each leap making it cheaper to produce a unit of hash. In theory, this should have driven prices down. In practice, it compressed margins until demand caught up, then expanded the system again. Efficiency did not kill value. It fed growth.
LLMs are undergoing their own acceleration. Engineers distil models, compress them, and then route computation through specialised pathways. Each innovation reduces the cost per token so that the same amount of language can be produced with less energy, less time, less hardware.
This should mean cheaper tokens. Eventually.
But not yet. Not fully. That’s because demand is still outrunning efficiency. Every gain in cost is met with an expansion in use. Cheaper tokens do not reduce total spending. Instead, they invite even more questions, more integrations, and more reliance. The system expands around its own improvements.
It is here that the analogy with Bitcoin begins to strain.
Bitcoin’s supply does not respond to demand. It cannot. The protocol enforces scarcity with indifference. Every four years, the system tightens itself through the halving, basically cutting the issuance of new coins in half. This is not a market response. It is a rule. It creates shocks, expectations, cycles. Historical data shows halvings often precede periods of heightened volatility and price appreciation, as reduced future issuance meets growing or speculative demand. In other words; shocks, expectations, and cycles. It is one of the primary engines of Bitcoin’s price history.
LLM tokens have no equivalent mechanism. Their supply is elastic. If demand rises, providers can build more data centres, acquire more GPUs, and expand capacity. Constraints like capital, supply chains, and geopolitics are real, but they are not absolute. They bend, and that changes everything.
Scarcity in Bitcoin is intrinsic. Scarcity in LLM compute is contingent – where one system resists abundance, the other moves inexorably toward it.
Out, brief candle!
And then there is the question of what is being priced. Bitcoin is held. It is accumulated, speculated on, and treated as a store of value. Like every pretty tulip, its price feeds back into its own demand, rising prices attract more buyers, which pushes prices higher. This reflexivity defines its extremes.
LLM tokens, on the other hand, strut and fret their moment upon the stage and are then consumed. They disappear the instant they are used. Nobody is hoarding tokens in the hope they will appreciate, and there is no secondary market to amplify their price through speculation. Instead, demand is anchored, for now at least, in utility – businesses pay for outcomes. Users pay for answers.
Power sits inside the pricing. Providers do not merely pass through costs, they set terms, and hey bundle services. They decide how much intelligence costs and who can afford it. The market is not perfectly competitive, but rather is shaped by a handful of actors with the capital to build and the leverage to charge. (Two decades of rapacious profiteering, following by vast anti-trust action tells you how this story ends.
The analogy to Bitcoin begins to dissolve here, not because it was wrong, but because it was incomplete.
Bitcoin teaches us that computation anchored in energy develops a floor. It teaches us that demand can lift price far above that floor, and it teaches us that efficiency does not necessarily destroy value, indeed it can expand it. But it cannot teach us how a system behaves when supply can grow, nor what happens when the thing being priced is not held, but used and gone.
So the real insight lives in the tension between these systems.
LLM tokens will likely drift toward commoditisation over time. Competition will increase. Costs will fall. Margins will compress – maybe. And yet, for as long as demand continues to surge faster than infrastructure can adapt, prices may remain elevated above their physical floor.
Not because they must, but because they can.
The question is not whether intelligence will become cheap. History suggests it will. The question is what happens in the long interval before it does – when the cost of thinking is still negotiable, still unevenly distributed, still controlled by those who own the machines.
Somewhere, a data centre is converting electricity into language. Somewhere else, a balance sheet is deciding what that language is worth.
The meter keeps ticking.
We have not yet decided what a thought should cost. The truth is, we are unlikely to be consulted.
Seraphine Vega is a human-powered Robo-reporter. She was the lead writer on this story.
Leave a Reply