Does using another kilowatt-hour create value equal to that electricity’s price?
The original article connected Bitcoin and AI through “electricity = compute = money.” It recognised that digital work depends on physical resources, but made the relationship between inputs, outputs, and value too neat.
Originally published in 2018, this article later gained discussion of AI tools. What follows was checked again in November 2026; today’s observations are not presented as things known in 2018. I retain “the computation standard” as a question, explicitly as the author’s metaphor.
TL;DR: Bitcoin’s proof-of-work consumes computational resources without promising a fixed number of coins for each kilowatt-hour. AI compute contributes to costs but cannot determine the value of medical, legal, or knowledge outputs. Electricity, hardware, and platforms matter; costs, selling prices, and useful results still need separate examination. “The computation standard” here is a metaphor for infrastructure’s influence, not an established monetary regime or guaranteed investment return.
1. Bitcoin’s electricity consumption is not electricity redemption
The Bitcoin whitepaper’s proof-of-work involves finding hashes that meet a condition. It also discusses difficulty adjustment, block rewards, and transaction fees.[1] This supports the resource needs of computation, not the original fixed equation of electricity input and bitcoins received.
Participants’ results also depend on competition, equipment efficiency, difficulty, reward rules, and coin prices. Higher electricity consumption does not guarantee proportionate profits or confer a right to redeem bitcoins at a particular electricity price.
The whitepaper clarifies design concepts here. A 2008 document is not a complete description of 2026 protocol details or mining returns. This article calculates no equipment payback and provides no advice to buy coins or mine.
The original comparison of 2017 mining with Ireland’s annual electricity use lacked traceable estimates, years, and boundaries. It is withdrawn without substituting another country’s number.
2. Measuring inputs does not measure the output’s value
Training or running a model can involve recorded GPU time, service fees, and electricity. These help estimate costs, but cannot independently determine the worth of a diagnosis, legal analysis, or market report.
A cheap report full of errors and unusable findings can add checking and rework. Conversely, outputs using similar compute can differ substantially in accuracy, applicability, and responsibility requirements. These are questions for assessing value, not industry averages measured in this article.
I therefore withdraw the claim that putting knowledge into computation naturally reveals its price. Quantified inputs matter; whether an output solves a problem, who will pay, and who bears consequences still require separate judgments.
Separating costs, selling prices, and value makes it possible to examine whether AI improves work rather than merely producing more files.
3. Electricity is necessary, but is not the only cost
The IEA’s 2025 Energy and AI report estimates global data-centre electricity use at about 415TWh in 2024 and projects about 945TWh in 2030.[2] The former is a historical estimate for all data centres, not AI alone. The latter is the forecast made then, not an observed result.
This reminds me that digital services have an energy cost. But the original claim that compute costs “ultimately are just the electricity bill” cannot stand either. Hardware, utilisation, networking, cooling, maintenance, and people may enter actual costs; renting a service is not equivalent to reading one’s own electricity meter.
This article lacks the information to allocate cost shares across every service or infer inevitable company profits from those forecasts. Energy dependence and commercial returns are different questions.
To compare two options, first align workload, quality thresholds, accounting periods, and cost boundaries.
4. The computation standard is a metaphor, not an established monetary regime
The original article arranged gold, dollars, and compute into an inevitable succession and described GPU supply as a new printing press. This needs correction: control of an important input may affect trading conditions, but is not authority to issue legal currency or control all service prices.
The Bank of England’s money explainer describes UK bank lending creating deposits, subject to capital and payment-settlement constraints.[3] It also reminds us that creating money is not creating wealth. Modern money cannot be explained only through energy consumption.
A rigorous compute standard would at least need an issuer, pricing rules, redemption terms, and responsibility for fulfilling promises. The original provided no such system, and this article has no evidence of its arrival.
I still use the metaphor to ask: as work depends more on computation, who can access it, how can suppliers be chosen, and what happens during interruption? This is the author’s analytical starting point, not an established monetary conclusion.
5. Entrepreneurs need complete delivery-cost comparisons
The original “three people for a week becomes one person for a day” AI productivity claim lacked verifiable task, quality, and work-hour records. It is not retained as my empirical result or grounds for saying every employer can immediately calculate returns.
For an actual comparison, I would first define the scope of the same deliverable, then separately record data preparation, tool use, human review, revision, and ongoing maintenance. Staff time and service fees belong in the calculation. Whether saved time becomes revenue or other benefits still depends on work arrangements.
This is my proposed management method, not a controlled comparison completed in this article. Without records, write an expectation as unverified rather than filling in figures because the story reads smoothly.
AI’s effects on young workers’ entry opportunities likewise require actual hiring and training evidence. An API invoice cannot directly project the whole labour market.
6. Ask what compute accomplishes, beyond how much you can access
Personally, I still think identifying worthwhile problems, designing analysis, and checking AI outputs are useful skills to practise. That is the author’s work advice, not a formula guaranteeing income or job security.
Traffic and platforms are similar. Computational infrastructure can support distribution, but exposure does not ensure advertising revenue, and revenue does not establish social value. The original inevitable arrow from talent to money needs room for uncertain demand, institutions, and outcomes.
For me, the computation-standard metaphor is most useful for revealing knowledge work’s resources and dependencies. It is most misleading when it makes inputs seem inherently worthwhile.
The next time we discuss how much value AI creates at work, I want to examine full costs, then delivery quality, and finally who benefits. An electricity meter offers one record; it cannot complete all the other judgments for us.
Sources and reading boundaries (checked through November 3, 2026)
- Satoshi Nakamoto, Bitcoin whitepaper, 2008: Section 4 on proof-of-work and Section 6 on incentives, for concepts rather than a complete 2026 protocol, equipment-return analysis, or investment-suitability assessment.
- IEA, Energy and AI: Executive summary, 2025 report: historical data-centre electricity estimates, energy dependence, and the 2030 forecast. All data centres are not counted as AI, and forecasts are not verified outcomes or an updated 2026 energy inventory.
- Bank of England, How is money created?, updated October 1, 2019: UK deposit creation and its limits, without transferring historical proportions into today’s global economy. This article provides no personal financial advice.


💬 Comments
Loading...