When Every Prompt Has a Price

Intelligence used to feel free. The infrastructure under it never was.

For most of the digital era, intelligence felt abundant. A search cost almost nothing. An email was free. Copying text or code could be done endlessly using a micro amount of physical resource. The internet taught a generation that information was also free and lived outside economics.

AI ends that illusion.

Every prompt sent to a large language model starts a chain of physical events. Data centers allocate compute. Processors draw power. Cooling systems dump heat. Networks haul the result across continents. The experience stays digital. The infrastructure underneath is industrial.

That changes the economics of intelligence. And it raises a question most businesses have not yet asked: should the consumption of AI sit inside the tax system?

The idea sounds strange at first. Tax has always followed income, profit, property, and physical resources. But every major technological shift forces a society to rethink what it measures, what it prices, and how it shares the gains. AI has reached that point.

The Proposal: Tax the Tokens

One idea is gaining traction – a tax on AI token consumption. The principle is plain. The more AI you use, the more you contribute. The gathered money could fund the thing the technology disrupts: retraining, education, support for workers whose roles change due to AI.

Linas Būtėnas, Co-Founder and EVP of Innovation & R&D at SH Group, frames it beyond revenue.

"The more AI you use, the more you would contribute," he says. "Those resources could help finance career transitions, retraining programs, and support systems for people affected by the shift."

The logic tracks history. Every transition built its own adjustment mechanism. The industrial revolution forced labor protections. Globalization funded regional development and retraining. AI will need its own. The disruption is not hypothetical: the World Economic Forum's Future of Jobs Report 2025 projects that the combined forces reshaping work this decade — technology and AI, the green transition, demographics, economic pressure — will create 170 million roles, displace 92 million, and leave 59 of every 100 workers needing reskilling or upskilling by 2030. AI is the largest single technology driver in that mix: the report attributes around 11 million new roles and 9 million displaced to AI and information processing — more than any other technology trend.

AI is an Infrastructure Story Now

The argument sharpens through the lens of infrastructure.

Today's AI runs on electricity grids, land, water, cooling, and data centers – resources built to serve whole economies, not one industry.  AI scaling consumes physical resources every time it runs, and that consumption shows up in someone's grid and someone's water supply.

The IEA expects data-center electricity demand to more than double from 2024 levels to around 945 TWh by 2030 — roughly Japan's entire electricity consumption – with AI-focused demand tripling. US data centers alone account for nearly half of the country's electricity-demand growth this decade. The water follows the power: US data centers directly consumed 17.4 billion gallons in 2023, a figure projected to rise to between 38 and 73 billion by 2028.

This is no longer a software story. To feed future compute, Microsoft signed a 20-year power deal that will bring a retired reactor at Three Mile Island back online — the Pennsylvania plant whose adjacent reactor caused America's worst nuclear accident in 1979 — and by 2026 every major hyperscaler had committed nuclear capacity to AI. Governments are examining what large data centers do to local energy and water systems. The constraint on AI is shifting from algorithms to amperes.

That reframes where value should sit. As Linas Būtėnas says:

"AI infrastructure uses resources that already belong to communities. Part of the value created by that infrastructure could reasonably return to the places where it operates."

Which opens the hardest question in the debate. Not whether to tax – but where and how.

Where Does an AI Tax Even Land?

Natural resources are straightforward to tax because they sit in one place. AI is global; its infrastructure is local. A data center draws on one country's energy, one country's grid, one country's communities – regardless of where its users sit.

So where is the tax paid? Where the model was trained? Where the company is headquartered? Where the user is? Or where the machines physically run and the heat comes out?

A fair answer is hard to design. A token from a lightweight local model and a token from a frontier system can represent wildly different amounts of computation. An enterprise deployment burns through resource an individual never touches. Measuring usage is easy. Measuring impact is not.

But difficulty has never stopped an economy forming around a resource that matters.

We Have Watched This from the Inside

This debate is not abstract for us. SH Group has run AI systems since 2018 – alongside our distributed-systems and cryptography work, not as a recent pivot. We built production AI for an EU-funded R&D programme, and we have specified AI systems for government-grade infrastructure: environments where the question of who carries the cost and who carries the risk is not a thought experiment.

That vantage point shapes how we read the token-tax argument. We build sovereign-grade infrastructure for a living, and one principle runs through all of it – technology should empower the people and places it touches, not extract from them. Sovereignty over dependency. An AI economy that takes energy, land, and water from a community and returns nothing fails that test, the same way a financial system that hides its provenance fails it.

We have made the opposite choice before. LBCOIN put 24,000 cryptographically verifiable tokens on the Bank of Lithuania's live systems – value that stayed accountable to the institution that issued it. Axiology runs today as a MiFID-licensed DLT venue under the EU Pilot Regime, where every transaction's provenance is verifiable from the architecture down.

The lesson carries straight into AI: where value is created should be visible and accountable by design, not asserted after the fact.

The Real Signal

The most revealing thing about this debate is that it exists at all.

Linas Būtėnas, Co-Founder and EVP of Innovation & R&D at SH Group

For decades, intelligence held a strange place in the economy. It created enormous value and was never measured as a resource. Human expertise could not be counted in units. Creativity could not be metered.

AI changes this equation. For the first time, intelligence can be quantified, priced, scaled, and optimized through infrastructure. Tokens are a unit of consumption. Compute is a strategic resource. Energy availability now sets the ceiling on ambition.

Whether governments introduce token taxes is not really the point. The point is that the conversation has moved – past models and capabilities, into energy, infrastructure, economics, and public responsibility.

For founders, this has a near-term edge: the cost of intelligence is becoming a line item, and the architecture you choose now decides whether you own that cost or rent it. For institutions, it is a governance question your board will raise before the regulator does.

Every technological revolution eventually reveals the resources that make it possible.

AI is now revealing its own.