What if the next reserve currency isn't mined from a blockchain — it's burned every time someone asks an AI a question?

AI Tokens as Currency: Is Claude the New Bitcoin?

Someone sent me a note this week that stopped me mid-scroll. The idea: Bitcoin is old crypto. The new crypto will be Claude tokens and ChatGPT tokens — and if AI becomes the infrastructure everything runs on, those tokens become the exchange currency of the future.

I've spent enough time at the intersection of operations, product, and early-stage investing to know when an idea is worth taking seriously versus when it just sounds good at a dinner party. This one deserves a real look.

TL;DR: The analogy between AI tokens and cryptocurrency is more structurally sound than it first appears — but it breaks down in a few critical places that matter for anyone thinking about it as an investment thesis. There's a real insight buried here about demand, scarcity, and infrastructure. There's also a category error that could cost you money if you don't see it clearly. Here's the full picture.

Nothing in this article is financial advice. These are my personal observations and opinions — think of it as me thinking out loud, not telling you what to do with your money.

Why This Matters

I've been building with AI tools seriously for the last two years — integrating them into Fleamio, using them daily for research, prototyping, and decision-making. What I've noticed is that the unit of consumption isn't hours or seats. It's tokens. Every prompt, every response, every background process — it burns tokens. And tokens cost compute, which costs money, which means they have real economic weight.

When I ran logistics operations at Foodora — scaling from a handful of cities to 20+ across Austria — the scarce resource was always the bottleneck that determined price. Riders in peak hours. Restaurant capacity on Friday nights. The insight from that experience: when demand scales faster than infrastructure, the price of the constraint goes up. Always.

That's the real thesis here. Not that Claude tokens are Bitcoin. But that AI compute capacity might behave like a scarce commodity during a demand spike the world hasn't seen yet.

What Are AI Tokens and Do They Have Real Economic Value?

First, a technical grounding — because the word "token" is doing a lot of work in two very different industries.

In crypto, a token is a unit on a blockchain — scarce by design, transferable peer-to-peer, and value is driven by speculation, utility, or both.

In AI, a token is a chunk of text — roughly 0.75 words in English. When you send a message to Claude or GPT-4, the model processes your input tokens and generates output tokens. You pay per token via API. As of mid-2025, GPT-4o costs around $5 per million input tokens. Claude 3.5 Sonnet sits at a similar range. These aren't blockchain tokens — they're billing units for compute.

Here's where the analogy gets interesting though. If AI becomes the operating layer of most software — coding, customer service, logistics, legal, medical — then token consumption becomes the economic heartbeat of the digital economy. Not metaphorically. Literally. Every automated decision, every generated document, every agent running in the background burns tokens. Demand becomes structural. The structural difference: Bitcoin tokens exist. AI tokens are consumed. One is a store. One is a flow.

Is There a Real Investment Thesis Here?

Let's stress-test the demand side first — because that's where the original idea has teeth.

Global AI compute demand is scaling faster than GPU production. NVIDIA's H100 chips have had multi-month waitlists. Microsoft, Google, Amazon, and Meta have each announced $50B+ in data centre capex for 2024-2025. The IEA estimates AI data centres could consume 1,000 TWh of electricity annually by 2026 — roughly the entire current consumption of Japan.

If demand for AI inference grows 10x over five years (conservative given current trajectory), and physical infrastructure can't keep pace, then the price of access to that compute goes up. That's not speculation. That's supply and demand.

So the question becomes: how do you think about holding exposure to that? I'll share how I'm framing it — but again, this is my own thinking, not a recommendation. Do your own research before doing anything with your money.

NVIDIA (NVDA) — the picks-and-shovels framing. You're betting on compute infrastructure.
Anthropic, OpenAI — private, not accessible to most retail investors yet.
Actual AI-linked crypto tokens — projects like Render Network (RNDR), Akash Network (AKT), or Bittensor (TAO) are genuinely trying to build decentralised AI compute markets. These are the closest real-world version of the idea in the original thesis.
The AI labs themselves going public — if OpenAI IPOs, that's the closest thing to buying the "new Bitcoin" the idea describes.

What Most People Get Wrong About This Idea

Here's the category error I mentioned at the top.

Bitcoin's value comes partly from what it isn't — it's not controlled by anyone. No company, no government. That decentralisation is the feature, not a bug.

Claude tokens and ChatGPT tokens are the opposite. They're controlled by private companies with investors, revenue targets, and the ability to change pricing at any time. Anthropic can reprice their API tomorrow. OpenAI already has. That's not a currency — that's a SaaS subscription with extra steps.

For something to function as an exchange currency, it needs neutrality and transferability. AI tokens have neither. You can't pay your landlord in Claude tokens. You can't send them to a wallet. They expire when you use them.

The deeper insight — and this is what I think the original idea is really pointing at — is that access to AI compute might become one of the most valuable economic inputs of the next decade. Like oil in the 20th century. The countries and companies that control it will have structural advantages. That's a real geopolitical and economic thesis. It just isn't best expressed as "tokens are the new Bitcoin."

I've written more about how infrastructure scarcity shapes market dynamics in my notes on longevity investing — the pattern is similar: when the bottleneck is physical and the demand is existential, price floors are sticky.

Bitcoin solved the problem of who to trust with money. AI tokens might solve the problem of who to trust with thinking.

What to Actually Do

Don't buy "AI tokens" expecting them to behave like crypto. They're usage-based billing units. They don't appreciate. They don't transfer. They're not an asset class.
Do look at decentralised AI compute networks — Render, Akash, Bittensor. These are genuine attempts to build a crypto-native AI infrastructure market. High risk, real thesis, volatile. Not a recommendation — just where the actual structural bet lives if this thesis interests you.
Do take the compute scarcity thesis seriously as a macro view. NVIDIA is the clearest expression of it in public markets, as is exposure to data centre REITs or energy infrastructure plays tied to AI demand. None of this is me telling you to buy anything — I'm mapping the landscape as I see it.
Watch for OpenAI or Anthropic IPO signals. That's when the "new crypto" thesis becomes more mainstream-investable — and early positioning in adjacent public companies may matter. Your call on whether and how to act on that.
If you're building anything — get serious about your token costs now, before pricing normalises upward. I'm already tracking token burn in Fleamio's cost model the way I tracked cost-per-delivery at Foodora.
Think in analogies but invest in specifics. The oil analogy for AI compute is useful for framing. It's not useful for picking tickers.

The idea isn't wrong. It's just early, and it's pointing at the right thing with slightly the wrong vocabulary.

One more time for clarity: nothing here is financial advice. I'm not a financial advisor. These are my own views, shaped by my experience in operations and product — not by any professional investment qualification. Do your own research.