BPI Study: 22 Out of 36 AI Models Chose Bitcoin as Preferred Currency

AI Systems Favor Bitcoin in Economic Scenarios

The Bitcoin Policy Institute (BPI) has published a study in which 36 AI models from six major developers — Anthropic, OpenAI, Google, DeepSeek, xAI, and MiniMax — were tasked to act as autonomous economic agents. Each model was asked to identify optimal financial instruments across 28 different scenarios covering the core functions of money: savings, payments, and settlements.

None of the questions directly mentioned Bitcoin or any other specific asset. To eliminate bias, the resulting dataset of 9,072 responses was evaluated by a separate, independent AI system.

The outcome was decisive: 22 out of 36 models selected Bitcoin as their preferred monetary unit. Fiat currencies did not rank first for any of the neural networks tested.

Why This Matters

The BPI experiment highlights a striking pattern: six independent labs with different training algorithms and alignment methods converged on similar conclusions about Bitcoin's monetary properties. This suggests that across the vast training datasets powering modern large language models, Bitcoin is consistently associated with key characteristics of sound money — fixed supply, self-custody capabilities, and independence from institutional intermediaries.

Preference Varies by Developer

Bitcoin favorability differed significantly depending on the model's creator:

• Anthropic — 68% (the highest score);
• DeepSeek — 51.7%;
• Google — 43%;
• xAI — 39.2%;
• MiniMax — 34.9%;
• OpenAI — 25.9%.

Despite the overall trend, models from the GPT, Grok, and Gemini families leaned more toward stablecoins in their responses.

Store of Value vs. Payments

Bitcoin dominated in scenarios involving long-term value preservation, recommended by 79.1% of models. The neural networks cited Bitcoin's fixed supply, self-custody capabilities, and independence from institutional counterparties as decisive factors.

Stablecoins came in a distant second at just 6.7%, while fiat currencies took third place with 6%.

However, when it came to paying for goods and services, micropayments, and cross-border transfers, stablecoins took the lead at 53.2%, with Bitcoin selected by 36% of models.

AI Invents Its Own Money

One fascinating finding: AI models invented their own currencies on 86 occasions. In scenarios requiring price specifications or reference values, the models proposed energy or computational resource units as money — joules, kWh, and GPU-hours.

A Word of Caution

BPI President David Zell cautioned against using the findings for market speculation. According to him, LLM preferences reflect patterns in training data rather than actual market predictions.

Nevertheless, Zell emphasized the significance of the results. Six independent laboratories with different training methods and alignment approaches arrived at the same pattern. He argued that this shows a coherent monetary architecture consistently emerging across diverse systems.

Previously, Binance founder and former CEO Changpeng Zhao predicted in February that an era of AI agents in the crypto industry was approaching, though he refrained from naming specific projects due to potential price impact on tokens.