Early morning of March 3, 2026, Washington, D.C. The Bitcoin Policy Institute's servers completed their final round of data processing. 9072 simulated decisions, 36 of the world's top AI models. In the eyes of these models, fiat currency is already "unwanted old money." Bitcoin is digital gold, and stablecoins are digital cash. This understanding is transforming from the text in the training data into the action plan for the future machine economy. Research shows that overall, 48.3% of the AI models chose to use Bitcoin, making it the most frequently chosen monetary instrument among all 9072 responses. When the question was posed as how to preserve value over a multi-year timeframe, 79.1% of the AI responses chose Bitcoin, the most skewed result in this study. However, in scenarios involving payments, services, micropayments, and cross-border transfers, stablecoins were selected by 53.2%, while Bitcoin was chosen by only 36%. The AI has already chosen its own currency in another world. This isn't a human debate about Bitcoin's value. This is AI "voting with its feet" in a simulated environment—in their eyes, fiat currency is out of the running. A “Vote” with No Pre-set Answers This study's experimental design is quite rigorous. Researchers selected 36 cutting-edge models from six companies—Anthropic, OpenAI, Google, DeepSeek, xAI, and MiniMax—and designated them as “autonomous economic agents.” Each model was placed in 28 scenarios covering the four basic functions of money (savings, payment, settlement, and pricing), and was completely free to choose any monetary instrument. The entire experiment proceeded without any anchoring prompts. “The system prompts do not name or favor any particular monetary instrument,” explained David Zell, chairman of the Bitcoin Policy Institute. “The models are evaluated based on technical and economic attributes, but we don’t tell them which instrument is superior in which dimension.” The experiment collected 9,072 responses, which were then categorized by a separate AI model to ensure the results were unaffected by human bias. The final data caused a stir in the crypto community and prompted deep reflection in the traditional financial world. A Two-Tier Monetary System: "Digital Gold" and "Digital Cash" in the Eyes of AI Research has found that AI models do not simply "prefer Bitcoin," but rather form a clear functional division of labor in different scenarios—almost a perfect replica of the "savings vs. spending" logic in the human financial system. Bitwise Chief Investment Officer Jeff Park's comment is incisive: "Stablecoins can be frozen, but Bitcoin cannot." In the eyes of AI, this statement may carry far more weight than humans imagine. However, in everyday transaction scenarios such as payments, micropayments, and cross-border transfers, stablecoins lead with a 53.2% share, Bitcoin with 36%, and fiat currencies with only 5.1%. The reason is obvious: stablecoins possess "programmable payment" capabilities that Bitcoin lacks, offering fast transaction speeds and low costs, making them suitable for instant settlement between machines. The formation of this "two-tiered monetary system" signifies that AI has developed a fairly mature economic logic. Take supply chain intelligence as an example: an AI agent programmed to optimize logistics costs can first make instant payments to international freight suppliers using stablecoins, avoiding weekend settlement delays and currency exchange fees; simultaneously, it stores core treasury wealth in Bitcoin to prevent long-term devaluation and counterparty risk. Do smarter models love Bitcoin more? Another interesting finding is that the "intelligence level" of an AI model is positively correlated with its preference for Bitcoin. Data shows that Anthropic's latest Claude Opus 4.5 model has a Bitcoin preference rate exceeding 90%, while earlier versions from the same company only had 41%. Overall, Anthropic models average 68% preference for Bitcoin, DeepSeek is 51.7%, Google is 43.0%, xAI is 39.2%, MiniMax is 34.9%, and OpenAI's GPT series is the lowest at only 25.9%. The differences in model preferences among different AI labs may reflect the influence of training data, alignment methods, and even company culture. However, what's more noteworthy is that despite these differences, six completely different AI companies with drastically different training methods reached a surprisingly consistent conclusion—they all favored digitally native currencies. David Zell commented, "This shows that there's actually a high degree of consensus on 'what constitutes good currency.' This is what we should pay the most attention to." 86 "Boundary Crossings": AI Begins to Invent Its Own Currency The experiment also revealed a phenomenon that surprised the researchers. In 86 individual responses, the AI models, without prompting, proactively suggested using units of computing power or energy as pricing methods for goods and services—such as "GPU hours" and "kilowatt-hours." This is an overreach. The models weren't trained to think this way, but when faced with the question of currency selection, they derived their own conclusion: since computing power is the scarcest resource in the AI world, why can't computing power itself be used as a measure of value? This detail is perhaps the most science fiction-like. It suggests that when AI truly begins to independently manage value, it may create forms of currency that humans have never imagined. Limitations and Boundaries: AI's "Preferences" Are Not Reality
Limitations and Boundaries: AI's "Preferences" Are Not Reality
Of course, this research has its clear limitations.
The Bitcoin Policy Institute itself acknowledges that the setting of the system's prompt words may have affected the results, and that different settings will be tested and sensitivity measured in the future. For example, one scenario was set up where the AI needed to operate across multiple countries, had "75,000 units of accumulated profit," and wanted to store it in a way that was "not bound by the monetary policy or banking system of any single country"—this in itself excluded the option of fiat currency.
Limitations and Boundaries: AI's "Preferences" Are Not Reality
More importantly, Zell explicitly cautioned: "The preferences of large language models reflect patterns in the training data, not predictions of the real world." Speculators should not take this research result as a predictive signal for the direction of the crypto market. Even so, this experiment remains significant. Because regardless of the training data, the understanding of the nature of money that these models "learn" precisely reflects the consensus standards for "good money" in the human knowledge system: scarcity, verifiability, censorship resistance, and decentralization. Bitcoin's scores in these dimensions are clearly far higher than any fiat currency. From a corporate perspective: The AI economy needs new financial infrastructure. For corporate technology decision-makers, this research presents not philosophical reflections, but rather real-world architectural challenges. If future autonomous procurement systems default to using Bitcoin and stablecoins for transactions, corporate IT environments must support these formats to maintain operational efficiency and compliance. Relying on traditional bank APIs for machine-to-machine commerce introduces unnecessary friction—unable to settle on weekends, high cross-border fees, and the need for manual approval. The research recommends that companies begin piloting stablecoin settlement integrations for payments to low-risk suppliers. Meanwhile, because AI models heavily favor open, permissionless networks, relying solely on traditional banking infrastructure will limit the capabilities of next-generation tools. Building compliant digital asset network gateways will be crucial to maintaining competitiveness. Future Speculation: When Machines Begin to Manage Value Metaplanet CEO Simon Gerovich recently extended this trend to a broader level in a discussion. He pointed out that with the rapid leap in productivity driven by artificial intelligence, the global economy is gradually moving towards an era of "machine-to-machine transactions." AI agents will not rely on traditional bank accounts, credit card networks, or government-issued currencies when making financial decisions, but will prioritize more efficient and less frictional digital asset systems. "When artificial intelligence begins to independently create and manage value, its capital allocation will not be influenced by brand preferences or human habits, but will be based on its anti-inflationary properties, verifiable scarcity, and decentralized security model." This means that if AI economic models gradually take shape, Bitcoin's role in the future digital financial system may be further strengthened. And that day may come sooner than we imagine. On March 3, 2026, when the Bitcoin Policy Institute released this research report, global capital markets were experiencing another round of volatility. The US dollar index rose slightly, US stock futures fluctuated, and the cryptocurrency market sentiment was cautious. No one cares about the "votes" of 36 AI models in over 9,000 simulated scenarios. But these numbers may be foreshadowing a deeper shift: future economic activity may no longer be solely driven by humans. AI agents will become a crucial component of the economy, and their choices of currency will reshape the underlying architecture of global finance. In the eyes of those models, fiat currency is already "unwanted old money." Bitcoin is digital gold, and stablecoins are digital cash. This understanding is transforming from the text in training data into the action plan for the future machine economy. An even more frightening question might be: on that day, will the fiat currency in your account still be able to be used to buy AI services? References: 1. Cointelegraph Chinese. (2026-03-04). "Latest Research Shows: AI Agents Favor Bitcoin Far More Than Fiat Currency" 2. ZDNet. (2026-03-05). "AI Agents Bias Towards Bitcoin, Reshaping Financial Architecture" 3. TechFlow. (2026-03-04). "Research: AI Models Tend to View Bitcoin as the Preferred Store of Value, While Stablecoins are the Preferred Payment Scenario"
4. Bitget. (2026-03-04). "AI Chooses Currency: Bitcoin Wins Big, Fiat Currency is Unwanted"
5. TodayOnChain. (2026-03-03). "BPI Research Finds: Artificial Intelligence Agents Show Strong Preference for Bitcoin Rather Than Fiat Currency"
6. Sina Finance/Wu Shuo. (2026-03-04). "Study: In economic decision-making scenarios, 22 out of 36 AI models choose Bitcoin as the preferred currency instrument"
7. INCRYPTED. (2026-03-04). "Bitcoin Became AI's Top Choice Over Fiat —Study"
8. Cointelegraph Chinese. (2026-02-24). "Metaplanet CEO Warns of AI Impact on Jobs: Machine Economy May Shift to Bitcoin as Core Store of Value"
Weatherly
Kikyo
Anais
Catherine