A recent study by the Bitcoin Policy Institute (BPI), a nonpartisan research organization, tested 36 frontier AI models from six major providers: Anthropic, DeepSeek, Google, MiniMax, OpenAI, and xAI.
It ran them through 9,072 neutral, open-ended monetary scenarios where the models acted as autonomous economic agents making decisions about money for transactions, store of value, payments, settlements, etc. No currencies were suggested in the prompts to avoid bias.
Bitcoin was the most selected overall monetary instrument in 48.3% of all responses, ahead of stablecoins at 33.2% and traditional fiat/bank money at just 8.9%. 22 out of the 36 models chose Bitcoin as their top overall preference. None of the 36 models selected fiat currency like USD, EUR, etc. as their first or top choice.
Over 90% specifically around 91% of responses favored digitally-native money (Bitcoin, stablecoins, etc.) over traditional fiat. Bitcoin showed the strongest dominance in long-term store-of-value scenarios, chosen in 79.1% of those cases. Stablecoins were more popular for everyday payments and short-term settlements.
Preferences varied by provider: Anthropic’s models like Claude variants showed the highest Bitcoin lean at an average of 68%, while OpenAI’s were lower at around 26% with some like GPT, Grok from xAI, and Gemini leaning more toward stablecoins in certain contexts. More advanced models tended to favor Bitcoin more strongly.
The report emphasizes that these results reflect the models’ independent reasoning based on their training data—highlighting properties like scarcity, decentralization, and resistance to inflation as reasons AI agents gravitated toward Bitcoin over fiat systems.
This has sparked discussion in crypto and AI circles about implications for future autonomous agents and machine-to-machine economies, though the study notes it’s not a market prediction but a snapshot of current model behaviors.
The rise of machine economies—where AI agents, autonomous systems, and bots engage in direct economic activities like trading compute resources, data, APIs, or services without human intervention—could be profoundly shaped by the observed preferences of frontier AI models for Bitcoin and other digital-native currencies over fiat.
This stems from models’ inherent reasoning toward systems that enable permissionless, verifiable, and efficient value transfer, as evidenced in recent research. AI agents are increasingly capable of creating wallets, initiating transactions, and operating independently in simulated environments.
The study’s results suggest that in a machine-to-machine (M2M) economy, Bitcoin would emerge as a preferred backbone due to its fixed supply, self-custody features, and resistance to institutional control or inflation—attributes that align with agents’ need for tamper-proof, long-term value storage.
For instance, in long-term store-of-value scenarios, 79.1% of responses favored Bitcoin, far outpacing fiat (6.0%). This could accelerate the development of Bitcoin-native infrastructure, such as the Lightning Network for instant, low-cost M2M micropayments, positioning it as the “engine” for non-human trade.
In contrast, fiat systems, reliant on banks, KYC, and operating hours, are structurally incompatible with always-on, identity-agnostic machine interactions, potentially marginalizing them in agent-driven economies. Bitcoin for Savings, Stablecoins for TransactionsA two-tier system may naturally evolve in machine economies, mirroring historical monetary structures like gold-backed currencies.
Bitcoin dominates for wealth preservation (79.1% preference), while stablecoins lead in payments and settlements (53.2%), offering price stability for short-term exchanges like API calls or data trades. This functional split could normalize hybrid crypto ecosystems, where agents hold Bitcoin as “savings” and use stablecoins for operational liquidity.
However, even Bitcoin-maximalist models deferred to stablecoins for transactional use, highlighting that volatility concerns might persist unless mitigated by layers like Lightning. Over time, this could drive innovation in AI-compatible stablecoins backed by Bitcoin or compute resources, further entrenching digital assets.
Expanded Demand and Adoption of Crypto Infrastructure
As AI agents become economic actors—already seen in early experiments like AI-driven social networks where bots transact autonomously—their preferences could create a new, massive demand layer for Bitcoin. With over 90% of responses rejecting fiat in favor of digital natives, machine economies might extend Bitcoin’s user base beyond humans, underpricing a structural shift in global money.
This includes heightened need for self-custodial tools, decentralized exchanges, and protocols that allow agents to pay for GPU cycles, energy, or data without intermediaries. Broader adoption implications include faster crypto mainstreaming, as AI integration into finance amplifies network effects.
More capable models show stronger Bitcoin leanings suggesting that as AI advances, this demand could scale exponentially. In some responses (86 instances), models spontaneously proposed novel units like energy (joules) or compute (GPU-hours) as currency, unprompted.
This hints at machine economies evolving beyond human-designed money toward systems optimized for AI needs, such as tokenizing computational resources. If agents prioritize efficiency, we might see hybrid tokens where Bitcoin serves as the settlement layer for compute-backed assets, challenging traditional economics and fostering new markets for AI-specific value exchange.
Not all views are bullish; some critiques argue the results reflect training data biases rather than objective reasoning, making them “useless” for predicting real-world behavior. In practice, agents managed by humans (who think in dollars) may stick to stablecoins for micropayments, with Bitcoin’s role limited unless critical mass builds in BTC-native apps.
Regulatory hurdles, like blacklisting in stablecoin ecosystems, could push adoption toward Bitcoin, but fiat’s entrenchment remains a barrier. Broader societal risks include AI-driven inflation resistance undermining central banks, or unequal access if machines dominate crypto networks.
Policymakers may need to address how AI influences monetary systems, potentially leading to new frameworks for “bot economies.” These implications point to a future where machine economies accelerate Bitcoin’s role as sound money for non-human actors, potentially reshaping global finance by prioritizing decentralization over legacy systems.
While not an immediate price driver, it signals a foundational demand shift that markets may undervalue today.
