Saylor Backs Bitcoin (BTC) as 75.8% of AI Models Pick It for Savings

Strategy executive chairman Michael Saylor has put fresh research behind his long-running case for Bitcoin (BTC), answering the Bitcoin Policy Institute's quarterly update on its “Money for AI” project with a four-word post: “Smart money saves in Bitcoin.” The reply went out on X on October 15, a message responding to the study update. Whatever the Bitcoin (BTC) price does this week, Saylor's line is not about the chart. It is a claim about the role Bitcoin might play in an economy that is increasingly run by software. The framing is textbook Bitcoin Maximalism, the position that Bitcoin is the terminal form of savings technology, and this time it arrives with numbers attached. The project the post refers to asks AI models to choose monetary tools as if they acted as autonomous economic agents rather than systems guided by human operators. In its first round, published in March, 36 models worked through 9,072 scenarios involving money. Bitcoin took 79.1% of responses on the question of preserving wealth over the long term, ahead of stablecoins, fiat money and other cryptocurrencies, and it ranked first across all four tested functions with 48.3% of all answers. The same round exposed the limits of that lead: stablecoins won the payments question outright at 53.2%, against 36% for Bitcoin (BTC). Saylor's corporate record gives the words weight beyond rhetoric. Strategy has built the largest corporate Bitcoin treasury in the market, a conviction position held through drawdowns in the style usually described as HODL, and the executive has argued repeatedly that companies and even nations should hold the asset as a Strategic Bitcoin Reserve rather than trade it. The reply matters less as market news than as positioning: a prominent corporate holder is citing research to argue that future demand for Bitcoin may come from machines, not only from treasuries and ETF products.

@bitcoinpolicy · X post

A message responding to the study update.

View on X

Nine Models, 2,268 Responses

The newest wave of the research supplies the detail behind Saylor's number. The institute ran a follow-up between September 1 and September 3, putting the same four tasks to nine additional AI models: saving for the future, paying, pricing and account keeping, and transaction settlement. The exercise collected 2,268 responses in total. Of the 567 answers filed under the value-storage function, 75.8% picked Bitcoin (BTC), the figure Saylor's post leans on. The aggregate picture is more mixed. Across all four roles combined, stablecoins led with 42.7% of responses against 34.0% for Bitcoin. On the unit-of-account function, the standard used to price goods and services, fiat and bank money ranked first at 41.4%, with stablecoins second at 35.8%. The institute said it reused the scenarios and classification method from the original study, but left temperature and seed values unspecified, so every model answered on its provider's default settings. The researchers also drew their own boundaries around the results. The two study cohorts differ in ways that make it impossible to isolate any change across model generations from other factors, and the percentages capture choices made in hypothetical situations rather than measured economic behavior. No model in the test bought, held or spent anything. The pattern itself repeated what came before: Bitcoin first on savings in both rounds, stablecoins first on payments in both, which the researchers read as evidence that the models can tell the difference between an asset meant to preserve value and an asset meant to move it. The savings rate alone should not be stretched into a claim that AI in general prefers Bitcoin, since payments, accounting and settlement answers pulled in other directions.

A Savings Vote, Not a Payments Vote

The split inside the numbers is what Saylor's four words compress. If autonomous agents ever manage real funds, the data points to Bitcoin for reserves and stablecoins for circulation, a division that mirrors how human holders already behave. That reading is ours, and the institute's own caveats bound it: the exercise recorded stated choices in hypothetical situations, with no wallets, no purchases and no holdings involved. The wider question of machine economies and the asset is the ground covered in the Galaxy Digital Podcast Weighs Bitcoin (BTC), AI and Tokenization Over the Next 10 Years. Whether the 75.8% savings preference would survive contact with real balances is the open question the research does not settle, and the institute has announced no next round.