ChatGPT played an unusual role in MicroStrategy’s 2025 fundraising: it helped Michael Saylor work through the design of a new financial instrument. The result, STRK, formed part of a capital raise of about $15 billion that also included related securities. The case shows AI being used beyond drafting and research, while drawing a distinction between generating a promising structure and proving that it can work in a real market.
The problem was a funding limit
In early 2025, MicroStrategy held roughly $30 billion in Bitcoin and wanted to raise more capital to buy more. It had already relied heavily on conventional equity and convertible bonds, leaving limited room to keep using those routes. Saylor, the company’s executive chairman, turned to OpenAI’s ChatGPT to explore another structure.
▲ Constraints behind a financial design
The objective was specific: design a convertible preferred stock that could trade steadily near a $100 par value, or stated face amount. Preferred stock is a class of shares with dividend terms; a conversion feature gives holders a way to exchange it under specified conditions. STRK was intended to combine those characteristics with a price profile closer to a stable, cash-like instrument.
That was not a generic request for fundraising ideas. The existing financing limits, desired trading price and terms available to investors all shaped the problem. Banking and securities-law advisers initially questioned the concept because they saw no established precedent for that combination of features.
What ChatGPT helped design
Saylor used iterative prompts—successive instructions that refine an AI model’s answer—to develop the proposed terms. A central feature was a variable dividend rate, adjusted monthly with the aim of keeping STRK’s market price near $100. The goal mattered as much as the mechanism: a dividend adjustment was meant to address how the security might trade, not merely how its terms looked on paper.
MicroStrategy brought the Bitcoin-backed convertible preferred stock to market through an initial public offering, the first public sale of that security. It then used a shelf registration, a filing that permits later securities sales, for further issuance. The amounts in the case break down as follows:
| Capital raised | Amount |
|---|---|
| Initial STRK offering | $2.5 billion |
| Subsequent STRK sales | $8 billion |
| Related securities | $4 billion |
| Total, rounded | About $15 billion |
The first two lines amount to $10.5 billion of STRK. Adding the $4 billion from related securities gives $14.5 billion, described in rounded terms as a roughly $15 billion raise. That distinction matters: the headline total was not raised through STRK alone, and it should not be read as a result another company could expect from using AI.
A design partner, not a final authority
The practical method starts with constraints. Saylor emphasized giving AI a defined state space—the relevant conditions and possible choices—rather than asking an open-ended question. In this case, those conditions included funding channels that had become difficult to expand, the $100 price objective and the need for terms that could function as a public-market security.
This approach changes what a useful answer looks like. Instead of asking ChatGPT to name a novel product, a team can ask it to work through how proposed terms serve a stated goal, then probe where those terms conflict with other requirements. Saylor’s iterative work on the dividend mechanism illustrates that process. His knowledge of the financing problem supplied the context; the model helped explore a structure within it.
The boundary is equally important. An AI-generated proposal is not legal approval, evidence of investor demand or proof that a target trading price will hold. Financial and legal specialists still need to examine proposed terms, and a market launch is a separate undertaking. STRK shows what can happen when AI contributes to design; it does not show that prompting alone can issue a security.
Why this use of AI may extend beyond finance
Consumer survey data put the share of households paying for AI subscriptions at about 2%. That figure suggests considerable room for wider use, though it says nothing about whether any particular AI-supported idea will succeed. The distinctive feature of this case is the combination of detailed domain knowledge and a question whose answer was not already sitting in a familiar template.
Saylor frames that opportunity through a technology S-curve: a period of rapid improvement that eventually levels off. He contrasts the swift progress from early powered flight in 1903 to the Moon landing in 1969 with his assessment that commercial jet efficiency improved by only about 15% over the following 50 years. Computing, in his view, has remained on a steeper path.
▲ Different stages of technological growth
That comparison is best used as a question, not a prediction. Which parts of a workflow have become routine, and where does an unresolved problem remain? In the STRK case, the opportunity was not to produce a faster version of an existing document. It was to consider a different financing structure under tight constraints.
What to take into your next AI project
- State the real constraint. Identify the goal, the approaches already tried and the conditions a workable answer must satisfy.
- Ask for a mechanism, not just an idea. Examine how each proposed feature is supposed to produce the desired result.
- Challenge the proposal with domain specialists. Treat AI output as a design to test, particularly where legal and market requirements matter.
- Keep the outcome precise. Separate what the AI helped design from what people subsequently reviewed, issued or sold.
MicroStrategy’s raise illustrates a demanding use of ChatGPT: working on a financial product when familiar options had narrowed. Its useful lesson is to pair precise prompts with domain judgment—and to keep testing the answer after the model responds.