OpenAI’s Project Mercury: How AI Is Reshaping Entry-Level Finance Roles

OpenAI’s secretive “Project Mercury” hires over 100 ex–investment bankers from firms like JPMorgan, Goldman Sachs, and Morgan Stanley to train its AI on Wall Street-style financial modeling.
These contractors earn about US$150 per hour to build weekly deal models—such as IPOs and restructurings—while encoding precise formatting, workflow, and quality-control norms into the system.
The initiative aims to automate junior bankers’ grunt work so human roles shift toward oversight, error-checking, and judgment rather than manual spreadsheet building.
Project Mercury reflects OpenAI’s push to monetize high-value enterprise workflows as it chases profitability despite a roughly US$10 billion revenue run-rate and a valuation above US$500 billion.

Project Mercury represents a strategic inflection point for finance and AI: OpenAI isn’t merely creating tools; it’s embedding domain expertise into foundational model training to automate rote tasks historically performed by junior investment bankers. By paying ex-bankers for high labor at high wages—US$150/hour—the company signals it values not just data, but real financial judgment, format precision, and domain habits vital to deal quality. [2] [4]

This shift is likely to accelerate a transition in what constitutes a “junior banker.” Where once entry-level hires learned via hours of manual modeling, formatting, and revision, in future they may instead act as overseers, reviewers, or validation experts who monitor AI outputs. The skills in demand will tend to be analytical judgment, ethical oversight, error detection, interpreting outputs rather than generating them from scratch. [10] [11]

From a commercial standpoint, Project Mercury aligns with OpenAI’s urgent need to monetize beyond consumer/subscription models: with an annualized revenue run-rate near US$10 billion as of mid-2025 and a valuation exceeding US$500 billion, yet with ongoing losses, initiatives that embed AI into high-billing sectors like investment banking offer high leverage. [13]

Yet the program faces several risks and open questions. Accuracy, trust, and liability issues arise when AI outputs feed into high-stakes financial transactions. Moreover, preserving institutional knowledge and capacity building is at stake; if junior roles are undercut too early, there may be gaps in training, client relationship development, and hierarchical expertise. Regulatory and ethical scrutiny is also likely as AI begins to influence markets tangibly. Finally, cost savings from automation might be offset by regulatory, compliance, or reputational costs.

Supporting Notes

OpenAI has engaged more than 100 former investment bankers to train its AI models in building financial models. [1] [2]
Recruited experts include prior employees of JPMorgan, Goldman Sachs, and Morgan Stanley. [1] [2]
Contractors are paid US$150 per hour and are expected to build financial models weekly for transaction types that include IPOs and restructurings. [2] [4]
The recruitment process features automated screening: a 20-minute AI chatbot interview based on the resume, followed by a financial statement test and a modeling challenge. [4] [6]
Tasks assigned encompass not only modeling but also maintaining formatting consistency and stylistic norms—italicizing percentages, setting margin sizes, etc.—to teach models both substance and presentation. [4] [6]
In contrast to fears of job loss, experts foresee that entry-level finance roles will evolve rather than disappear immediately, with emphasis shifting to supervision, interpreting AI outputs, and maintaining ethical, accuracy, and client-facing skills. [10] [11]
OpenAI’s revenue run-rate stood at approximately US$10 billion by June 2025, up from US$5.5 billion in December 2024; however, the company still had not become profitable. [13]
OpenAI’s valuation is reported to exceed US$500 billion, placing high expectations on delivering actionable enterprise use cases. [1] [13]

Sources

Sources

Related Reading

Leave a Comment Cancel Reply