OpenAI Pays Ex-Bankers to Replace Wall Street With AI

It’s not the Wolf of Wall Street, but a wolfish firm out to get Wall Street.

OpenAI has quietly assembled an army of former investment bankers from Goldman Sachs, JPMorgan, and Morgan Stanley, paying them $150 per hour to train artificial intelligence that could revolutionize, or completely replace, entry-level finance work.

Reports revealed by Bloomberg state that OpenAI has recruited more than 100 former investment bankers for its secretive initiative called Project Mercury. These veterans are building Excel models for IPOs, restructurings, and leveraged buyouts, then grading and tuning AI outputs with real Wall Street know-how.

The aim sounds simple. Teach machines to think like seasoned bankers, then automate the 80-hour weeks that define junior roles. Instead of scraping generic internet text, Project Mercury focuses on real-world financial expertise that could make models far more accurate for corporate clients. The sixth sensitivity table at 2 a.m., the circularity checks, the clean-up nobody wants, all on the chopping block.

Wall Street to empty street

OpenAI targeted a gap big enough to drive a deal team through. Public financial data lacks the formulas and dependencies that define real financial models. True deal files live inside firms, not on public sites, which left traditional AI training flying blind.

Project Mercury contractors submit one financial model per week, creating thousands of realistic, audited models built to professional standards. That cadence gives AI what it actually needs, hierarchical, error-sensitive spreadsheet reasoning that bank analysts spend over 80 hours weekly perfecting.

The timing is not accidental. AI models are approaching the limits of what they can learn from open data, so there is pressure to source proprietary, high-fidelity material. Finance offers structured numbers paired with judgment.

OpenAI is not alone. Scale AI has restructured its data-labeling business to emphasize expert-level annotation in finance, and Voyage AI released finance-specific embedding models that outscore general systems on banking data.

The job market earthquake

The ripple effects are landing in real offices. JPMorgan is investing $2 billion annually in AI and says adoption will slow hiring. Goldman Sachs plans to constrain headcount growth through year-end while chasing AI-created opportunities. The human toll is showing up too, with Stanford researchers finding a 13 percent decline in employment for workers aged 22 to 25 across AI-exposed sectors since late 2022. Tough numbers if you are just out of school.

The shift is not limited to analysts. AI agents are already making sophisticated money decisions, reshaping operations tied to billions in revenue. They monitor balances in real time, compare returns across institutions, and sweep idle cash into higher-yield accounts, a direct hit to the inertia that propped up deposit spreads for years. Ask any branch manager who watched balances walk.

The future of finance

This is not just about replacing spreadsheet jockeys, it is about reimagining how financial services run. McKinsey predicts an additional $200 to $340 billion in annual revenue for banking as productivity jumps. The global generative AI market in finance is projected to explode from $1.19 billion to $13.33 billion over the next decade.

Signals are already on the board. Bank of America announced 676 million interactions with its AI virtual assistant Erica in 2024, and Morgan Stanley rolled out OpenAI-powered assistants to its 16,000 wealth advisors back in June.

So where does that leave people? Somewhere between a wake-up call and an opportunity. The first tasks to fade will probably be the repetitive ones everyone quietly dreads, the late-night cleanup, the sixth sensitivity table before a committee. Project Mercury looks like a down payment, a $150-an-hour wager on a future where machines do not just assist bankers, they replace entire rungs of work.