OpenAI's Wall Street Gambit: How AI is Set to Eclipse Junior Bankers' Grind

OpenAI has quietly assembled more than 100 former investment bankers to train its models on the arcane art of financial modelling. Dubbed Project Mercury, this initiative is not just about automating grunt work—it could redefine the entry gates to the world's most lucrative industry, forcing banks to rethink talent pipelines and productivity.

The project, shrouded in secrecy until Bloomberg's revelations, pays ex-bankers from blue-chip firms such as JPMorgan Chase, Morgan Stanley and Goldman Sachs $150 an hour to generate prompts and build models for everything from initial public offerings to corporate restructurings. These contractors, including MBA candidates from Harvard and the Massachusetts Institute of Technology, submit one Excel-based model weekly, adhering to industry conventions like italicised percentages and precise margins.

The workflow is deceptively simple: draft a prompt in plain language, execute the model, incorporate reviewer feedback, and feed the refined output into OpenAI's systems to hone its AI for more accurate, first-draft financial documents. OpenAI's spokesperson stresses that such experts are engaged through third parties to bolster model capabilities across domains, a nod to the company's broader quest for commercial viability amid mounting losses. For the first half of 2025, OpenAI reported $13.5 billion in losses, despite an income of $4.3 billion.

Despite the ambitious goals of the OpenAI project, a MIT Sloan Management Review study from August 2025 notes that in finance, embedded AI in back-office functions like reporting can yield benefits, but standalone pilots—often tied to layoffs—rarely deliver measurable ROI, with 95% stalling due to integration failures or overestimated capabilities. JPMorgan Chase, for instance, has shuttered hundreds of AI initiatives since 2024, including those aimed at automating advisory tasks, though not directly linked to post-layoff regrets; executives admitted many prototypes contributed minimally to bottom-line efficiencies despite initial hype. 

The Mechanics of Mercury: Codifying Expertise

At its core, Project Mercury leverages reinforcement learning to embed Wall Street's tacit knowledge into AI, allowing the technology to produce deal-ready analyses with minimal human intervention. Participants, drawn from outposts like Brookfield, Evercore and KKR, undergo a near-fully automated recruitment: a 20-minute AI chatbot interview, a financial statements quiz, and a modelling test. This efficiency mirrors the very drudgery it aims to supplant—the endless "pls fix" emails that have become a meme for junior analysts' plight. Early access to the AI tools grants these trainers a front-row seat to their own obsolescence, yet it underscores OpenAI's strategy: turn domain specialists into data annotators, accelerating AI's leap from novelty to necessity in high-stakes finance.

Supporting evidence from peers amplifies Mercury's ambition. Morgan Stanley, for instance, already deploys OpenAI-derived technology in wealth management, merging proprietary data with generative AI for portfolio summaries and market insights. Raj Bakhru, general manager at BlueFlame AI, observes that "nobody writes financial documents better than highly trained analysts at investment banks," positioning their input as the key to making AI outputs "much more useful" for pitches and valuations. With OpenAI securing a $4bn credit facility from banks including JPMorgan, the lines between innovator and incumbent are blurring, hinting at symbiotic partnerships rather than outright rivalry.

Reshaping Banking's Labour Model

The implications for investment banking are profound. Junior roles, long a rite of passage marked by exhaustion and hierarchical hazing, face existential pressure as AI tackles sensitivity analyses, circularity checks and overnight tweaks. Productivity gains could be transformative: banks might compress deal timelines, slashing errors in mergers or buyouts while redirecting juniors toward client-facing strategy and execution.

Yet this shift risks exacerbating talent shortages; without robust upskilling, the 80-hour desk marathons that build institutional lore could vanish, leaving a skills chasm in an industry already grappling with burnout. Broader sectors—consulting, legal and technology—may follow suit, templating AI for any process demanding precision and structure, potentially amplifying economic efficiencies but widening inequality if adoption favours the resourced few.fortune+3​

Critics, including junior bankers themselves, counter that such "grunt work" fosters indispensable intuition, even as AI encroaches. Regulators, too, will scrutinise: how does automated modelling square with accountability in audited financials? OpenAI's push, amid a valuation surge, reflects Sam Altman's urgency to monetise AI beyond hype, but profitability hinges on navigating these tensions.eweek+1​

Forging New Frontiers in Finance Careers

Project Mercury is not merely disruptive—it is generative, spawning hybrid careers at the AI-finance intersection. Ex-bankers could evolve into prompt engineers or AI oversight specialists, commanding premiums for ensuring models comply with regulations or tailor to niche transactions. A gig economy of flexible expertise beckons, with roles in AI ethics auditing or fintech innovation drawing displaced talent. Business schools may pivot curricula toward "financial AI literacy," equipping graduates for human-machine teams that amplify deal-making acumen.