OpenAI courts Wall Street with ChatGPT for financial services, developed with Morgan Stanley

OpenAI launched a finance-specific ChatGPT product for eligible banks and financial institutions, powered by its GPT-6 Astra model and developed with input from Morgan Stanley and Evercore. The platform connects to Bloomberg, FactSet, Daloopa, PitchBook, Crunchbase and LSEG News data, offers roughly 50 MCP connectors, and can generate editable Excel models, PowerPoint pitchbooks and dashboards; OpenAI said sample slides took about 10 minutes to create. The targeted vertical product is intended to accelerate OpenAI's enterprise revenue growth while competing with Anthropic's finance-focused Claude offering, with enterprise security controls including encryption, SAML SSO, SCIM and role-based access.

Analysis

The investable implication is less about a single enterprise-seat win and more about whether regulated workflows begin shifting from document production to AI-native research, modeling, and scenario analysis. That transition raises utilization of proprietary data rather than necessarily displacing it: FDS and LSEG should benefit if AI increases query volume and makes their datasets more embedded in daily workflows. The medium-term risk is pricing power—if the AI interface becomes the primary user experience, data vendors could face weaker terminal stickiness and greater pressure to unbundle premium workflow products over 6-18 months.

MS and EVR are useful read-throughs for adoption, but revenue upside will initially be indirect: faster pitchbook and diligence throughput can improve banker capacity and operating leverage only if deal activity recovers. The more immediate 1-3 month catalyst is evidence of production deployment, especially disclosed user counts, compliance approvals, and measurable reductions in analyst-to-associate preparation time. A pilot with curated datasets is not equivalent to broad rollout; bank security, model-risk governance, citation accuracy, and information-barrier controls remain the gating items.

MSFT is the clearest relative loser at the margin because this expands competition from AI assistants into Office-adjacent artifacts and enterprise workflow orchestration. That said, the market should not extrapolate a near-term collapse in Copilot economics: financial institutions have long procurement cycles and deeply embedded Microsoft identity, document, and collaboration infrastructure. The contrarian view is that incumbent data vendors and MSFT may monetize the integration layer while model providers absorb high inference costs, particularly if high-quality outputs require consistently expensive reasoning settings.

Key Decisions for Investors

• Maintain a 6-12 month long FDS/LSEG relative-value basket versus a broad software ETF: AI-assisted workflows can lift data consumption and reinforce dataset importance; reassess if either vendor reports declining desktop retention, material API pricing concessions, or AI-native product cannibalization.
• Use MS as an adoption watch rather than a directional AI trade: add only following disclosed firmwide deployment or evidence that AI productivity supports compensation-ratio improvement. Thesis fails if governance restrictions confine usage to low-value drafting tasks through 2027.
• Initiate a small 3-6 month long EVR versus a diversified financials hedge only if M&A activity and announced AI-enabled workflow adoption coincide; productivity has greatest earnings leverage at boutiques, but weak advisory volumes can overwhelm efficiency gains. Exit on renewed fee-pool deterioration or lack of operating-margin guidance support.
• For MSFT, avoid chasing a headline-driven short. Instead, monitor enterprise AI renewal data and Copilot attach rates; a sustained deceleration alongside confirmed finance-assistant displacement would justify a 6-12 month MSFT underweight versus large-cap software.