OpenAI launched ChatGPT for Financial Services on Sept. 10, embedding Daloopa, PitchBook, LSEG News and Crunchbase datasets directly into the product on its own infrastructure, bypassing the separate data contracts incumbents monetize. FactSet and S&P Global shares weakened on launch day as investors reassessed the value of the data-access layer.
OpenAI is now hosting other companies' financial data on its own servers, and the vendors that used to sell that access are paying for it in the market. ChatGPT for Financial Services, launched Sept. 10 and powered by the new GPT-6 Astra model, ships with Daloopa, PitchBook, LSEG News and Crunchbase datasets already indexed inside the product — no separate contracts, no connector setup, no per-seat data negotiation.
"We are effectively teaching ChatGPT to research like an analyst and to justify its conclusions like an analyst," OpenAI product vice president Nick Turley said. Asked whether the tool would cut junior banking hires, Turley compared it to Microsoft Excel, saying the technology would let teams "do analysis faster and better" rather than replace them.
The architecture is the part that matters. OpenAI indexes and hosts the data on its own infrastructure, which the company says improves retrieval speed and latency. Every figure carries a citation traceable to a specific table or passage, so a banker normalizing a P&L can inspect the reconciliation behind an adjusted Ebitda number, see which costs were excluded, and decide how to use it in a valuation. Administrators can publish firm Excel, Word and PowerPoint templates, meaning output lands in house-formatted valuation models, research notes and pitchbooks rather than generic documents.
On OfficeQA Pro, a benchmark that tests how well systems locate and analyze information across complex U.S. Treasury documents, OpenAI reported GPT-6 Astra scored 69.9%, against 60.2% for GPT-5.6 Sol and 62.4% for Anthropic's Claude Fable 5.1. In internal slide-generation tests judged by human blind comparison against Anthropic's Opus 5, GPT-6 Astra won 55.6% of the time, versus 21.6% for GPT-5.6 Sol.
The bundling decision landed on the incumbents immediately. FactSet and S&P Global shares weakened on launch day as investors reassessed what the data-access layer is worth when a model provider embeds the underlying datasets itself. Both companies are named in the product as integration partners — firms holding existing subscriptions can connect them through shared sign-in and entitlement arrangements, alongside S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva and Moody's. That framing cuts both ways: it preserves the incumbents' seat revenue inside ChatGPT, but it also makes them a feature of someone else's interface rather than the interface itself.
The pressure extends past the large vendors. Vertical AI startups built on single-purpose earnings parsing or research-report generation now compete against a product that bundles the same inputs with a frontier model on top. The information-organizing middle layer — the businesses that charged for assembling and formatting data — is the layer being repriced.
Morgan Stanley and Evercore served as design partners, shaping features around regulated finance workflows. The product builds on ChatGPT Enterprise controls: SAML single sign-on, SCIM provisioning, role-based access, encryption at rest and in transit, configurable retention, and multiple workspaces to enforce information barriers. Compliance teams can export workspace logs into existing audit workflows, and OpenAI said institutional business data is not used to train its models by default. More than 50 connectors, including Box, Datasite, Preqin and Intapp, are supported.
The labor angle is where the launch gets uncomfortable. Entry-level banking roles have functioned for decades as an apprenticeship in exactly the tasks the product automates — pulling data, building models, formatting decks. Packing research, modeling and pitchbook generation into one workspace compresses the workload that historically justified those seats. Turley's Excel comparison is the company's answer; the market's answer will show up in bank headcount plans over the next several hiring cycles, not in a press release.
The competitive context explains the timing. OpenAI and Anthropic are both pushing into regulated enterprise verticals where data governance and audit trails matter as much as raw model quality, and financial services is the highest-value beachhead. Anthropic's Claude Fable 5.1 already appears in OpenAI's own benchmark table, which tells you how closely the two are tracking each other.
For investors, the read-through is directional rather than settled. FactSet and S&P Global still own the entitlements, the historical depth and the compliance relationships that banks will not abandon quickly, and the integration deals keep their content inside the workflow. What changed on Sept. 10 is who controls the surface the analyst actually looks at. OpenAI said it plans to widen the data available in the product and expand beyond investment banking and equity research across the broader financial services sector — a roadmap that points at more of the incumbents' revenue per user, not less.
This article is for informational purposes only and does not constitute investment advice.
