Blog/Self-hosted AI for financial analysis
Private AI

Self-hosted AI for financial analysis

By Samuel Seidel · September 9, 2026 · 8 min read

An LLM can summarize a forecast model, flag inconsistencies across a data room and draft the first pass of an analysis. The documents involved (an unannounced quarterly forecast, a diligence data room, a margin analysis that would move the stock) are material non-public information, so the model should run on a dedicated instance scoped to the deal team, not a shared AI product.

Why this data is unusually sensitive

Material non-public information carries a specific kind of risk that a generic "confidential documents" label doesn't fully capture. A competitor seeing it early is one problem. The bigger one is that anyone who sees it early, an AI vendor's systems or staff included, holds information that could be traded on before it's public. That's true whether the leak is a person acting in bad faith or simply a document sitting somewhere it shouldn't, retained by a third-party system longer than intended.

M&A due diligence compounds this. A data room during an acquisition often contains the target's own MNPI plus the acquirer's strategic thinking about the deal, in one place, at the exact moment secrecy matters most before signing. Running it through a general-purpose AI product means a document a handful of named people are cleared to see also passes through a shared service with many more hands that could, in principle, touch it.

What a dedicated instance changes

The mechanics match any self-hosted document workflow: an instruct model, possibly with retrieval over the data room the way semantic search works for a general document store, running on hardware assigned only to your organization. Forecasts, board decks and diligence documents go to your own model, so there's no AI vendor retention or training policy to weigh against insider-trading exposure.

On a rented Spark, the hardware is GPUwerk's, in EU-Central, and GPUwerk keeps standard infrastructure administrator access under a DPA. Put the hosting provider in your information-barrier analysis like any other processor and clear it with your compliance function before live MNPI goes on the machine.

None of this replaces the practices finance and legal teams already use around MNPI: restricted lists, need-to-know distribution, deal code names. It's the same discipline applied to a new tool. If a document wouldn't go in an email to someone outside the deal team, it shouldn't go into an AI product whose access model you haven't reviewed.

Where this fits in a workflow

Useful, lower-risk starting points: summarizing a long forecast model into a two-page brief for a board update, checking a data room for internal inconsistencies between documents (a revenue figure in the deck that doesn't match the underlying financials), or drafting the first pass of a variance analysis that a human analyst then checks and owns. None of these needs the model to make a judgment call. They speed up the reading and cross-referencing that eats an analyst's day.

Keep the model's output as a draft an analyst reviews, not a number that goes straight into a report. LLMs read and summarize text well; they don't replace the modeling and judgment a forecast or valuation requires. Treating their output as authoritative on numbers they didn't verify introduces errors that are hard to trace.

Access boundaries need to mirror the deal team, not the company

A company-wide internal AI tool is the wrong place to run deal analysis, even a self-hosted one, if the deal team is a smaller set of people than everyone with access to that tool. A separate instance scoped to the deal team makes the access boundary match who's cleared to see the documents, instead of relying on everyone with the shared tool to show restraint about a deal they shouldn't know exists. When the deal closes or falls through, wind that instance and its documents down deliberately, not left running with logins for ex-deal-team members.

Logging matters more here than in most internal AI use cases. Because MNPI carries insider-trading exposure, being able to show who queried which document and when, if that question ever comes up, is worth having in place before it's needed rather than reconstructed after the fact.

One Spark per deal

A single Spark's 128 GB of unified memory runs a large instruct model for document analysis, with room for a retrieval layer over the data room on the same machine. Giving each active deal its own instance, separate from other internal AI work, keeps the access boundary as tight as the deal requires. At $0.79/hour in Europe, North America or East Asia, billed per minute, the instance costs only what it runs for the length of the deal.

Related pages

First top-up: pay $20, get $30 in credit

Give each deal its own isolated model.

A dedicated DGX Spark in EU-Central, $0.79/hour, isolated to the deal or team that needs it.

Your data stays on a machine assigned only to you. GPUwerk processes it under its published DPA.

Create account Private LLM hosting