If Everyone Has AI, Who Wins?

If everyone in investing has AI, AI itself stops being the edge.

The winners will be the investors and platforms with something around the model that others cannot easily copy: better context, more reliable data, repeatable research workflows, institutional memory and stronger judgement.

That is the important change in the AI investment debate. The question is no longer who has access to the smartest chatbot. Powerful foundation models will be widely available. The question is what those models can see, what work they can reliably perform, what the organisation remembers and how the resulting evidence changes an actual portfolio decision.

AI becomes the interface. The investment system underneath it becomes the advantage.

The model is moving towards the commodity layer

For a short period, simply having access to a capable language model felt differentiated. An analyst could summarise a long annual report, compare two conference calls or draft an investment memo in a fraction of the usual time.

That capability is rapidly becoming normal.

Claude, GPT, Gemini and other models will all be able to read documents, analyse spreadsheets, write code and produce professional-looking research. Every investor will be able to ask for a summary of HDFC Bank, a list of Bajaj Finance’s risks or a first-pass valuation of an industrial company.

The quality of those outputs will continue to improve. Their scarcity will continue to fall.

If everybody can generate a polished 30-page memo in 30 seconds, the memo itself becomes much less valuable. The cost of producing research prose approaches zero. The value moves to the parts the prose cannot settle:

• Which source is correct?
• What was actually known on the decision date?
• Which KPI has historically driven the business?
• Did management deliver what it previously promised?
• Which assumption does the valuation depend on?
• What does this new information change inside the portfolio?
• What would prove the original thesis wrong?

These are system questions, not chatbot questions.

The edge moves into five layers

1. Proprietary context

Everyone can ask the same model to analyse the same company. They will not all give it the same context.

One investor may provide the latest annual report. Another may have a clean history of company-specific KPIs, every change in management guidance, point-in-time estimates, previous investment notes, portfolio exposures and a record of the decisions the team made before each result.

The second investor is not necessarily using a better model. The model is working inside a better information environment.

This distinction is easy to miss because the answer appears in the same chat window. But the answer is shaped by everything behind that window.

Generic public context might include:

• the latest filings;
• recent financial statements;
• conference-call transcripts;
• current valuation ratios;
• recent news and presentations.

A richer investment context can add:

• historical statements as they were originally reported;
• company-specific operating KPIs;
• management-guidance history;
• changes in accounting quality and forensic signals;
• the assumptions in the last model;
• the original investment thesis;
• current portfolio exposure and risk limits;
• a record of what the team previously expected.

The foundation model may be commoditised. The context layer is not.

2. Better research infrastructure

More data is not automatically better context.

Financial information arrives through filings, transcripts, presentations, exchange announcements, spreadsheets, emails and market feeds. Periods differ. Definitions change. A company may restate history. Standalone and consolidated figures can look similar while describing different businesses. An annual figure available today may not have been available on the date being studied.

An AI system connected to an untidy pile of documents is simply a faster way to misunderstand the pile.

Research infrastructure determines whether the model receives information that is comparable, dated, sourced and fit for the question. It should make four things explicit:

1. Provenance: where did this number or statement come from?
2. Time: when did it become knowable?
3. Basis: which period, accounting definition and consolidation basis does it use?
4. Consistency: can it be compared with the other observations in the analysis?

This work is less visible than a clever answer. It is also more defensible.

A general chatbot can explain a P/E ratio. A proper research system can show the exact earnings measure behind it, the date on which that measure became available, the price used in the calculation and whether an exceptional item distorted the denominator.

The intelligence in the second answer comes partly from the model. Much of it comes from the structure underneath.

3. Proprietary workflows

Two analysts can have identical information and identical access to Claude. Their outcomes can still be very different.

Analyst A types:

Analyse Bajaj Finance.

Analyst B works inside a system that, when a result arrives:

1. detects what changed from the previous quarter;
2. compares the result with management’s earlier guidance;
3. refreshes the historical spreadsheet and model inputs;
4. flags unusual changes in cash conversion or accounting lines;
5. reruns the research scorecard;
6. checks how valuation changed;
7. maps the result to portfolio exposure;
8. retrieves the original thesis and its failure conditions;
9. produces a sourced update for the analyst to review.

Both analysts may use the same underlying language model. Analyst B has a much stronger process.

The important difference is that the second workflow is repeatable. It does not depend on remembering every check during a busy results week. It makes the organisation’s standard of analysis part of the system.

That is what a proprietary workflow really is. It is not a secret prompt. It is a sequence of checks, sources, calculations, decisions and review gates that reflects how a particular investor thinks.

Prompts are easy to copy. A mature research process is not.

4. Feedback loops and investment memory

The strongest long-term advantage may be the ability to learn from the firm’s own history.

Imagine a research system that can preserve statements such as:

• We forecast 18% revenue growth. The company delivered 12%.
• Management’s guidance has historically exceeded the eventual result.
• Our margin assumptions in this sector have repeatedly been too optimistic.
• This factor behaved poorly during rising-rate periods.
• We bought this company for one reason, but later justified holding it for another.

Each observation turns an isolated decision into a feedback loop.

Most investment organisations possess some version of this knowledge. It is scattered across old spreadsheets, committee memos, email threads and the memories of individual analysts. When a person leaves, much of the context leaves too. When a new quarter begins, the team often rebuilds the same understanding from scratch.

A durable research system keeps the original forecast, the evidence available at the time, the decision that followed and the eventual outcome. It does not quietly overwrite the past with a cleaner present-day story.

This is difficult for a new competitor to reproduce. Buying access to the same model and the same public data cannot recreate years of decisions, mistakes, overrides and institutional learning.

The compounding asset is not merely a larger database. It is a better record of how the organisation thinks and where that thinking has been wrong.

5. Human judgement

AI makes analysis cheaper. It does not make capital allocation trivial.

A model can produce arguments for and against the same company. It can build scenarios, highlight anomalies and show how an assumption changes valuation. It cannot remove the need to decide:

• Which assumptions do you believe?
• Which risks are tolerable?
• Which signal matters now?
• What information is noise?
• When should you wait?
• When should you act?
• How much capital should the idea receive?

These are not flaws left behind because the technology is incomplete. They are the substance of investing under uncertainty.

Judgement also includes restraint. A better system may reveal that the evidence is insufficient, that the model is too sensitive to one assumption or that no action is required. Producing more analysis is not the same as making a better decision.

The scarce analyst is not the person who can generate the most material. It is the person who can use abundant material without losing the hierarchy of what matters.

The AI stack and where the edge begins

The investment research stack increasingly looks like this:

Foundation models
        ↓
Generic AI agents
        ↓
Public financial information
        ↓
────────────────────────────────
        THE EDGE STARTS HERE
────────────────────────────────
        ↓
Reliable point-in-time research data
        ↓
Firm-specific workflows and models
        ↓
Portfolio context
        ↓
Investment memory and feedback
        ↓
Human judgement
        ↓
Capital-allocation decisions

The line is not perfectly fixed. Some public information can be assembled by anyone, and some workflows will become standard software features. The point is that defensibility moves upward as the lower layers become abundant.

The model helps at every layer. It is not the moat by itself.

The Altys angle: build the system around the AI

This is the distinction that matters for Altys.

“AI-powered equity research” is not a strong enough thesis on its own. Models will improve, prices will fall and access will widen. A prettier chat interface is not durable differentiation.

The stronger thesis is:

AI becomes the interface. The research infrastructure underneath it becomes the advantage.

Altys is built around that system view. Its public product today brings together several parts of the investment process:

Point-in-time datasets

Research and backtests should use what was actually known at the time, not a history cleaned up with later restatements. Altys keeps dated financial and market information so an investor can distinguish the current record from the information set that existed on an earlier decision date.

Strategy and backtesting infrastructure

An investment rule should be tested on the full historical record rather than judged by the examples an investor remembers. Altys lets users define strategy rules, test them on point-in-time data, inspect historical holdings and carry a selected strategy into forward monitoring.

GenGrid and parallel research

Traditional research moves one company at a time. GenGrid puts companies down the rows and the questions or metrics across the columns, allowing an investor to compare financials, valuation, guidance, risks and other evidence across a universe rather than reconstructing the same worksheet repeatedly.

Company intelligence and workbooks

Filings, concalls, guidances, financials, valuations, factors and forensic evidence belong in one research surface with sources attached. Workbooks then let the investor ask questions, run analysis and model scenarios using the same underlying company context.

Monitoring

Research should not stop when an idea enters a portfolio or watchlist. A monitored strategy or company should be re-evaluated as new filings, results and relevant data arrive. The system should surface what changed and take the analyst back to the evidence.

None of these parts eliminates judgement. Together, they make judgement better informed, more consistent and easier to audit.

That is a more useful role for AI than asking it to pronounce whether a stock is “good.”

Financial data alone is not enough either

It would be equally misleading to say that financial data by itself is the moat. Much of that information is public and can ultimately be obtained elsewhere.

The harder-to-copy asset is the combination:

clean point-in-time data × derived financial context × company-specific KPIs × guidance history × strategy tests × models × portfolio context × institutional workflows × accumulated research memory

Each component improves the usefulness of the others.

Point-in-time data makes a backtest more honest. Guidance history makes a model assumption more informed. A model makes a result update actionable. Portfolio context shows whether the update matters. Research memory reveals whether the analyst has seen the same pattern before and misread it.

The combination turns information into a process.

Why infrastructure may become more valuable as AI improves

This is the apparent paradox.

If AI becomes dramatically more capable, it may make institutional research infrastructure more valuable, not less.

More intelligence creates more possible analysis. That increases the value of reliable context, clear provenance and disciplined workflow. When a model can run hundreds of investigations at once, an error in the source or framing can also scale hundreds of times. When it can produce any argument on demand, the record of what was believed before the outcome becomes more important.

Abundant intelligence does not remove scarcity. It relocates it.

Context becomes scarce. Reliability becomes scarce. Institutional memory becomes scarce. The ability to convert evidence into a proportionate portfolio decision remains scarce.

So, who wins?

Not whoever has the best chatbot this quarter.

The winner is the investor with the best investment system around it: the context that improves the question, the infrastructure that keeps the evidence reliable, the workflows that make research repeatable, the memory that turns mistakes into learning and the judgement that decides what deserves capital.

That is the bet behind Altys. AI will become ubiquitous. Institutional-grade investment infrastructure will not.

Public references

• Altys institutional research platform
• How to use Altys: strategies, GenGrid, workbooks and monitoring

Related reading:

• Investing then vs now: what AI actually changed
• Why point-in-time data matters
• Why citations are non-negotiable in financial AI
• Financial modelling with AI

This article is educational. Altys Labs is not a registered research analyst or investment adviser, and nothing here is investment advice or a recommendation to buy, sell or hold any security.