Why Is Microsoft Buying Finance, Legal, and Data AI Companies?

Among this year’s EU Digital Markets Act gatekeeper acquisition disclosures, Microsoft’s movement is strikingly enterprise-oriented.

If Google is focused on enterprise-data-based agents, cloud security, and 3D generative AI; Apple is embedding sensors and creative tools into device experiences; Amazon is moving toward robotics, logistics, and delivery infrastructure; and Meta is looking at AI agents and the future of social interaction, Microsoft is aiming most clearly at the center of enterprise work.

Financial AI research.

Legal-process automation.

AI application quality evaluation.

Agent performance improvement.

Data cleaning, transformation, and integration automation.

Natural-language web research.

Differential privacy.

The Microsoft transactions appearing in the DMA list reveal the layers required for AI to actually perform work inside enterprises.

The most recent item is Fintool.

In March 2026, Microsoft notified the acquisition of sole control over Blocktool Inc., known as Fintool, through the acquisition of all its shares. Fintool is described as a San Francisco-based company that develops financial AI research tools.

This entry clearly shows where Microsoft’s AI strategy is heading.

Microsoft already owns the enterprise-work platform: Office, Teams, Outlook, Excel, PowerPoint, Dynamics, LinkedIn, GitHub, and Azure. When AI enters this environment, the first things to change are document drafting and email summarization. But the next stage is far more specialized: financial research, legal review, data analysis, quality evaluation, and risk management, where accuracy and verification matter.

Fintool points directly to this professional-work domain.

Financial research is not simple summarization. It requires synthesizing corporate performance, market data, financial statements, regulatory filings, investment logic, and risk factors. For AI to operate in this domain, language ability alone is not enough. Source tracing, calculation accuracy, data connectivity, and auditability become essential.

Microsoft’s acquisition of a financial AI research-tool company shows that Copilot is expanding beyond general productivity assistance into tools for specialized professional work.

The second axis is legal automation.

In December 2025, Microsoft hired several former employees of Robin AI Limited and stated that, as of the notification date, it was evaluating the potential acquisition of Robin’s assets. Robin is described as a UK company that developed AI-powered software designed to automate and simplify legal processes.

Law is an area where AI adoption is both fast and cautious.

Contract review, clause comparison, risk detection, legal drafting, case and regulation search, and due-diligence work are all areas where AI can assist. But legal AI carries serious risks of hallucination and responsibility. If an AI recommends a wrong clause or cites a non-existent case, the damage can be significant. Legal automation therefore requires verification, sources, permissions, and audit records.

Microsoft’s interest in Robin AI’s people and possible assets shows how enterprise Copilot could expand into legal work.

Microsoft 365 already contains Word, Outlook, Teams, and SharePoint. Legal work happens inside these tools. Contracts are written in Word. Negotiations take place through email and Teams. Documents are stored in SharePoint. If legal AI is combined with this environment, Microsoft can expand from a document-tool provider into a legal-workflow platform.

The third axis is AI quality evaluation and agent performance.

In November 2025, Microsoft acquired Pi Labs. Pi Labs is a startup that develops tools for evaluating and improving AI application quality and agent performance.

This item is small but extremely important.

Once AI enters enterprise work, the question changes from “Does the AI answer well?” to “Does the AI perform work reliably?” Agentic AI does not simply answer once and stop. It can plan multiple steps, call tools, read data, create files, send emails, and update systems.

In this context, quality evaluation becomes essential.

Did the agent correctly understand the goal?

Did it call the wrong tool in the middle of the task?

Did it act beyond its permissions?

Does the output match the user’s intent?

What error patterns repeat?

How should task-specific success and failure rates be measured?

The tools Pi Labs develops for AI application quality and agent performance evaluation deal precisely with these questions. Microsoft’s acquisition of Pi Labs shows that evaluation and verification are becoming core infrastructure in enterprise AI competition.

This also connects to Satya Nadella’s recent argument that AI models should be treated as insider risks.

If AI agents perform important actions inside enterprises, trusting the model itself is not enough. Organizations need to observe what the agent did, evaluate its performance, test failures, and verify it independently. Technologies such as Pi Labs become part of the quality-management layer needed to place AI inside real business systems.

The fourth axis is data-engineering automation.

In November 2025, Microsoft acquired 42Layers Corp., known as Osmos. Osmos is described as a U.S. company providing data-engineering tools that automate data cleaning, transformation, and integration.

AI cannot function without data.

One of the most common reasons enterprise AI projects fail is not that the model is weak, but that the data is not ready. Data is scattered across multiple systems, stored in different formats, full of duplication and errors, and entangled with permission and quality problems. For AI to perform work, it must be able to read, connect, and clean this data.

Osmos is connected to this foundational work.

The ability to automate data cleaning, transformation, and integration can connect to Microsoft Azure, Fabric, Power BI, Dynamics, and Microsoft 365. For AI to operate inside enterprises, enterprise data must first be organized into a form that AI can use. Data-engineering automation is therefore the invisible foundation of AI applications.

The fifth axis is the natural-language web.

In April 2025, Microsoft acquired Mahi Labs. Mahi Labs is described as a startup conducting Natural Language Web research.

The available description is limited.

But the phrase “Natural Language Web” is suggestive. It points toward a future in which users operate the web and software not through commands, menus, and clicks, but through natural language. When connected to Microsoft’s Copilot strategy, this suggests a shift in which users state their goals in natural language and AI moves across web and work tools on their behalf.

Here, Microsoft’s assets are powerful.

Windows, Edge, Bing, Microsoft 365, Teams, GitHub, Azure, and LinkedIn are all surfaces where users work, search, develop, and communicate. Natural Language Web research could become a technical foundation for connecting these surfaces through a natural-language interface.

The sixth axis is privacy technology.

In March 2025, Microsoft’s subsidiary LinkedIn hired employees from Tumult Labs and secured a non-exclusive IP license from the company. Tumult Labs is described as a provider of differential privacy solutions that enable sensitive data to be shared and analyzed safely. The notification was submitted on the basis of Microsoft’s view that the hiring and non-exclusive license agreement did not constitute a concentration under the EU Merger Regulation.

This item connects to Microsoft’s data strategy.

In enterprise AI, the most sensitive issue is data. LinkedIn, in particular, holds vast professional, career, company, and network data. If AI is to use such data, there must be a balance between privacy protection and statistical analysis. Differential privacy is a technology that enables aggregate analysis while reducing the possibility of identifying individuals.

The Tumult Labs entry can be read as part of Microsoft’s effort to secure privacy-preserving technology as AI and data analysis expand.

Taken together, Microsoft’s 2025–2026 list looks like an enterprise AI stack.

Fintool is about specialized financial work.

Robin AI is about legal automation.

Pi Labs is about AI application quality and agent performance evaluation.

Osmos is about data cleaning, transformation, and integration.

Mahi Labs is about the natural-language web.

Tumult Labs is about differential privacy.

This combination shows where Microsoft wants to place AI.

Microsoft is not limiting AI to consumer conversation services. It wants to insert AI into real enterprise work, professional processes, data pipelines, evaluation systems, privacy protection, and the way users operate the web. This is a strategy to extend Copilot from a document-writing assistant into an interface for enterprise operations.

Microsoft’s strength is that it is already deeply embedded inside enterprises.

Employees send emails through Outlook, hold meetings in Teams, create documents in Word and Excel, report through PowerPoint, store files in SharePoint, develop on GitHub, and run systems on Azure. Microsoft already owns the work surface. If AI is layered onto that surface, users do not need to move to a separate AI application.

For Microsoft, therefore, the key is not a single model.

It is permission to access work data.

Domain tools that understand specialized tasks.

Quality systems that evaluate AI-generated outcomes.

Methods for measuring agent performance.

Engineering that cleans and connects data.

Analytical technologies that protect privacy.

The Microsoft entries in the DMA list fill these layers one by one.

Pi Labs and Osmos are especially revealing of the reality of AI adoption.

Enterprises want stable operations more than impressive demos. They need AI not only to produce one plausible answer, but to process repeated tasks accurately over time, trace why failures occur, manage data quality, and respond to regulation and audits. Pi Labs addresses quality and agent performance. Osmos addresses the data foundation.

This is a very Microsoft-like way to compete in AI.

Microsoft’s relationship with OpenAI, Azure infrastructure, and the Copilot product family form the front stage of its AI strategy. But the DMA list reveals the small components Microsoft is gathering behind that stage: finance, law, data, evaluation, privacy, and the natural-language web. These are the pieces enterprise AI needs in order to make real money.

Microsoft is trying to make AI the operating system of work.

If Windows was the operating system of the PC, Copilot could become the natural-language operating system of enterprise work. Users search for files, summarize meetings, review contracts, analyze financial data, clean data, prepare customer responses, write code, and create reports. AI becomes the interface connecting these processes.

But for that interface to be trusted, it must be verifiable.

If financial research is wrong, investment judgment can be distorted.

If legal automation is wrong, contract risk can increase.

If data integration is wrong, the entire analysis can fail.

If agent performance evaluation is weak, automation failures may go undetected.

If privacy technology is insufficient, sensitive data can be exposed.

Microsoft’s AI strategy is therefore not only about productivity improvement. It is also connected to responsible operational systems.

That is why this year’s Microsoft entries matter.

Fintool alone looks like an acquisition of a financial AI tool.

Robin AI alone looks like the securing of legal-automation talent.

Pi Labs alone looks like the acquisition of an AI evaluation tool.

Osmos alone looks like the acquisition of a data-engineering tool.

But when they are connected, Microsoft’s post-Copilot strategy becomes visible.

AI is moving beyond the stage of writing documents on behalf of users and entering specialized domains of enterprise work. In those domains, what is needed is not only a larger model, but also data, quality evaluation, privacy, and domain-specific workflows.

Microsoft is buying precisely this execution layer.

This strategy also matters from a regulatory perspective.

In the 2024 Inflection case, Microsoft hired Mustafa Suleyman and 64 Inflection employees and entered into an agreement that included a non-exclusive IP license from Inflection. The DMA list explains that the notification was submitted on the basis of Microsoft’s view that the transaction was not a concentration.

The Inflection case was a symbolic event showing that AI talent and IP can move without the acquisition of an entire company.

Similar structures appear again in 2025 entries such as Robin AI and Tumult Labs, where hiring and non-exclusive IP licensing are used. This shows that Microsoft is absorbing the technologies and people it needs not only through full-company acquisitions, but also through more selective structures. In the AI era, competitive capability does not move only through equity acquisitions.

Microsoft is looking at the most realistic battlefield of enterprise AI.

That battlefield is not the consumer chatbot app.

It is the legal, finance, data, development, sales, and customer-support teams of large enterprises.

It is not a single answer, but daily repeated workflows.

It is not model performance alone, but data connectivity, permissions, quality, auditability, and privacy.

The Microsoft entries in the DMA list make this battlefield clear.

The next stage of AI competition is not who gives the most impressive answer. It is who can embed AI into enterprise work more safely and verifiably. Microsoft is gathering finance, legal, data, evaluation, and privacy technologies to solve that problem.

Ultimately, Microsoft’s movement this year shows the next stage of Copilot.

Copilot began as AI that helps with work.

The next Copilot becomes AI that performs work.

The stage after that must become an enterprise operating layer that evaluates work, connects to data, and responds to regulation and audits.

Fintool, Robin AI, Pi Labs, Osmos, Mahi Labs, and Tumult Labs are pieces of that direction.

Microsoft is not a company that leads with one AI model alone. It already owns the work surface into which AI will enter. What it now needs is to fill that surface with data, evaluation, privacy, and domain knowledge so that AI does not fail when it performs real work.

This year’s DMA list shows that Microsoft is buying those pieces.

As AI enters the everyday work of enterprises, the contest will not be decided by the eloquence of a model.

It will be decided by the ability to take responsibility for work from beginning to end.

Microsoft is building the infrastructure for that responsibility.