Signals Inbox·August 27, 2026·AI Infrastructure

What startups Microsoft bought recently?

Microsoft has bought only three startups outright lately: Fintool, Osmos and Pi Labs. But the acquisition list hides a much broader push to absorb AI teams, technology and expertise through hiring and licensing deals.

Q1Which startups has Microsoft actually bought lately?

As of today, Microsoft’s official acquisition history shows only three recent startup purchases: Fintool, Osmos and Pi Labs.

That is much less than Microsoft used to buy. Fintool was added to Microsoft’s acquisition record in April 2026, Osmos in January 2026 and Pi Labs in December 2025. Before those three, the list jumps back almost three years to Fungible in January 2023. Microsoft recorded no acquisition at all in 2024.

The three deals are also unusually small and focused compared with Microsoft’s older blockbuster purchases. Fintool built AI agents for financial research and Microsoft 365 workflows. Osmos built agentic data-engineering software that already worked inside Microsoft Fabric. Pi Labs worked on measuring and improving the quality of AI applications.

So the short answer is three. But that answer becomes misleading once we look at how Microsoft has been getting startup talent lately. Several other startup teams have moved into Microsoft through hiring and licensing deals without appearing on Microsoft’s acquisition list.

Q2Has Microsoft suddenly stopped buying companies?

Microsoft has dramatically slowed conventional acquisitions: its official deal rate has fallen by roughly 80% from the pace we saw before 2022.

We counted 52 acquisitions on Microsoft’s own list from 2018 through 2021. That works out to 13 per year. From the beginning of 2022 through today, the same list contains only 10, or a little over two per year.

The drop gets even sharper recently. Microsoft bought six companies in 2022, one in 2023, zero in 2024, one in 2025 and two so far in 2026. This is a real break from Microsoft’s previous behavior, when years with eight, 13 or even 17 acquisitions were normal.

Yet Microsoft has continued pulling external technology and teams into the company during this quieter M&A period. The missing piece is that some of its most important startup deals no longer look like normal acquisitions.

Q3Why did Microsoft buy Fintool?

Microsoft bought Fintool because the startup had already built working AI agents for tasks that fit almost perfectly inside Excel, PowerPoint and Word.

Fintool was founded in 2023 by Nicolas Bustamante and Edouard Godfrey. Its agents read earnings calls and company filings, synthesize research and perform work normally done manually by investors and analysts. Bustamante said after the acquisition that thousands of professionals across hedge funds, asset managers and investment banks were using Fintool for research.

The more revealing development came shortly before Microsoft bought the company. Fintool V5 could autonomously build a DCF model in Excel, generate an earnings presentation in PowerPoint or write a research memo in Word. Those are already Microsoft workflows.

Fintool had raised about $7.2 million according to CB Insights, while the acquisition price was not disclosed. The startup therefore brought Microsoft something more useful than financial scale: a small team that had already learned how professionals want agents to behave inside the exact software Microsoft owns.

Bustamante also said Microsoft’s Office Product Group, including the team responsible for Excel, was directly involved in the deal. He plans to bring Fintool’s expertise into Microsoft 365, first around financial services and then into other professional workflows.

Q4Why did Microsoft buy Osmos?

Microsoft bought Osmos because Fabric still has a basic problem: companies cannot do much with AI until their messy data has been cleaned, connected and prepared.

Osmos had spent years working on that problem. Founded in 2019, the startup initially helped companies ingest external data from customers, suppliers and other sources. It later built AI agents that could turn raw information into structured data ready for analytics and AI.

By the time Microsoft acquired Osmos, the startup was already deeply embedded in Microsoft Fabric. Microsoft had published a customer story about the company several months earlier, and Osmos had built native Fabric products including an AI data wrangler and an AI data-engineering agent. Microsoft product executive Roy Hasson said those tools were cutting customers’ development and maintenance work by about 50%.

Microsoft then moved the Osmos team into its Fabric engineering organization. The startup began shutting down its standalone products as its technology moved directly into Fabric and OneLake.

Osmos had raised $13 million in a round led by Lightspeed Venture Partners. Microsoft never disclosed the purchase price. Again, the interesting part is how little integration risk Microsoft was taking: Osmos was already building on Fabric, already serving Fabric users and already solving a problem Microsoft wanted Fabric itself to handle.

Q5Why did Microsoft buy Pi Labs?

Microsoft bought Pi Labs because AI products are getting harder to judge, and Pi Labs had built tools that help developers figure out whether an AI system is actually giving good answers.

Pi Labs was tiny, with LinkedIn listing fewer than 10 employees, but its product attacked a very current problem. Pi Studio took product requirements, system prompts, user feedback and A/B test results and turned them into evaluation criteria for AI applications.

That becomes particularly useful as software moves from simple chatbots toward agents. A developer can easily test whether traditional software returns the expected number or opens the right page. Judging whether an agent was helpful, accurate or good at completing a multi-step task is much messier.

Pi Labs tried to automate that judgment. Its software could generate evaluation rubrics, run them against model outputs and keep adjusting the metrics as more user feedback arrived.

There was also a strong people angle. Microsoft executive R.V. Guha described founders Achint Srivastava and David Karam as former collaborators and publicly celebrated the chance to work with the team again. Microsoft’s own acquisition ledger records Pi Labs as a LinkedIn acquisition. The price was never disclosed.

Q6Did Microsoft actually buy Inflection AI?

Microsoft never bought the Inflection AI company, but it hired almost the entire team and licensed its technology in a deal reported at roughly $650 million.

The distinction sounds technical until we look at what moved. Microsoft hired Inflection co-founder Mustafa Suleyman, co-founder and chief scientist Karén Simonyan and most of a roughly 70-person team. Suleyman became CEO of Microsoft AI, while Simonyan became chief scientist.

Microsoft also entered into a non-exclusive licensing arrangement covering Inflection’s models. Reuters and Bloomberg reported a payment of roughly $650 million around the transaction.

Inflection itself survived. It kept its corporate entity and could continue selling or licensing technology. Calling it a normal acquisition would therefore be inaccurate.

The UK Competition and Markets Authority still decided that the arrangement qualified as a relevant merger situation. Its reasoning was unusually important: Microsoft had acquired the team’s collective know-how, and in frontier AI the researchers building and improving the models can be one of the most valuable parts of the business.

The regulator eventually cleared the transaction. But the case showed how far Microsoft can go toward absorbing a startup without buying its shares.

Q7Which startup teams has Microsoft hired without buying the company?

Microsoft has hired several entire or substantial startup teams lately, including Cove, Robin AI, Bluesky Data and Tumult Labs, without recording those companies as normal acquisitions.

Cove is the cleanest recent example. The Sequoia-backed startup built a visual workspace where people could work with AI on an editable canvas rather than through a standard chatbot. Cove announced in March 2026 that its entire team had accepted offers to join Microsoft AI. The product was subsequently shut down.

Robin AI followed a different route. Microsoft hired at least 18 employees from the legal-AI startup, according to Bloomberg Law. Former Robin AI CTO Carina Negreanu moved into Microsoft Word, and former AI director Diana Mincu joined as a principal applied science manager. Microsoft explicitly said it had no plan to acquire Robin AI.

Earlier transactions show that this approach had already become part of Microsoft’s playbook. Disclosures collected under Europe’s Digital Markets Act classify Microsoft’s Bluesky Data transaction as “hiring, license” and its Tumult Labs transaction the same way. Bluesky worked on cloud-data efficiency, while Tumult specialized in differential privacy.

As we saw previously with Inflection, Microsoft can therefore move a startup’s people and technical knowledge inside the company while leaving the corporate shell outside.

Q8Is Microsoft basically only shopping for AI startups now?

Microsoft’s recent startup activity is overwhelmingly about AI, with nearly every formal acquisition or startup-team deal we found tied to AI itself or to the infrastructure around it.

The range is broader than foundation models. Microsoft has gone after financial agents, AI evaluation, data engineering, privacy, legal workflows, collaboration interfaces and frontier-model researchers.

The mix is revealing. Microsoft already has access to several leading model families through Azure and its partnerships. What it keeps reaching outside the company for are specialist teams that know how to make those models useful in specific products.

There is also very little comparable recent buying around gaming, advertising or conventional SaaS. That is a big change from the Microsoft acquisition lists of 2018 to 2021, which included everything from GitHub and game studios to cybersecurity companies and productivity software.

These days, Microsoft’s startup shopping is much narrower. AI is driving almost all of it.

Q9Does Microsoft care more about startup teams than startup products?

Microsoft increasingly seems more interested in a startup’s team and technology than in keeping its standalone product alive.

The post-deal behavior is revealing. Osmos started winding down standalone products as its technology moved into Fabric. Cove closed entirely when the team joined Microsoft AI. Tumult Labs said its existing operation would wind down as team members moved to LinkedIn. The Inflection deal transferred most of the people who had built its original AI effort while the startup itself continued under a different strategy.

Fintool is another variation. Its founder specifically described bringing the company’s expertise and intellectual property into Office and Microsoft 365. Microsoft already owns the distribution layer, so preserving a separate startup application is often unnecessary.

That advantage is enormous. A startup can spend years persuading companies to adopt a new tool. Microsoft can put the capability inside Word, Excel, Fabric, LinkedIn, GitHub or Copilot and reach an existing user base immediately.

The recent deals look more like targeted injections of technical expertise into major Microsoft products than attempts to build a collection of new standalone brands.

Q10Is Microsoft buying startups to become less dependent on OpenAI?

Yes. Microsoft is actively looking at AI startup acquisitions partly because it wants the ability to compete without depending so heavily on OpenAI.

Reuters reported in May 2026, citing five people familiar with the matter, that Microsoft was shopping for AI startups as it prepared for a future where it could operate more independently from OpenAI. Three of those sources said acquisitions could bring in researchers needed for Microsoft’s goal of building a cutting-edge model internally.

Microsoft’s relationship with OpenAI has also loosened substantially. Under the amended partnership announced in April 2026, Microsoft’s license to OpenAI models and products became non-exclusive, OpenAI gained more freedom to serve products through other cloud providers, and Microsoft kept the right to develop competing AI independently.

Microsoft’s latest annual filing makes the relationship easier to size. The company says it now owns roughly 25% of OpenAI on an as-converted basis and recorded $24.1 billion of fiscal-year revenue from commercial arrangements with OpenAI, including revenue-sharing payments. OpenAI therefore remains extremely important to Microsoft financially.

At the same time, Microsoft is diversifying. It has invested in Anthropic, distributes competing models through Azure and is building more models itself. Startup acquisitions give it another route to secure researchers and technology it directly controls.

Microsoft is preparing for much greater AI independence while keeping an OpenAI relationship that remains worth tens of billions of dollars to both sides.

Q11Why did Microsoft give up on buying Cursor?

Microsoft considered buying Cursor but walked away over regulatory concerns, and the opportunity is now gone because SpaceX has since completed a $60 billion acquisition of the AI coding startup.

Reuters reported that Microsoft explored buying Cursor earlier in 2026. According to people familiar with the talks, executives worried regulators would object because Microsoft already owns GitHub and sells GitHub Copilot, one of Cursor’s major competitors.

That concern stopped looking hypothetical once Cursor’s value exploded. The startup went from a $2.5 billion valuation in early 2025 to $9 billion several months later and then $29.3 billion by the end of that year.

SpaceX ultimately moved much more aggressively. It agreed to acquire Cursor for an implied $60 billion equity value, and the deal officially closed in August 2026 according to Cursor itself and SpaceX’s SEC filing.

This changes Microsoft’s options today. One of the most obvious large AI software targets Microsoft considered is now owned by another giant technology company.

It also gives us unusually direct evidence that regulation can affect Microsoft’s acquisition strategy before a deal even reaches regulators. Microsoft reportedly killed the idea internally rather than spending months trying to get it approved.

Q12Could Inception be Microsoft’s next AI startup acquisition?

Inception remains the most concrete publicly reported Microsoft AI acquisition target, but as of now there is no evidence that Microsoft has actually bought it.

Reuters reported that Microsoft had held discussions with Inception, a Stanford-linked startup building diffusion-based language models. Microsoft’s M12 venture arm had already invested in the company’s $50 million seed financing.

Inception is interesting because its technology could strengthen Microsoft closer to the model layer itself. Its Mercury models generate text using diffusion, allowing parts of the output to be produced in parallel rather than token by token in the conventional autoregressive way.

The relationship has kept developing without an acquisition. Inception recently made Mercury 2 available through Azure Foundry, giving Microsoft customers access to the model while Inception remains independent.

We checked Microsoft’s current acquisition ledger again, and Inception still does not appear there. The last confirmed Microsoft acquisition remains Fintool.

That makes Inception worth watching, especially after Microsoft lost the chance to buy Cursor. But calling it Microsoft’s next acquisition today would get ahead of the evidence.

Q13How much is Microsoft spending on these startup deals?

We cannot produce a credible total for Microsoft’s recent startup buying because the company has left the prices of its three formal acquisitions undisclosed.

Microsoft has never published what it paid for Fintool, Osmos or Pi Labs. Their venture funding gives us some sense of scale but tells us very little about the final prices. Fintool had raised about $7.2 million, while Osmos had raised $13 million. Pi Labs was a very small team.

The unusual Inflection transaction is the exception. Reuters and Bloomberg reported approximately $650 million, mostly linked to licensing technology while Microsoft hired most of the team.

Cove’s team move also came without a disclosed purchase price, and Microsoft described the employees as joining the company rather than announcing an acquisition.

There is no defensible way to add these transactions together and claim Microsoft spent a specific amount. What we can say is that Microsoft’s visible recent strategy looks radically cheaper than the era of $26.2 billion for LinkedIn, $19.7 billion for Nuance and $68.7 billion for Activision Blizzard.

Q14Is regulation changing how Microsoft buys startups?

Regulation is clearly constraining some Microsoft deals, although we cannot prove that it caused the whole shift toward smaller acquisitions and acqui-hires.

Cursor gives us the strongest evidence. Reuters reported that Microsoft backed away because executives feared a purchase would face antitrust trouble alongside GitHub Copilot.

Inflection shows what happens on the other side. Microsoft avoided buying the company outright, yet the UK Competition and Markets Authority still investigated the hiring and licensing arrangement and decided it was substantial enough to qualify as a merger under UK law.

European disclosures now make several similar transactions visible. Microsoft’s dealings with Inflection, Bluesky Data, Tumult Labs and Robin AI all involved employee transfers, licensing or both instead of conventional full-company purchases.

Still, Microsoft has plenty of other reasons to prefer these deals. AI researchers are expensive and scarce, technologies change extremely quickly, and Microsoft already has huge distribution channels where acquired technology can be inserted.

Regulators are one part of the explanation. They make large overlapping acquisitions harder, while Microsoft’s product structure makes small team-and-technology deals unusually attractive anyway.

Q15So what startups has Microsoft bought recently?

Microsoft has bought very few startups outright lately: Fintool, Osmos and Pi Labs are the only three recent names on its official acquisition record.

That small number understates how much startup technology Microsoft is absorbing. Beyond those purchases, Microsoft has repeatedly hired startup teams, licensed technology and moved the people directly into Microsoft AI, Word, Fabric and LinkedIn.

The overall direction is now easy to see. Microsoft wants AI researchers, agent builders, data engineers, evaluation experts and people who have already figured out useful AI workflows. Often, Microsoft has little reason to preserve the startup around them.

The formal acquisition slowdown is dramatic. Microsoft went from 52 listed acquisitions across 2018 to 2021 to just 10 since the beginning of 2022. Yet Reuters reported this year that Microsoft is actively shopping for more AI startups, and its attempt to buy Cursor shows that the company is willing to consider much larger targets when the strategic fit is strong.

Our answer is direct: Microsoft currently buys surprisingly few startups as complete companies, but it is still aggressively buying access to startup talent and technology. These days, the legal acquisition list only shows part of what Microsoft is taking home.