Microsoft Research

What Microsoft Research does

Microsoft Research is the fundamental science and technology research division of Microsoft Corporation, founded in 1991 and headquartered in Redmond, Washington. It conducts basic and applied research across the full computing stack through 12+ global labs in North America (Redmond, NYC, New England, Montreal), Europe (Cambridge UK, Munich), Asia (Beijing, Bangalore, Singapore), and Africa (Nairobi), employing 1,001-5,000 researchers across AI, quantum computing, healthcare, robotics (Physical AI), security, programming languages, and data infrastructure. Its research portfolio includes proprietary models such as the Phi-4 small language models, Rho-alpha Physical AI foundation model, NatureLM scientific model (56B parameters), Fara1.5 browser-use agents, and Miro/Mirage video world models, alongside research tools (mimalloc memory allocator, DoWhy/EconML causal AI) and unconventional computing platforms (Analog Optical Computer, Project Silica glass storage). The division has produced 777 AI architecture patents, placing it in Tier 1 globally behind only Google DeepMind.

The organization operates as a corporate research division rather than a commercial entity, funded through Microsoft's operating budget and generating value through technology transfer to Microsoft's commercial products (Azure AI services, GitHub Copilot, Microsoft 365) and strategic partnerships. Its GTM motion is hybrid: enterprise co-development partnerships (Siemens, NVIDIA, Barclays, Paige.ai, ChinaAMC), academic collaborations (MIT, Harvard, Tsinghua, Cornell, Renmin, Columbia, Zhejiang), and community-led research programs (Agentic AI Research and Innovation Program). Customer-facing outputs are routed through three channels: OEM/embedded partnerships (Tyger into Siemens MRI scanners), Azure cloud platform integration (Rho-alpha early access, Azure credits for academic programs), and open-source self-serve distribution (GitHub, Hugging Face). Recent expansion signals include entry into the Greater Bay Area via the HKUST-CalmCar Physical AI Innovation Center, deepening Africa presence via Paza ASR for 39 languages, and new commercial pathways for healthcare AI (Providence's GigaTIME, Virchow with Paige.ai).

The division serves a dual constituency: Microsoft Corporation (as internal R&D engine feeding Azure, GitHub, Microsoft 365, and Xbox product lines) and the broader global research community (via open-source releases, academic publications, fellowship programs, and conference presence such as CVPR 2026 Best Student Paper). Its strategic role within Microsoft is to produce foundational IP and de-risk long-horizon bets in AI infrastructure, robotics, healthcare, and unconventional computing that would be uneconomic for product teams to pursue directly.

Name

Microsoft Research

Legal name

Microsoft Corporation

Website

https://research.microsoft.com

Company type

Public

Founded year

1991

Operating status

Operating

Headcount range

1,001–5,000 employees

Short description

Microsoft Research is Microsoft's corporate research division, conducting fundamental AI, robotics, healthcare, and computing research across 12+ global labs and producing proprietary models and tools (Phi, Rho-alpha, Fara, mimalloc, Project Silica) that feed Microsoft commercial products.

Ownership category

akta.pro rank

Product category

AI Research and Development

NAICS

Software Publishers (513210), Software Publishers (5132), Software Publishers (51321), Computer Training (611420), Research and Development in the Physical, Engineering, and Life Sciences (except Nanotechnology and Biotechnology) (541715), Research and Development in the Physical, Engineering, and Life Sciences (54171), Scientific Research and Development Services (5417), Research and Development in Biotechnology (except Nanobiotechnology) (541714)

SIC

Electronic Computers (3571), Services-Computer Programming Services (7371), Services-Computer Integrated Systems Design (7373), Services-Prepackaged Software (7372)

akta.pro primary industry

Computer Science, IT & Data Science (EDAGABAB)

akta.pro secondary industries

Information Technology & Software (EDAAAPAB), Computer Science Theory (Computational Thinking, Logic, Discrete Math) (EDAMACAK), Knowledge Management & Scientific Collaboration / Literature Intelligence (HLAGAJAN)

Keywords

• Artificial intelligence research
• Machine learning models
• Quantum computing research
• Healthcare AI systems
• Open source research tools

Headquarters

HQ city

Redmond

HQ country

United States

HQ region

North America

Offices10 records

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Technology or R&D, Infrastructure, Operations, Marketing or Sales

Revenue model

1. Research Funding and Grants: Microsoft Research operates as a corporate research organization funded by Microsoft's operating budget. The division generates value through advancing technology that enables commercial products, attracting talent, and building partnerships.
2. Technology Transfer: Research innovations transition to Microsoft commercial products including Azure AI services, healthcare solutions, and developer tools like GitHub Copilot.
3. Healthcare AI Partnerships: Medical imaging AI frameworks like Tyger partnered with Siemens for MRI scanner integration, representing commercial healthcare technology pathways.

Go-to-market motion2 records

Distribution channels5 records

Marketing channels10 records

Core offering

Microsoft Research conducts basic and applied research in computer science, software engineering, artificial intelligence, machine learning, quantum computing, healthcare, robotics, and related fields. It develops foundation models, AI agent frameworks, programming tools, and specialized systems that transition into Microsoft commercial products. Outputs are delivered through open-source releases, academic publications, patents, and direct technology transfer to Microsoft business units and enterprise partners.

Product overview

Microsoft Research is the fundamental science and technology research division of Microsoft, conducting research across the full computing and AI stack through labs in Africa, Asia-Pacific, Cambridge, India, Montreal, New England, New York, and Redmond. Its product and technology portfolio spans several categories: (1) AI Agents & Automation — Fara1.5 (browser computer-use agents, 4B/9B/27B), MagenticLite, Webwright (terminal-native web agent framework), GitHub Agentic Workflows, and Arbor (persistent hypothesis tree for coding agents); (2) Physical AI & Robotics — Rho-alpha (Microsoft's first Physical AI foundation model combining vision, language, and tactile sensing), Argos (multimodal RL verification), VITRA (robotic VLA pre-training), GroundedPlanBench, and datacenter maintenance robotics; (3) Generative AI — Phi series small language models, Mirage and World-R1 (video world model consistency), Dayhoff Atlas (protein generation), QFANG (chemical synthesis planning), CleaveNet (peptide sensor design); (4) Healthcare AI — UniRG (medical imaging report generation), Tyger (MRI reconstruction), GigaTIME (tumor analysis), Talos (genomic rare disease detection); (5) Infrastructure & Storage — Project Silica (glass-based 10,000-year data storage), Analog Optical Computer (AOC, light-based AI inference), mimalloc (high-performance memory allocator), U.S. Power Grid Dataset; (6) Speech & Language — Paza/PazaBench (ASR for 39 African languages), Vega (zero-knowledge proofs for digital identity); (7) Research Tools & Benchmarks — AsgardBench, DELEGATE-52, GroundedPlanBench, and GeoMap (geologic analysis). The portfolio is structured as a broad research organization with individually named tools, models, and benchmarks rather than a unified commercial product platform.

Differentiator

Problem solved

Functional benefit

Brands

• Microsoft Research Asia: Asia-Pacific research lab of Microsoft Research focusing on AI research with applications in healthcare, finance, and logistics.
• Microsoft Research Cambridge
• Microsoft Research India
• Microsoft Research Africa
• Microsoft Health Futures
• AI Frontiers
• AI for Science

Products and services

• Phi-4 language model An open-weights small language model developed by Microsoft Research for reasoning, code generation, and general natural language tasks, available via Azure AI Foundry and Hugging Face. Targeted at developers, researchers, and enterprise AI builders.
• Rho-alpha (Physical AI foundation model) Microsoft Research's first foundation model purpose-built for physical AI and robotics, providing general-purpose embodied reasoning, motor control, and manipulation skills. Aimed at researchers and developers building robotic systems and embodied agents.
• Fara1.5 A Microsoft Research browser-based computer-use agent model that performs actions in a web browser on behalf of a user, available via Hugging Face. Aimed at developers building agentic browser automation.
• MagenticLite An experimental Microsoft Research agentic application that uses vision and tool use to complete web-based tasks autonomously. Aimed at researchers and developers exploring agentic workflows.
• Webwright A terminal-native web agent framework released by Microsoft Research for building, testing, and running browser-based AI agents. Aimed at developers and researchers building web automation.
• Arbor A Microsoft Research hypothesis tree framework for AI coding agents that explores multiple code-change hypotheses in parallel. Targeted at developers building advanced code-generation agents.
• Mirage A Microsoft Research video world model system that synthesizes video from text prompts. Targeted at researchers and creators exploring generative video.
• Tyger (MRI reconstruction framework) An open-source MRI reconstruction framework released by Microsoft Research in collaboration with Mass General Brigham, designed to accelerate MRI scan reconstruction. Targeted at clinical researchers and radiology teams.
• mimalloc An open-source, general-purpose memory allocator developed by Microsoft Research, widely used in high-performance runtimes and systems software. Targeted at systems software developers.
• Project Silica A long-term archival storage technology developed by Microsoft Research that uses ultrafast laser writing on quartz glass, originally demonstrated with Warner Bros. Targeted at archival storage use cases.
• Analog Optical Computer (AOC) An analog optical computing platform developed by Microsoft Research that performs continuous-variable optimization computations using photons, demonstrated for problems such as blood-coagulation inference. Targeted at researchers and customers exploring analog and optical computing.
• GitHub Agentic Workflows A workflow capability released by Microsoft Research in collaboration with GitHub that allows developers to define and run agentic AI workflows within GitHub repositories. Targeted at software developers.
• UniRG (medical imaging model) A universal medical imaging model released by Microsoft Research that generates radiology reports from medical images across modalities. Targeted at clinical researchers and radiology AI developers.
• Paza / PazaBench A Microsoft Research automatic speech recognition model and benchmark suite for African languages, released on Hugging Face and Azure AI Foundry. Targeted at speech and language researchers and developers serving African-language use cases.
• Health Podcast CT A Microsoft Research contrastive training approach for pretraining on medical imaging data, enabling improved downstream medical imaging models. Targeted at clinical AI researchers.
• Majorana 1 Microsoft Research's first quantum computing chip based on a topological qubit architecture, announced as a foundational step toward a million-qubit quantum machine. Targeted at quantum computing researchers and enterprise partners.
• World-R1 An open-source post-training recipe released by Microsoft Research for building video world models. Targeted at researchers developing generative video and world models.
• VITRA A Microsoft Research video world model architecture designed for embodied agents and physical AI. Targeted at researchers building embodied AI systems.
• Argos A Microsoft Research multimodal AI model for real-time translation across modalities. Targeted at translation researchers and developers.
• Atom reasoning model family A Microsoft Research family of reasoning-focused AI models with tool-use and graph-aware capabilities. Targeted at researchers and enterprise developers building reasoning applications.

Quantifiable outcome

• Arbor achieves 2.5x performance gains over Codex and Claude Code for same compute budget
• +7 more outcomes

Named customers15 records

Segments5 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

Integration8 records

AI capability54 records

Feature27 records

Partnerships

31 partnerships are on record, tiered core, flagship and minor.

• Broad InstitutecoreStrategic or Co-development Partner · 24 June 2026Co-led development of Talos genomic tool for automated rare disease detection, achieving 5.1% additional diagnostic yield by reanalysis of stored DNA data.
• MCRI (Murdoch Children's Research Institute)coreStrategic or Co-development Partner · 24 June 2026Led research validating Talos genomic tool across Australian cohorts, demonstrating automated reanalysis translating new scientific knowledge into clinical benefits within median 32 days.
• Renmin University of ChinacoreStrategic or Co-development Partner · 19 June 2026Joint development of Arbor hypothesis tree framework for AI coding agents with Gaoling School of Artificial Intelligence, achieving 2.5x performance gains over standard agents.
• Tsinghua UniversityflagshipStrategic or Co-development Partner · 11 June 2026Won CVPR 2026 Best Student Paper with O-Voxel, a novel 3D generative modeling approach improving AI-generated 3D asset quality through joint research with Microsoft AI team.
• HKUST and CalmCarcoreStrategic or Co-development Partner · 3 June 2026Physical AI Innovation Center established in Hong Kong with Microsoft Research Asia as inaugural Industry-Research Alliance partner. Focus on full-stack physical AI for autonomous driving, robotics, and smart manufacturing, leveraging Greater Bay Area supply chain.
• MITflagshipStrategic or Co-development Partner · 31 May 2026Joint research on model compression, pruning, quantization, and data compression algorithms for efficient AI deployment, including MetaEase verification tool for cloud algorithms.
• University HospitalscoreStrategic or Co-development Partner · 8 May 2026Co-authored editorial with Dr. Peter Pronovost on AI integration in healthcare for Patient Safety national campaign, advancing AI safety practices in medical settings.
• Zhejiang UniversitycoreStrategic or Co-development Partner · 1 May 2026Joint development of World-R1 reinforcement learning framework for geometric consistency in video generation models without architectural changes.
• Harvard UniversitycoreStrategic or Co-development Partner · 14 April 2026Joint development of energy-based fine-tuning (EBFT) method for language models, matching RLVR while outperforming SFT on Q&A, coding, and translation tasks.
• Google DeepMindflagshipStrategic or Co-development Partner · 8 April 2026Joint proposal of Agentic Risk Standard (ARS) framework for AI agent transactions, applying financial risk management principles to protect users from autonomous system errors.
• Columbia UniversitycoreStrategic or Co-development Partner · 8 April 2026Co-developed Agentic Risk Standard framework for AI agent financial transactions, with research showing framework reduces user losses by 24-61% in 5,000-round simulations.
• Virtuals ProtocolcoreStrategic or Co-development Partner · 8 April 2026Collaborated on Agentic Risk Standard framework and strategic integration of agentic finance infrastructure for autonomous treasury operations on XRP Ledger.
• t54 LabscoreStrategic or Co-development Partner · 8 April 2026Cross-institutional research partnership proposing Agentic Risk Standard, with t54's simulation finding 61% reduction in financial transaction losses using the framework.
• MediaTekcoreTechnology or Integration · 16 March 2026Joint development of Active Optical Cable using MicroLED technology for AI data centers, achieving 50% lower power consumption than conventional solutions. The partnership combines MediaTek's semiconductor expertise with Microsoft Research's data center infrastructure research.
• EPAM SystemsminorImplementation/ SI/ Consulting Partner · 11 March 2026EPAM joined Microsoft Intelligent Security Association (MISA), collaborating on cybersecurity solutions centered on Microsoft technologies to improve security outcomes for customers.
• University of LilleminorStrategic or Co-development Partner · 12 February 2026Research collaboration advancing brain-computer interface (BCI) technology integration with immersive VR/AR systems for usability and cybersickness mitigation.
• Digital GreencoreStrategic or Co-development Partner · 5 February 2026Project Gecko collaboration testing Paza ASR models with farmers in real-world agricultural settings across Kenya, improving transcription accuracy for low-resource languages.
• NVIDIAflagshipTechnology or Integration · 24 January 2026Strategic partnership for Rho-alpha robotics model training using NVIDIA Isaac Sim simulation framework on Azure, accelerating deployment for manufacturing and warehouse applications.
• Hexagon RoboticscoreStrategic or Co-development Partner · 24 January 2026Partnership accelerating Rho-alpha deployment for manufacturing, warehouses, and humanoid robotics applications through robotic system integration expertise.
• Johns Hopkins APLcoreStrategic or Co-development Partner · 24 January 2026Strategic partnership for Rho-alpha deployment in humanoid robotics applications, bringing research institution expertise to real-world robotic systems.
• Paige.aiflagshipStrategic or Co-development Partner · 8 January 2026Provided computational resources and expertise for Virchow multi-cancer detection AI, built from 4 million+ digitized pathology slides, representing first application of foundation models in cancer diagnostics.
• Cornell UniversitycoreStrategic or Co-development Partner · 7 January 2026Selected as one of twelve global research groups in Agentic AI Research and Innovation program, examining psychological security in high-autonomy AI systems for mental health interventions.
• Providence HealthflagshipStrategic or Co-development Partner · 10 December 2025Co-developed GigaTIME AI model for tumor analysis, capable of analyzing tumors in fraction of time and cost compared to existing methods, in collaboration with University of Washington.
• University of WashingtonflagshipStrategic or Co-development Partner · 10 December 2025Joint development of GigaTIME AI model with Providence for rapid, cost-effective tumor analysis using unique computational resources provided by Microsoft Research.
• Siemens HealthineersflagshipStrategic or Co-development Partner · 9 December 2025Partnership to embed Tyger MRI image reconstruction framework into Siemens MRI scanners, enabling faster scan times and real-time cloud-based processing for improved image quality.
• BarclaysflagshipStrategic or Co-development Partner · 26 November 2025Partnership to solve financial optimization problems using Analog Optical Computer technology, demonstrating AOC's capability for real-world enterprise applications beyond AI inference.
• Microsoft Research Asia SingaporecoreStrategic or Co-development Partner · 24 July 2025Singapore presence bridging AI research with real-world applications in healthcare, finance, and logistics through collaborations with local institutions and industry partners.
• PhysicsWallahcoreStrategic or Co-development Partner · 22 July 2025Collaboration with Microsoft Research India supporting Aryabhata AI model development for mathematics exam preparation, scoring 90.2% on JEE Main 2025 papers.
• MITflagshipStrategic or Co-development Partner · 1 November 2024Joint development of MetaEase verification tool for stress-testing cloud computing algorithms before deployment, using symbolic execution to identify worst-case failure scenarios.
• ChinaAMCflagshipStrategic or Co-development Partner · 1 March 2024Partnership since 2017 integrating AI into asset management operations, managing approximately RMB3.245 trillion (US$464.5 billion) in assets under management as of end 2025.
• Global Health Drug Discovery InstitutecoreStrategic or Co-development Partner · 1 January 2024Partnership to use AI for designing molecule inhibitors for COVID-19 and tuberculosis, advancing global health drug discovery capabilities.

Scale indicators5 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Anthropic: Anthropic is a direct peer in foundation model research (Claude series) and AI safety, with comparable research depth in agents, computer use, and enterprise AI. Anthropic's published research, talent base, and commercial trajectory place it in direct competition with MSR's AI Frontiers work.
• IBM Research: IBM Research is the closest historical analog to MSR: a 75+ year-old corporate research division conducting fundamental research in AI (watsonx), quantum computing, healthcare, and scientific computing. Both organizations have similar global lab networks, long operating histories, enterprise-focused commercialization paths, and deep involvement in healthcare AI and programming languages research.
• Google Research: Google Research is a direct peer conducting broad fundamental and applied research across AI, systems, and scientific computing. Both organizations share publishing models, open-source releases, university partnerships, and integration with major commercial product lines (Azure vs. Google Cloud, Microsoft 365 vs. Google Workspace).
• Google DeepMind: Google DeepMind is the most directly comparable peer to Microsoft Research: a large-scale corporate AI research organization conducting fundamental and applied research across deep learning, reinforcement learning, healthcare, and scientific computing. DeepMind holds the leading 1,093 AI architecture patents versus MSR's 777, and competes head-to-head in foundation models, agents, and life sciences.
• OpenAI: OpenAI is a direct peer in AI research and frontier model development (GPT series, Operator computer-use agents), competing with MSR's Fara1.5, Webwright, and Phi models. Both organizations target enterprise AI monetization and publish peer-reviewed research, though OpenAI operates with a more product-centric commercial model.
• Apple Machine Learning Research: Apple Machine Learning Research is a direct peer conducting fundamental AI research with emphasis on efficient on-device models, privacy-preserving ML, and multimodal systems. Like MSR, Apple ML Research publishes openly, releases open-source frameworks, and integrates research breakthroughs into commercial products.
• Meta AI (FAIR): Meta's Fundamental AI Research (FAIR) lab is a direct peer operating at the same scale as MSR in fundamental AI research across computer vision, NLP, generative AI, and open-source foundation models (LLaMA family). Like MSR, FAIR leverages its parent company's distribution (Facebook/Instagram/WhatsApp) and releases open-source research outputs.

Regional players

• Alibaba DAMO Academy: Alibaba DAMO Academy is a regional peer conducting fundamental research in AI, quantum computing, and semiconductor technology, with labs across Asia and global research partnerships. Its scope and mission closely parallel MSR Asia's research agenda and integration with a major cloud platform (Alibaba Cloud).
• Tencent AI Lab: Tencent AI Lab is a regional peer conducting fundamental AI research in computer vision, NLP, speech, and gaming AI, primarily serving Asia-Pacific markets. Comparable to MSR Asia in scope and scale, with similar partnerships across Chinese academia and industry.
• Baidu Research: Baidu Research is a regional peer focused on AI fundamentals including deep learning, NLP, autonomous driving, and quantum computing, primarily serving the Chinese market. Like MSR, Baidu Research invests heavily in long-horizon fundamental research with both academic publications and commercial applications.

Market position

Weaknesses5 records

Competitive moat6 records

Key risks6 records

Key highlights8 records

Customer concentration

• Xx.com/MSFTResearch
• YouTubeyoutube.com/user/MicrosoftResearch
• Facebookfacebook.com/microsoftresearch
• LinkedInlinkedin.com/company/microsoft
• Instagraminstagram.com/msft_research

Revenue estimate

Valuation estimate

Number of profiles

Profiles7 records

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments2 records

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Microsoft Research