What Microsoft AI does

Microsoft AI is a division of Microsoft Corporation, founded in 2024 and led by CEO Mustafa Suleyman (co-founder of DeepMind), that develops the in-house MAI model family and consumer-facing AI applications under a 'Humanist Superintelligence' philosophy. The division has built seven first-party models covering reasoning (MAI-Thinking-1, a 35B-active/~1T-total sparse MoE), coding (MAI-Code-1-Flash), image generation/editing (MAI-Image-2.5), transcription across 43 languages (MAI-Transcribe-1.5), and expressive speech synthesis (MAI-Voice-2), all trained from scratch on clean, traceable data without distillation and co-designed with custom Maia 200 silicon. These models are distributed through Microsoft Foundry (Azure AI Foundry) as the primary commercial API platform, embedded into first-party products including Microsoft 365 Copilot, GitHub Copilot, VS Code, Teams, PowerPoint, OneDrive, Bing, Dynamics 365 Contact Center, and the consumer-facing Copilot Health, with additional reach via OpenRouter, Fireworks AI, and Baseten. R&D operations span Redmond, Mountain View, New York, London, and Zurich.

The commercial model is hybrid: usage-based API pricing on Foundry (e.g., $0.36/hr for MAI-Transcribe-1.5, $22/1M characters for MAI-Voice-2, $5–$47/1M tokens for MAI-Image-2.5), per-seat subscriptions via Microsoft 365 Copilot and GitHub Copilot, and enterprise contracts for Microsoft Frontier Tuning — a reinforcement learning-based customization service that delivers up to 10x better token efficiency than GPT-5.5 and lets customers like McKinsey and Land O'Lakes own their tuned model variants exclusively. Customer base is horizontally distributed across enterprise developers, healthcare (Mayo Clinic flagship partnership), creative agencies (WPP, Shutterstock), professional services, and consumer health (50M+ health questions daily via Copilot), with no single logo concentration. The division is ISO/IEC 42001 certified and reports into Microsoft's broader commercial restructuring under Judson Althoff, with Mustafa Suleyman driving the AI agenda.

Name

Microsoft AI

Legal name

Microsoft Corporation

Website

https://microsoft.ai

Company type

Private

Founded year

2024

Operating status

Operating

Headcount range

5,001–10,000 employees

Short description

Microsoft AI is a division of Microsoft Corporation that builds in-house MAI foundation models (reasoning, coding, image, transcription, voice) co-designed with custom Maia 200 silicon, distributes them via Microsoft Foundry and across Microsoft 365, GitHub Copilot, and third-party platforms, and serves enterprise customers (Mayo Clinic, McKinsey, WPP) and billions of consumer end-users.

Ownership category

akta.pro rank

Product category

Enterprise AI Foundation Models

NAICS

Software Publishers (5132), Computer Systems Design Services (541512)

SIC

Services-Prepackaged Software (7372), Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

Foundation Model Developers (LLM/Multimodal Model Labs) (HDAAACAA)

akta.pro secondary industries

Enterprise Foundation Model Integration & APIs (Connectors, Governance, Deployment) (HDAAACAO), Agents & Autonomous Workflows (Tool Use, Planning, Multi-Agent) (HDAAACAF), End-to-End Enterprise AI Platforms (MLOps & Model Lifecycle Management) (HDAEANAA)

Keywords

• Foundation AI models
• Enterprise AI platform
• Multimodal AI models
• AI customization services
• Cloud AI APIs

Headquarters

HQ city

Redmond

HQ country

United States

HQ region

North America

Offices5 records

Markets served

GTM type

B2B

Offering type

Software

Cost components

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

Revenue model

1. Microsoft Azure AI / Foundry Platform: MAI models are offered as API-accessible services through Microsoft Foundry (Azure AI Foundry), a cloud-based enterprise AI development and deployment platform. Customers pay for API usage based on tokens processed or compute consumed. This is the primary commercial distribution mechanism for MAI models to external developers and enterprises.
2. Microsoft 365 Copilot Subscriptions: Copilot AI features including Copilot Health are distributed through Microsoft 365 Personal, Family, and Premium subscription tiers, generating recurring subscription revenue per seat.
3. GitHub Copilot: MAI-Code-1-Flash is integrated as a default model in GitHub Copilot within VS Code, contributing to Microsoft's commercial AI coding product revenue on a per-seat subscription basis.
4. Enterprise Customization (Frontier Tuning): Organizations including McKinsey and Land O'Lakes use Microsoft Frontier Tuning to create proprietary customized MAI models, generating enterprise contract revenue from co-creation and platform access.

Pricing tiers

Go-to-market motion4 records

Distribution channels6 records

Marketing channels7 records

Core offering

Microsoft AI develops and sells the proprietary MAI model family — seven in-house multimodal foundation models spanning reasoning, coding, image generation/editing, speech-to-text, and text-to-speech — distributed via the Microsoft Foundry enterprise API platform, embedded across Microsoft 365 products, and offered with enterprise customization through Microsoft Frontier Tuning. Models are co-designed with Microsoft's Maia 200 custom silicon and built without distillation from third-party models.

Product overview

Microsoft AI (MAI) operates as a unified AI research and product division centered on its Humanist Superintelligence philosophy. The core offering is the MAI model family — a multimodal ecosystem of seven in-house developed models including MAI-Thinking-1 (reasoning), MAI-Code-1-Flash (coding), MAI-Image-2.5 (image generation/editing), MAI-Transcribe-1.5 (speech-to-text), and MAI-Voice-2 (text-to-speech). These models are made available through Microsoft Foundry (the development platform with API access) and MAI Playground (interactive testing), with Microsoft Frontier Tuning offering enterprise customization via reinforcement learning. Consumer-facing products include Copilot Health (health insights app) and integration into Microsoft 365 products including Copilot, Teams, PowerPoint, and OneDrive. The MAI models differentiate through clean data lineage (no distillation from third-party models), co-design with Maia 200 silicon, and enterprise-grade security.

Differentiator

Problem solved

Functional benefit

Brands

• MAI: Microsoft AI model family including MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2, MAI-Transcribe-1, MAI-Voice-1, and related variants.
• Copilot
• Humanist Superintelligence
• Copilot Health
• Microsoft Frontier Tuning
• MAI Playground

Products and services

• MAI-Thinking-1 Flagship reasoning foundation model: 35B active / ~1T total parameters sparse Mixture-of-Experts model with 256K token context window, competitive with Claude Opus 4.6 on SWE-Bench Pro and preferred over Sonnet 4.6 in blind human evaluations. Sold to enterprise developers and integrated into Microsoft products.
• MAI-Code-1-Flash Inference-efficient agentic coding model with 5B active parameters, optimized for software engineering tasks and integrated as the default model in GitHub Copilot and VS Code. Outperforms Claude Haiku 4.5 on all coding benchmarks.
• MAI-Image-2.5 Text-to-image and image editing foundation model ranked No. 2 on the Arena image editing leaderboard and No. 3 on text-to-image. Supports fine-grained localized edits, face and identity consistency, and reliable text rendering for enterprise and creative use.
• MAI-Transcribe-1.5 State-of-the-art multilingual speech-to-text foundation model with the lowest Word Error Rate on the FLEURS benchmark across 43 languages. Priced at $0.36 per hour of audio; 5x faster than competing models; supports keyword biasing for domain-specific terminology.
• MAI-Voice-2 Expressive text-to-speech foundation model supporting 15 languages with granular emotion control (joy, anger, disgust, fear, sadness), zero-shot voice cloning from 5-60 second reference clips, code-switching for Hindi-English and Spanish-English, and consent guardrails. Priced at $22 per 1M characters.
• Microsoft Frontier Tuning Enterprise model customization service that uses reinforcement learning environments (RLEs) to fine-tune MAI models on customer-specific data, workflows, and institutional knowledge within the customer's secure environment. Tuned models remain exclusively owned by the customer and achieve up to 10x better token efficiency than GPT-5.5.
• Microsoft Foundry (Azure AI Foundry) Enterprise AI development and deployment platform providing API access to the MAI model family with built-in enterprise security, compliance, governance, and pricing. Primary commercial distribution channel for MAI models to external developers and enterprises.
• Copilot Health Secure consumer health application within Microsoft Copilot providing personalized health insights by combining electronic health records, wearable data, lab results, and a medical literature-grounded knowledge base. Distributed through Microsoft 365 Personal, Family, and Premium subscriptions for US users.

Quantifiable outcome

• MAI-Thinking-1 achieves 97.0% on AIME 2025 and 94.5% on AIME 2026 math benchmarks; 53% on SWE-Bench Pro (competitive with Claude Opus 4.6)
• +6 more outcomes

Named customers5 records

Segments5 records

Ideal customer profiles5 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration7 records

AI capability14 records

Feature9 records

Partnerships

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

• International Organisation for Migration (IOM)minorStrategic or Co-development Partner · 27 June 2026The International Organisation for Migration (UN migration agency) is working with Microsoft AI for Good Lab on satellite mapping following the Venezuela earthquake disaster. Microsoft AI for Good Lab's satellite analysis contributed to damage assessment showing 31.5% of buildings in Catia La Mar were damaged, supporting humanitarian response efforts.
• Mayo ClinicflagshipStrategic or Co-development Partner · 2 June 2026Microsoft and Mayo Clinic are co-creating a frontier AI model for healthcare that combines Mayo Clinic's world-leading clinical expertise, de-identified clinical data, and longitudinal health insights with Microsoft AI's foundational capabilities. The model will first be deployed within Mayo Clinic's own environment, and once validated, made available to other organizations via Microsoft Foundry. Mayo Clinic will own the resulting model, reinforcing mutual commitment to patient trust, clinical rigor, and responsible data stewardship. This represents Microsoft AI's highest-stakes healthcare partnership and a major strategic investment in domain-specific AI.
• McKinsey & CompanycoreStrategic or Co-development Partner · 2 June 2026McKinsey is an early adopter of Microsoft Frontier Tuning, using reinforcement learning environments to tune MAI models on their specific enterprise tasks and workflows. When tuned for McKinsey's exacting standards, MAI achieved the highest win rate of any model tested — outperforming GPT-5.5 on quality at roughly 10x lower cost. This demonstrates Frontier Tuning's enterprise value proposition for professional services firms.
• Land O'LakescoreStrategic or Co-development Partner · 2 June 2026Land O'Lakes is an early adopter of Microsoft Frontier Tuning, using it to strengthen grounded outputs and style compliance for production deployment. Their Product Development Scientist noted Frontier Tuning made a strong MAI model 'even stronger' with superior token efficiency. Land O'Lakes represents an agricultural cooperative using AI for product development workflows.
• OpenRoutercoreTechnology or Integration · 2 June 2026OpenRouter is making MAI models (MAI-Image-2.5, MAI-Voice-2, MAI-Transcribe-1.5) available to its community of 9 million developers through its marketplace API. For the first time, developers can tune model weights directly through OpenRouter. OpenRouter's CEO noted the goal is to make great new models immediately accessible to developers through the same API they already use.
• LINGUA Africa (Bill & Melinda Gates Foundation, Masakhane African Languages Hub, Google.org)minorStrategic or Co-development Partner · 27 May 2026Microsoft AI for Good Lab is a co-lead of LINGUA Africa, a funding and technical support initiative for projects building language resources, models, and AI applications for underrepresented African languages. Microsoft AI for Good Lab contributes technical expertise alongside Gates Foundation funding, Masakhane's linguistic expertise, and Google.org's support. Applications open until June 15, 2026 with funding up to $250,000 cash plus $400,000 compute credits.
• American Academy of Family Physicians (AAFP)minorStrategic or Co-development Partner · 26 May 2026AAFP launched the Primary Care Innovation Network (PCIN) with Microsoft AI as a represented technology partner on the inaugural Advisory Committee. Microsoft AI representatives will meet quarterly to provide strategic guidance on responsible AI deployment in primary care. The network includes webinars, hackathons, research, and national convenings to shape AI adoption at the frontline of healthcare.
• Taylor GeospatialminorStrategic or Co-development Partner · 21 April 2026Taylor Geospatial and Microsoft AI for Good Lab released the first global dataset of agricultural field boundaries ('Fields of the World') following an 18-month collaboration. The open dataset advances food security, carbon accounting, precision agriculture, and water-quality analysis. The UN Food and Agriculture Organization and NASA Harvest are applying the dataset. A follow-up 'Features of the World' project will expand to global infrastructure mapping.
• ShutterstockminorStrategic or Co-development Partner · 14 April 2026Shutterstock is evaluating MAI-Image-2-Efficient for production workflows in their evaluation work assessing prompt fidelity and creative usability. Their Principal Product Manager noted the model is 'trending in the right direction' for reliability in transitioning from experimentation to real-world use. Shutterstock represents a major stock media company exploring MAI image models.
• WPPcoreStrategic or Co-development Partner · 2 April 2026WPP, one of the world's largest marketing and communications groups, is among the first enterprise partners building with MAI-Image-2 at scale for campaign-ready image generation. WPP's Global Chief Creative Officer called MAI-Image-2 'a genuine game-changer' that responds to creative direction nuance and respects the craft of generating real-world images. WPP's creative talent is using the model to enhance global marketing campaigns.

Scale indicators10 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Anthropic: Frontier model lab behind the Claude family, cited directly as MAI-Thinking-1's performance benchmark (Claude Opus 4.6, Sonnet 4.6, Haiku 4.5). Competes head-to-head with MAI on reasoning, coding, and enterprise deployment.
• Google DeepMind: Google's frontier AI research division building Gemini multimodal models. Competes with MAI across reasoning, image, and voice modalities and is cited as a benchmark comparator (e.g., Gemini 3.1 Flash for transcription speed).
• Mistral AI: European frontier model lab offering open-weight and commercial foundation models via API. Competes with MAI in enterprise model-serving, particularly for European customers and open-weight use cases where Microsoft Azure deployment is not preferred.
• xAI: Elon Musk's frontier AI lab building Grok models with comparable reasoning and multimodal ambitions. A direct peer in the race for frontier-scale foundation models, with similar capital-intensity profile and enterprise GTM.
• OpenAI: Frontier foundation model developer and Microsoft AI's most direct comparable. OpenAI builds GPT-class foundation models and distributes them via API and ChatGPT, overlapping with MAI's reasoning, coding, image, and voice offerings across both consumer and enterprise channels.

Broad incumbents

• Meta AI (Llama): Meta's open-weight Llama family competes with MAI models in the open-source and self-hosted enterprise segments. While not a direct frontier reasoning competitor, Llama's breadth and licensing model shape the pricing envelope Microsoft AI must navigate.
• Amazon (AWS Bedrock / Anthropic partnership): AWS Bedrock is the closest hyperscaler competitor to Microsoft Foundry, offering hosted foundation models from multiple providers including Anthropic. Competes with MAI for enterprise AI platform spend rather than model quality alone.

Others

• Hugging Face: Open-source model hub and inference provider enabling enterprise access to thousands of models including open-weight competitors to MAI. Functions as an adjacent ecosystem and indirect distribution channel rather than a direct model competitor.
• NVIDIA (NeMo / NIM): NVIDIA is both a key infrastructure provider and a competitor through NeMo and NIM, which offer model customization and serving stacks to enterprises. Highly relevant as the silicon counterpoint to Microsoft AI's Maia 200 strategy.

Emerging players

• Cohere: Enterprise-focused foundation model provider offering Command models and retrieval/agents tooling. Targets the same enterprise customization segment that Frontier Tuning addresses for Microsoft AI, though at smaller scale.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat7 records

Key risks6 records

Key highlights7 records

Customer concentration

• Xx.com/MicrosoftAI
• LinkedInlinkedin.com/company/microsoft-ai
• Instagraminstagram.com/microsoftai

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Profiles3 records

Funding overview

Funding rounds

Investors

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

M&A

Investments1 record

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

Frequently asked questions about Microsoft AI